arXiv:2001.04235v2 [cs.

DC] 3 Jan 2021

Notes on Theory of Distributed Systems

James Aspnes

2021-01-05 01:59
 i

Copyright © 2002–2020 by James Aspnes. Distributed under a Cre-

ative Commons Attribution-ShareAlike 4.0 International license: https:
//creativecommons.org/licenses/by-sa/4.0/.
Contents

Table of contents ii

List of figures xv

List of tables xvi

List of algorithms xvii

Preface xxi

Lecture schedule xxii

1 Introduction 1
1.1 Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

I Message passing 7

2 Model 8
2.1 Basic message-passing model . . . . . . . . . . . . . . . . . . 8
2.1.1 Formal details . . . . . . . . . . . . . . . . . . . . . . 9
2.1.2 Network structure . . . . . . . . . . . . . . . . . . . . 10
2.2 Asynchronous systems . . . . . . . . . . . . . . . . . . . . . . 10
2.2.1 Example: client-server computing . . . . . . . . . . . . 11
2.3 Synchronous systems . . . . . . . . . . . . . . . . . . . . . . . 12
2.4 Drawing message-passing executions . . . . . . . . . . . . . . 12
2.5 Complexity measures . . . . . . . . . . . . . . . . . . . . . . . 14

ii
CONTENTS iii

3 Broadcast and convergecast 16

3.1 Flooding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.1.1 Basic algorithm . . . . . . . . . . . . . . . . . . . . . . 16
3.1.2 Adding parent pointers . . . . . . . . . . . . . . . . . 18
3.1.3 Termination . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 Convergecast . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.3 Flooding and convergecast together . . . . . . . . . . . . . . . 21

4 Distributed breadth-first search 23

4.1 Using explicit distances . . . . . . . . . . . . . . . . . . . . . 23
4.2 Using layering . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.3 Using local synchronization . . . . . . . . . . . . . . . . . . . 25

5 Leader election 29
5.1 Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
5.2 Leader election in rings . . . . . . . . . . . . . . . . . . . . . 31
5.2.1 The Le Lann-Chang-Roberts algorithm . . . . . . . . 31
5.2.1.1 Proof of correctness for synchronous executions 32
5.2.1.2 Performance . . . . . . . . . . . . . . . . . . 33
5.2.2 The Hirschberg-Sinclair algorithm . . . . . . . . . . . 33
5.2.3 Peterson’s algorithm for the unidirectional ring . . . . 34
5.2.4 A simple randomized O(n log n)-message algorithm . . 35
5.3 Leader election in general networks . . . . . . . . . . . . . . . 36
5.4 Lower bounds . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
5.4.1 Lower bound on asynchronous message complexity . . 37
5.4.2 Lower bound for comparison-based algorithms . . . . 38

6 Logical clocks 41
6.1 Causal ordering . . . . . . . . . . . . . . . . . . . . . . . . . . 41
6.2 Implementations . . . . . . . . . . . . . . . . . . . . . . . . . 43
6.2.1 Lamport clock . . . . . . . . . . . . . . . . . . . . . . 44
6.2.2 Neiger-Toueg-Welch clock . . . . . . . . . . . . . . . . 45
6.2.3 Vector clocks . . . . . . . . . . . . . . . . . . . . . . . 46
6.3 Consistent snapshots . . . . . . . . . . . . . . . . . . . . . . . 46
6.3.1 Property testing . . . . . . . . . . . . . . . . . . . . . 48

7 Synchronizers 49
7.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
7.2 Implementations . . . . . . . . . . . . . . . . . . . . . . . . . 50
7.2.1 The alpha synchronizer . . . . . . . . . . . . . . . . . 51
CONTENTS iv

7.2.2 The beta synchronizer . . . . . . . . . . . . . . . . . . 51

7.2.3 The gamma synchronizer . . . . . . . . . . . . . . . . 52
7.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
7.4 Limitations of synchronizers . . . . . . . . . . . . . . . . . . . 53
7.4.1 Impossibility with crash failures . . . . . . . . . . . . 53
7.4.2 Unavoidable slowdown with global synchronization . . 54

8 Coordinated attack 57
8.1 Formal description . . . . . . . . . . . . . . . . . . . . . . . . 57
8.2 Impossibility proof . . . . . . . . . . . . . . . . . . . . . . . . 58
8.3 Randomized coordinated attack . . . . . . . . . . . . . . . . . 60
8.3.1 An algorithm . . . . . . . . . . . . . . . . . . . . . . . 60
8.3.2 Why it works . . . . . . . . . . . . . . . . . . . . . . . 61
8.3.3 Almost-matching lower bound . . . . . . . . . . . . . . 62

9 Synchronous agreement 63
9.1 Problem definition . . . . . . . . . . . . . . . . . . . . . . . . 63
9.2 Solution using flooding . . . . . . . . . . . . . . . . . . . . . . 64
9.3 Lower bound on rounds . . . . . . . . . . . . . . . . . . . . . 65
9.4 Variants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

10 Byzantine agreement 69
10.1 Lower bounds . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
10.1.1 Minimum number of rounds . . . . . . . . . . . . . . . 69
10.1.2 Minimum number of processes . . . . . . . . . . . . . 69
10.1.3 Minimum connectivity . . . . . . . . . . . . . . . . . . 71
10.1.4 Weak Byzantine agreement . . . . . . . . . . . . . . . 72
10.2 Upper bounds . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
10.2.1 Exponential information gathering gets n = 3f + 1 . . 73
10.2.1.1 Proof of correctness . . . . . . . . . . . . . . 75
10.2.2 Phase king gets constant-size messages . . . . . . . . . 76
10.2.2.1 The algorithm . . . . . . . . . . . . . . . . . 77
10.2.2.2 Proof of correctness . . . . . . . . . . . . . . 78
10.2.2.3 Performance of phase king . . . . . . . . . . 79

11 Impossibility of asynchronous agreement 80

11.1 Agreement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
11.2 Failures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
11.3 Steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
11.4 Bivalence and univalence . . . . . . . . . . . . . . . . . . . . . 82
CONTENTS v

11.5 Existence of an initial bivalent configuration . . . . . . . . . . 83

11.6 Staying in a bivalent configuration . . . . . . . . . . . . . . . 83
11.7 Generalization to other models . . . . . . . . . . . . . . . . . 84

12 Paxos 85
12.1 The Paxos algorithm . . . . . . . . . . . . . . . . . . . . . . . 85
12.2 Informal analysis: how information flows between rounds . . 89
12.2.1 Example execution . . . . . . . . . . . . . . . . . . . . 89
12.2.2 Safety properties . . . . . . . . . . . . . . . . . . . . . 91
12.2.3 Learning the results . . . . . . . . . . . . . . . . . . . 92
12.2.4 Liveness properties . . . . . . . . . . . . . . . . . . . . 92
12.3 Replicated state machines and multi-Paxos . . . . . . . . . . 93

13 Failure detectors 94
13.1 How to build a failure detector . . . . . . . . . . . . . . . . . 95
13.2 Classification of failure detectors . . . . . . . . . . . . . . . . 95
13.2.1 Degrees of completeness . . . . . . . . . . . . . . . . . 95
13.2.2 Degrees of accuracy . . . . . . . . . . . . . . . . . . . 95
13.2.3 Boosting completeness . . . . . . . . . . . . . . . . . . 96
13.2.4 Failure detector classes . . . . . . . . . . . . . . . . . . 97
13.3 Consensus with S . . . . . . . . . . . . . . . . . . . . . . . . . 98
13.3.1 Proof of correctness . . . . . . . . . . . . . . . . . . . 99
13.4 Consensus with ♦S and f < n/2 . . . . . . . . . . . . . . . . 100
13.4.1 Proof of correctness . . . . . . . . . . . . . . . . . . . 102
13.5 f < n/2 is still required even with ♦P . . . . . . . . . . . . . 103
13.6 Relationships among the classes . . . . . . . . . . . . . . . . . 103

14 Quorum systems 105

14.1 Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
14.2 Simple quorum systems . . . . . . . . . . . . . . . . . . . . . 105
14.3 Goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
14.4 Paths system . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
14.5 Byzantine quorum systems . . . . . . . . . . . . . . . . . . . 108
14.6 Probabilistic quorum systems . . . . . . . . . . . . . . . . . . 109
14.6.1 Example . . . . . . . . . . . . . . . . . . . . . . . . . . 110
14.6.2 Performance . . . . . . . . . . . . . . . . . . . . . . . 110
14.7 Signed quorum systems . . . . . . . . . . . . . . . . . . . . . 111
CONTENTS vi

II Shared memory 112

15 Model 113
15.1 Atomic registers . . . . . . . . . . . . . . . . . . . . . . . . . 113
15.2 Single-writer versus multi-writer registers . . . . . . . . . . . 114
15.3 Fairness and crashes . . . . . . . . . . . . . . . . . . . . . . . 115
15.4 Concurrent executions . . . . . . . . . . . . . . . . . . . . . . 115
15.5 Consistency properties . . . . . . . . . . . . . . . . . . . . . . 116
15.6 Complexity measures . . . . . . . . . . . . . . . . . . . . . . . 117
15.7 Fancier registers . . . . . . . . . . . . . . . . . . . . . . . . . 118

16 Distributed shared memory 120

16.1 Message passing from shared memory . . . . . . . . . . . . . 121
16.2 Shared memory from message passing: the Attiya-Bar-Noy-
Dolev algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . 121
16.3 Proof of linearizability . . . . . . . . . . . . . . . . . . . . . . 123
16.4 Proof that f < n/2 is necessary . . . . . . . . . . . . . . . . . 124
16.5 Multiple writers . . . . . . . . . . . . . . . . . . . . . . . . . . 124
16.6 Other operations . . . . . . . . . . . . . . . . . . . . . . . . . 125

17 Mutual exclusion 126

17.1 The problem . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
17.2 Goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
17.3 Mutual exclusion using strong primitives . . . . . . . . . . . . 127
17.3.1 Test and set . . . . . . . . . . . . . . . . . . . . . . . . 128
17.3.2 A lockout-free algorithm using an atomic queue . . . . 129
17.3.2.1 Reducing space complexity . . . . . . . . . . 130
17.4 Mutual exclusion using only atomic registers . . . . . . . . . 130
17.4.1 Peterson’s algorithm . . . . . . . . . . . . . . . . . . . 130
17.4.1.1 Correctness of Peterson’s protocol . . . . . . 131
17.4.1.2 Generalization to n processes . . . . . . . . . 135
17.4.2 Fast mutual exclusion . . . . . . . . . . . . . . . . . . 135
17.4.3 Lamport’s Bakery algorithm . . . . . . . . . . . . . . 138
17.4.4 Lower bound on the number of registers . . . . . . . . 139
17.5 RMR complexity . . . . . . . . . . . . . . . . . . . . . . . . . 141
17.5.1 Cache-coherence vs. distributed shared memory . . . . 141
17.5.2 RMR complexity of Peterson’s algorithm . . . . . . . 142
17.5.3 Mutual exclusion in the DSM model . . . . . . . . . . 143
17.5.4 Lower bounds . . . . . . . . . . . . . . . . . . . . . . . 145
CONTENTS vii

18 The wait-free hierarchy 146

18.1 Classification by consensus number . . . . . . . . . . . . . . . 148
18.1.1 Level 1: registers etc. . . . . . . . . . . . . . . . . . . 149
18.1.2 Level 2: interfering RMW objects etc. . . . . . . . . . 151
18.1.3 Level ∞: objects where the first write wins . . . . . . 152
18.1.4 Level 2m − 2: simultaneous m-register write . . . . . . 154
18.1.4.1 Matching impossibility result . . . . . . . . . 156
18.1.5 Level m: m-process consensus objects . . . . . . . . . 157
18.2 Universality of consensus . . . . . . . . . . . . . . . . . . . . . 158

19 Atomic snapshots 161

19.1 The basic trick: two identical collects equals a snapshot . . . 161
19.2 Snapshots using double collects with helping . . . . . . . . . . 162
19.2.1 Linearizability . . . . . . . . . . . . . . . . . . . . . . 163
19.2.2 Using bounded registers . . . . . . . . . . . . . . . . . 164
19.3 Faster snapshots using lattice agreement . . . . . . . . . . . . 167
19.3.1 Lattice agreement . . . . . . . . . . . . . . . . . . . . 167
19.3.2 Connection to vector clocks . . . . . . . . . . . . . . . 168
19.3.3 The full reduction . . . . . . . . . . . . . . . . . . . . 169
19.3.4 Why this works . . . . . . . . . . . . . . . . . . . . . . 170
19.3.5 Implementing lattice agreement . . . . . . . . . . . . . 171
19.4 Practical snapshots using LL/SC . . . . . . . . . . . . . . . . 174
19.4.1 Details of the single-scanner snapshot . . . . . . . . . 175
19.4.2 Extension to multiple scanners . . . . . . . . . . . . . 178
19.5 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
19.5.1 Multi-writer registers from single-writer registers . . . 178
19.5.2 Counters . . . . . . . . . . . . . . . . . . . . . . . . . 179
19.5.3 Resilient snapshot objects . . . . . . . . . . . . . . . . 179

20 Lower bounds on perturbable objects 181

21 Restricted-use objects 185

21.1 Max registers . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
21.2 Implementing bounded max registers . . . . . . . . . . . . . . 186
21.3 Encoding the set of values . . . . . . . . . . . . . . . . . . . . 188
21.4 Unbounded max registers . . . . . . . . . . . . . . . . . . . . 188
21.5 Lower bound . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
21.6 Max-register snapshots . . . . . . . . . . . . . . . . . . . . . . 190
21.6.1 Linearizability . . . . . . . . . . . . . . . . . . . . . . 191
21.7 Restricted-use snapshots . . . . . . . . . . . . . . . . . . . . . 193
CONTENTS viii

21.7.1 Randomized and amortized snapshots . . . . . . . . . 195

22 Common2 197
22.1 Test-and-set and swap for two processes . . . . . . . . . . . . 198
22.2 Building n-process TAS from 2-process TAS . . . . . . . . . . 198
22.3 Obstruction-free swap from test-and-set . . . . . . . . . . . . 200
22.4 Wait-free swap from test-and-set . . . . . . . . . . . . . . . . 202
22.5 Implementations using stronger base objects . . . . . . . . . . 205

23 Randomized consensus and test-and-set 207

23.1 Role of the adversary in randomized algorithms . . . . . . . . 207
23.2 History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
23.3 Reduction to simpler primitives . . . . . . . . . . . . . . . . . 210
23.3.1 Adopt-commit objects . . . . . . . . . . . . . . . . . . 210
23.3.2 Conciliators . . . . . . . . . . . . . . . . . . . . . . . . 211
23.4 Implementing an adopt-commit object . . . . . . . . . . . . . 212
23.5 Conciliators and shared coins . . . . . . . . . . . . . . . . . . 212
23.6 A one-register conciliator for an oblivious adversary . . . . . 214
23.7 Sifters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
23.7.1 Test-and-set using sifters . . . . . . . . . . . . . . . . 218
23.7.2 Consensus using sifters . . . . . . . . . . . . . . . . . . 218
23.7.3 A better sifter for test-and-set . . . . . . . . . . . . . 220
23.8 Space bounds . . . . . . . . . . . . . . . . . . . . . . . . . . . 222

24 Renaming 224
24.1 Renaming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
24.2 Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
24.3 Order-preserving renaming . . . . . . . . . . . . . . . . . . . 226
24.4 Deterministic renaming . . . . . . . . . . . . . . . . . . . . . 226
24.4.1 Wait-free renaming with 2n − 1 names . . . . . . . . . 227
24.4.2 Long-lived renaming . . . . . . . . . . . . . . . . . . . 228
24.4.3 Renaming without snapshots . . . . . . . . . . . . . . 229
24.4.3.1 Splitters . . . . . . . . . . . . . . . . . . . . . 229
24.4.3.2 Splitters in a grid . . . . . . . . . . . . . . . 230
24.4.4 Getting to 2n − 1 names in polynomial space . . . . . 232
24.4.5 Renaming with test-and-set . . . . . . . . . . . . . . . 233
24.5 Randomized renaming . . . . . . . . . . . . . . . . . . . . . . 233
24.5.1 Randomized splitters . . . . . . . . . . . . . . . . . . . 234
24.5.2 Randomized test-and-set plus sampling . . . . . . . . 234
24.5.3 Renaming with sorting networks . . . . . . . . . . . . 235
CONTENTS ix

24.5.3.1 Sorting networks . . . . . . . . . . . . . . . . 235

24.5.3.2 Renaming networks . . . . . . . . . . . . . . 236
24.5.4 Randomized loose renaming . . . . . . . . . . . . . . . 238

25 Software transactional memory 240

25.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
25.2 Basic approaches . . . . . . . . . . . . . . . . . . . . . . . . . 241
25.3 Implementing multi-word RMW . . . . . . . . . . . . . . . . 242
25.3.1 Overlapping LL/SC . . . . . . . . . . . . . . . . . . . 243
25.3.2 Representing a transaction . . . . . . . . . . . . . . . 243
25.3.3 Executing a transaction . . . . . . . . . . . . . . . . . 244
25.3.4 Proof of linearizability . . . . . . . . . . . . . . . . . . 244
25.3.5 Proof of non-blockingness . . . . . . . . . . . . . . . . 245
25.4 Improvements . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
25.5 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245

26 Obstruction-freedom 247
26.1 Why build obstruction-free algorithms? . . . . . . . . . . . . 248
26.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
26.2.1 Lock-free implementations . . . . . . . . . . . . . . . . 248
26.2.2 Double-collect snapshots . . . . . . . . . . . . . . . . . 248
26.2.3 Software transactional memory . . . . . . . . . . . . . 249
26.2.4 Obstruction-free test-and-set . . . . . . . . . . . . . . 249
26.2.5 An obstruction-free deque . . . . . . . . . . . . . . . . 251
26.3 Boosting obstruction-freedom to wait-freedom . . . . . . . . . 253
26.3.1 Cost . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
26.4 Lower bounds for lock-free protocols . . . . . . . . . . . . . . 258
26.4.1 Contention . . . . . . . . . . . . . . . . . . . . . . . . 258
26.4.2 The class G . . . . . . . . . . . . . . . . . . . . . . . . 259
26.4.3 The lower bound proof . . . . . . . . . . . . . . . . . . 261
26.4.4 Consequences . . . . . . . . . . . . . . . . . . . . . . . 264
26.4.5 More lower bounds . . . . . . . . . . . . . . . . . . . . 265
26.5 Practical considerations . . . . . . . . . . . . . . . . . . . . . 265

27 BG simulation 266
27.1 High-level strategy . . . . . . . . . . . . . . . . . . . . . . . . 266
27.2 Safe agreement . . . . . . . . . . . . . . . . . . . . . . . . . . 267
27.3 The basic simulation algorithm . . . . . . . . . . . . . . . . . 269
27.4 Effect of failures . . . . . . . . . . . . . . . . . . . . . . . . . 270
27.5 Inputs and outputs . . . . . . . . . . . . . . . . . . . . . . . . 270
CONTENTS x

27.6 Correctness of the simulation . . . . . . . . . . . . . . . . . . 271

27.7 BG simulation and consensus . . . . . . . . . . . . . . . . . . 271

28 Topological methods 273

28.1 Basic idea . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
28.2 k-set agreement . . . . . . . . . . . . . . . . . . . . . . . . . . 274
28.3 Representing distributed computations using topology . . . . 275
28.3.1 Simplicial complexes and process states . . . . . . . . 276
28.3.2 Subdivisions . . . . . . . . . . . . . . . . . . . . . . . 279
28.4 Impossibility of k-set agreement . . . . . . . . . . . . . . . . . 283
28.5 Simplicial maps and specifications . . . . . . . . . . . . . . . 285
28.5.1 Mapping inputs to outputs . . . . . . . . . . . . . . . 286
28.6 The asynchronous computability theorem . . . . . . . . . . . 286
28.6.1 The participating set protocol . . . . . . . . . . . . . . 287
28.7 Proving impossibility results . . . . . . . . . . . . . . . . . . . 289
28.7.1 k-connectivity . . . . . . . . . . . . . . . . . . . . . . . 289
28.7.2 Impossibility proofs for specific problems . . . . . . . 290

29 Approximate agreement 292

29.1 Algorithms for approximate agreement . . . . . . . . . . . . . 292
29.2 Lower bound on step complexity . . . . . . . . . . . . . . . . 295

III Other communication models 297

30 Overview 298

31 Self-stabilization 299
31.1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
31.2 Token ring circulation . . . . . . . . . . . . . . . . . . . . . . 300
31.3 Synchronizers . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
31.4 Spanning trees . . . . . . . . . . . . . . . . . . . . . . . . . . 306
31.5 Self-stabilization and local algorithms . . . . . . . . . . . . . 307

32 Distributed graph algorithms 309

32.1 The LOCAL and CONGEST models . . . . . . . . . . . . . . 309
32.2 Local graph coloring . . . . . . . . . . . . . . . . . . . . . . . 310
32.2.1 Coloring graphs with out-degree 1 . . . . . . . . . . . 310
32.2.2 Lower bound for rings . . . . . . . . . . . . . . . . . . 311
32.2.3 Coloring bounded-degree graphs . . . . . . . . . . . . 312
CONTENTS xi

33 Population protocols 314

33.1 Definition of a population protocol . . . . . . . . . . . . . . . 315
33.2 Stably computable predicates . . . . . . . . . . . . . . . . . . 316
33.2.1 Time complexity . . . . . . . . . . . . . . . . . . . . . 316
33.2.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . 317
33.2.2.1 Leader election . . . . . . . . . . . . . . . . . 317
33.2.2.2 Distributing the output . . . . . . . . . . . . 318
33.2.2.3 Remainder mod m . . . . . . . . . . . . . . . 318
33.2.2.4 Linear threshold functions . . . . . . . . . . 318
33.2.3 Presburger arithmetic and semilinear sets . . . . . . . 319
33.2.3.1 Semilinear predicates are stably computable 320
33.2.3.2 Stably computable predicates are semilinear 321
33.3 Random interactions . . . . . . . . . . . . . . . . . . . . . . . 321

34 Mobile robots 325

34.1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
34.2 Two robots, no faults . . . . . . . . . . . . . . . . . . . . . . . 327
34.3 Three robots . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
34.4 Many robots, with crash failures . . . . . . . . . . . . . . . . 330

35 Beeping 332
35.1 Interval coloring . . . . . . . . . . . . . . . . . . . . . . . . . 333
35.1.1 Estimating the degree . . . . . . . . . . . . . . . . . . 334
35.1.2 Picking slots . . . . . . . . . . . . . . . . . . . . . . . 334
35.1.3 Detecting collisions . . . . . . . . . . . . . . . . . . . . 334
35.2 Maximal independent set . . . . . . . . . . . . . . . . . . . . 335
35.2.1 Lower bound . . . . . . . . . . . . . . . . . . . . . . . 335
35.2.2 Upper bound with known bound on n . . . . . . . . . 337

Appendix 341

A Assignments 341
A.0 Assignment 0: due as soon as possible . . . . . . . . . . . . . 341
A.1 Assignment 1: due Wednesday, 2020-09-23, at 5:00pm Eastern
US time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342
A.1.1 A token-passing game . . . . . . . . . . . . . . . . . . 342
A.1.2 A load-balancing problem . . . . . . . . . . . . . . . . 344
A.2 Assignment 2: due Wednesday, 2020-10-07, at 5:00pm Eastern
US time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
CONTENTS xii

A.2.1 Synchronous agreement with limited broadcast . . . . 345

A.2.2 Asynchronous agreement with limited failures . . . . . 346
A.3 Assignment 3: due Wednesday, 2020-10-21, at 5:00pm Eastern
US time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
A.3.1 Too many Byzantine processes . . . . . . . . . . . . . 348
A.3.2 Committee election . . . . . . . . . . . . . . . . . . . . 349
A.4 Assignment 4: due Wednesday, 2020-11-04, at 5:00pm Eastern
US time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
A.4.1 Counting without snapshots . . . . . . . . . . . . . . . 350
A.4.2 Rock-paper-scissors . . . . . . . . . . . . . . . . . . . . 352
A.5 Assignment 5: due Wednesday, 2020-11-18, at 5:00pm Eastern
US time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 354
A.5.1 Randomized consensus with one max register . . . . . 354
A.5.2 A plurality object . . . . . . . . . . . . . . . . . . . . 355
A.6 Presentation (for students taking CPSC 565): due Tuesday,
2020-12-01, or Thursday, 2020-12-03, in class; paper selection
due Friday, 2020-11-20. . . . . . . . . . . . . . . . . . . . . . . 355

B Sample assignments from Spring 2019 358

B.1 Assignment 1: due Wednesday, 2019-02-13, at 5:00pm . . . . 358
B.1.1 A message-passing bureaucracy . . . . . . . . . . . . . 358
Time complexity . . . . . . . . . . . . . . . . . 358
Message complexity . . . . . . . . . . . . . . . 360
B.1.2 Algorithms on rings . . . . . . . . . . . . . . . . . . . 360
B.1.3 Shutting down . . . . . . . . . . . . . . . . . . . . . . 362
B.2 Assignment 2: due Wednesday, 2019-03-06, at 5:00pm . . . . 363
B.2.1 A non-failure detector . . . . . . . . . . . . . . . . . . 363
B.2.2 Ordered partial broadcast . . . . . . . . . . . . . . . . 364
B.2.3 Mutual exclusion using a counter . . . . . . . . . . . . 366
B.3 Assignment 3: due Wednesday, 2019-04-17, at 5:00pm . . . . 369
B.3.1 Zero, one, many . . . . . . . . . . . . . . . . . . . . . 369
B.3.2 A very slow counter . . . . . . . . . . . . . . . . . . . 370
B.3.3 Double-entry bookkeeping . . . . . . . . . . . . . . . . 371
B.4 Presentation (for students taking CPSC 565): due Monday,
2019-04-22, or Wednesday, 2019-04-24, in class . . . . . . . . 372
B.5 CS465/CS565 Final Exam, May 7th, 2019 . . . . . . . . . . . 373
B.5.1 A roster (20 points) . . . . . . . . . . . . . . . . . . . 373
B.5.2 Self-stabilizing consensus (20 points) . . . . . . . . . . 374
B.5.3 All-or-nothing intermittent faults (20 points) . . . . . 375
B.5.4 A tamper-proof register (20 points) . . . . . . . . . . . 376
CONTENTS xiii

C Sample assignments from Spring 2016 377

C.1 Assignment 1: due Wednesday, 2016-02-17, at 5:00pm . . . . 377
C.1.1 Sharing the wealth . . . . . . . . . . . . . . . . . . . . 377
C.1.2 Eccentricity . . . . . . . . . . . . . . . . . . . . . . . . 380
C.1.3 Leader election on an augmented ring . . . . . . . . . 383
C.2 Assignment 2: due Wednesday, 2016-03-09, at 5:00pm . . . . 383
C.2.1 A rotor array . . . . . . . . . . . . . . . . . . . . . . . 383
C.2.2 Set registers . . . . . . . . . . . . . . . . . . . . . . . . 385
C.2.3 Bounded failure detectors . . . . . . . . . . . . . . . . 386
C.3 Assignment 3: due Wednesday, 2016-04-20, at 5:00pm . . . . 387
C.3.1 Fetch-and-max . . . . . . . . . . . . . . . . . . . . . . 387
C.3.2 Median . . . . . . . . . . . . . . . . . . . . . . . . . . 388
C.3.3 Randomized two-process test-and-set with small registers390
C.4 Presentation (for students taking CPSC 565): due Wednesday,
2016-04-27 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 392
C.5 CS465/CS565 Final Exam, May 10th, 2016 . . . . . . . . . . 393
C.5.1 A slow register (20 points) . . . . . . . . . . . . . . . . 393
C.5.2 Two leaders (20 points) . . . . . . . . . . . . . . . . . 394
C.5.3 A splitter using one-bit registers (20 points) . . . . . . 395
C.5.4 Symmetric self-stabilizing consensus (20 points) . . . . 396

D Sample assignments from Spring 2014 398

D.1 Assignment 1: due Wednesday, 2014-01-29, at 5:00pm . . . . 398
D.1.1 Counting evil processes . . . . . . . . . . . . . . . . . 398
D.1.2 Avoiding expensive processes . . . . . . . . . . . . . . 399
D.2 Assignment 2: due Wednesday, 2014-02-12, at 5:00pm . . . . 401
D.2.1 Synchronous agreement with weak failures . . . . . . . 401
D.2.2 Byzantine agreement with contiguous faults . . . . . . 402
D.3 Assignment 3: due Wednesday, 2014-02-26, at 5:00pm . . . . 403
D.3.1 Among the elect . . . . . . . . . . . . . . . . . . . . . 403
D.3.2 Failure detectors on the cheap . . . . . . . . . . . . . . 404
D.4 Assignment 4: due Wednesday, 2014-03-26, at 5:00pm . . . . 405
D.4.1 A global synchronizer with a global clock . . . . . . . 405
D.4.2 A message-passing counter . . . . . . . . . . . . . . . 406
D.5 Assignment 5: due Wednesday, 2014-04-09, at 5:00pm . . . . 407
D.5.1 A concurrency detector . . . . . . . . . . . . . . . . . 407
D.5.2 Two-writer sticky bits . . . . . . . . . . . . . . . . . . 409
D.6 Assignment 6: due Wednesday, 2014-04-23, at 5:00pm . . . . 410
D.6.1 A rotate register . . . . . . . . . . . . . . . . . . . . . 410
D.6.2 A randomized two-process test-and-set . . . . . . . . . 412
CONTENTS xiv

D.7 CS465/CS565 Final Exam, May 2nd, 2014 . . . . . . . . . . . 414

D.7.1 Maxima (20 points) . . . . . . . . . . . . . . . . . . . 414
D.7.2 Historyless objects (20 points) . . . . . . . . . . . . . 415
D.7.3 Hams (20 points) . . . . . . . . . . . . . . . . . . . . . 415
D.7.4 Mutexes (20 points) . . . . . . . . . . . . . . . . . . . 417

E Sample assignments from Fall 2011 419

E.1 Assignment 1: due Wednesday, 2011-09-28, at 17:00 . . . . . 419
E.1.1 Anonymous algorithms on a torus . . . . . . . . . . . 419
E.1.2 Clustering . . . . . . . . . . . . . . . . . . . . . . . . . 420
E.1.3 Negotiation . . . . . . . . . . . . . . . . . . . . . . . . 421
E.2 Assignment 2: due Wednesday, 2011-11-02, at 17:00 . . . . . 422
E.2.1 Consensus with delivery notifications . . . . . . . . . . 422
E.2.2 A circular failure detector . . . . . . . . . . . . . . . . 423
E.2.3 An odd problem . . . . . . . . . . . . . . . . . . . . . 425
E.3 Assignment 3: due Friday, 2011-12-02, at 17:00 . . . . . . . . 426
E.3.1 A restricted queue . . . . . . . . . . . . . . . . . . . . 426
E.3.2 Writable fetch-and-increment . . . . . . . . . . . . . . 427
E.3.3 A box object . . . . . . . . . . . . . . . . . . . . . . . 428
E.4 CS465/CS565 Final Exam, December 12th, 2011 . . . . . . . 429
E.4.1 Lockable registers (20 points) . . . . . . . . . . . . . . 429
E.4.2 Byzantine timestamps (20 points) . . . . . . . . . . . 430
E.4.3 Failure detectors and k-set agreement (20 points) . . . 431
E.4.4 A set data structure (20 points) . . . . . . . . . . . . . 432

F Additional sample final exams 433

F.1 CS425/CS525 Final Exam, December 15th, 2005 . . . . . . . 433
F.1.1 Consensus by attrition (20 points) . . . . . . . . . . . 433
F.1.2 Long-distance agreement (20 points) . . . . . . . . . . 434
F.1.3 Mutex appendages (20 points) . . . . . . . . . . . . . 436
F.2 CS425/CS525 Final Exam, May 8th, 2008 . . . . . . . . . . . 437
F.2.1 Message passing without failures (20 points) . . . . . . 437
F.2.2 A ring buffer (20 points) . . . . . . . . . . . . . . . . . 437
F.2.3 Leader election on a torus (20 points) . . . . . . . . . 438
F.2.4 An overlay network (20 points) . . . . . . . . . . . . . 439
F.3 CS425/CS525 Final Exam, May 10th, 2010 . . . . . . . . . . 440
F.3.1 Anti-consensus (20 points) . . . . . . . . . . . . . . . . 440
F.3.2 Odd or even (20 points) . . . . . . . . . . . . . . . . . 441
F.3.3 Atomic snapshot arrays using message-passing (20 points)441
F.3.4 Priority queues (20 points) . . . . . . . . . . . . . . . 442
CONTENTS xv

G I/O automata 444

G.1 Low-level view: I/O automata . . . . . . . . . . . . . . . . . . 444
G.1.1 Enabled actions . . . . . . . . . . . . . . . . . . . . . . 444
G.1.2 Executions, fairness, and traces . . . . . . . . . . . . . 445
G.1.3 Composition of automata . . . . . . . . . . . . . . . . 445
G.1.4 Hiding actions . . . . . . . . . . . . . . . . . . . . . . 446
G.1.5 Fairness . . . . . . . . . . . . . . . . . . . . . . . . . . 446
G.1.6 Specifying an automaton . . . . . . . . . . . . . . . . 447
G.2 High-level view: traces . . . . . . . . . . . . . . . . . . . . . . 447
G.2.1 Example . . . . . . . . . . . . . . . . . . . . . . . . . . 448
G.2.2 Types of trace properties . . . . . . . . . . . . . . . . 448
G.2.2.1 Safety properties . . . . . . . . . . . . . . . . 448
G.2.2.2 Liveness properties . . . . . . . . . . . . . . . 449
G.2.2.3 Other properties . . . . . . . . . . . . . . . . 450
G.2.3 Compositional arguments . . . . . . . . . . . . . . . . 450
G.2.3.1 Example . . . . . . . . . . . . . . . . . . . . 451
G.2.4 Simulation arguments . . . . . . . . . . . . . . . . . . 451
G.2.4.1 Example . . . . . . . . . . . . . . . . . . . . 452

Bibliography 453

Index 478
List of Figures

2.1 Asynchronous message-passing execution . . . . . . . . . . . . 13

2.2 Asynchronous message-passing execution with FIFO channels 13
2.3 Synchronous message-passing execution . . . . . . . . . . . . 14
2.4 Asynchronous time . . . . . . . . . . . . . . . . . . . . . . . . 15

5.1 Labels in the bit-reversal ring with n = 32 . . . . . . . . . . . 40

10.1 Synthetic execution for Byzantine agreement lower bound . . 70

10.2 Synthetic execution for Byzantine agreement connectivity . . 71

12.1 Example execution of Paxos . . . . . . . . . . . . . . . . . . . 90

13.1 Failure detector classes . . . . . . . . . . . . . . . . . . . . . . 98

14.1 Figure 2 from [NW98] . . . . . . . . . . . . . . . . . . . . . . 107

21.1 Snapshot from max arrays; see also [AACHE15, Fig. 2] . . . . 195

24.1 A 6 × 6 Moir-Anderson grid . . . . . . . . . . . . . . . . . . . 231

24.2 Path through a Moir-Anderson grid . . . . . . . . . . . . . . 232
24.3 A sorting network . . . . . . . . . . . . . . . . . . . . . . . . 236

28.1 Subdivision corresponding to one round of immediate snapshot281

28.2 Subdivision corresponding to two rounds of immediate snapshot282
28.3 An attempt at 2-set agreement . . . . . . . . . . . . . . . . . 284
28.4 Output complex for renaming with n = 3, m = 4 . . . . . . . 291

D.1 Connected Byzantine nodes take over half a cut . . . . . . . . 402

xvi
List of Tables

18.1 Position of various types in the wait-free hierarchy . . . . . . 149

xvii
List of Algorithms

2.1 Client-server computation: client code . . . . . . . . . . . . . . 11

2.2 Client-server computation: server code . . . . . . . . . . . . . 11

3.1 Basic flooding algorithm . . . . . . . . . . . . . . . . . . . . . 17

3.2 Flooding with parent pointers . . . . . . . . . . . . . . . . . . 18
3.3 Convergecast . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.4 Flooding and convergecast combined . . . . . . . . . . . . . . . 22

4.1 AsynchBFS algorithm (from [Lyn96]) . . . . . . . . . . . . . . 24

5.1 LCR leader election . . . . . . . . . . . . . . . . . . . . . . . . 32

5.2 Peterson’s leader-election algorithm . . . . . . . . . . . . . . . 35

10.1 Exponential information gathering . . . . . . . . . . . . . . . . 74

10.2 Byzantine agreement: phase king . . . . . . . . . . . . . . . . . 77

12.1 Paxos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

13.1 Boosting completeness . . . . . . . . . . . . . . . . . . . . . . . 96

13.2 Consensus with a strong failure detector . . . . . . . . . . . . . 99
13.3 Reliable broadcast . . . . . . . . . . . . . . . . . . . . . . . . . 101

17.1 Mutual exclusion using test-and-set . . . . . . . . . . . . . . . 128

17.2 Mutual exclusion using a queue . . . . . . . . . . . . . . . . . 129
17.3 Mutual exclusion using read-modify-write . . . . . . . . . . . . 131
17.4 Peterson’s mutual exclusion algorithm for two processes . . . . 132
17.5 Implementation of a splitter . . . . . . . . . . . . . . . . . . . 136
17.6 Lamport’s Bakery algorithm . . . . . . . . . . . . . . . . . . . 138
17.7 Yang-Anderson mutex for two processes . . . . . . . . . . . . . 143

18.1 Determining the winner of a race between 2-register writes . . 155

xviii
LIST OF ALGORITHMS xix

18.2 A universal construction based on consensus . . . . . . . . . . 159

19.1 Snapshot of [AAD+ 93] using unbounded registers . . . . . . . 163

19.2 Lattice agreement snapshot . . . . . . . . . . . . . . . . . . . . 169
19.3 Update for lattice agreement snapshot . . . . . . . . . . . . . . 170
19.4 Increasing set data structure . . . . . . . . . . . . . . . . . . . 173
19.5 Single-scanner snapshot: scan . . . . . . . . . . . . . . . . . . 176
19.6 Single-scanner snapshot: update . . . . . . . . . . . . . . . . . 176

21.1 Max register read operation . . . . . . . . . . . . . . . . . . . . 186

21.2 Max register write operations . . . . . . . . . . . . . . . . . . . 187
21.3 Recursive construction of a 2-component max array . . . . . . 192

22.1 Building 2-process TAS from 2-process consensus . . . . . . . . 198

22.2 Two-process one-shot swap from TAS . . . . . . . . . . . . . . 199
22.3 Tournament algorithm with gate . . . . . . . . . . . . . . . . . 200
22.4 Obstruction-free swap from test-and-set . . . . . . . . . . . . . 201
22.5 Wait-free swap from test-and-set [AMW11] . . . . . . . . . . . 204

23.1 Consensus using adopt-commit . . . . . . . . . . . . . . . . . . 211

23.2 A 2-valued adopt-commit object . . . . . . . . . . . . . . . . . 212
23.3 Shared coin conciliator from [Asp12b] . . . . . . . . . . . . . . 213
23.4 Impatient first-mover conciliator from [Asp12b] . . . . . . . . . 214
23.5 A sifter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
23.6 Test-and-set in O(log log n) expected time . . . . . . . . . . . . 219
23.7 Sifting conciliator (from [Asp12a]) . . . . . . . . . . . . . . . . 220
23.8 Giakkoupis-Woelfel sifter [GW12a] . . . . . . . . . . . . . . . . 221

24.1 Wait-free deterministic renaming . . . . . . . . . . . . . . . . . 227

24.2 Releasing a name . . . . . . . . . . . . . . . . . . . . . . . . . 229
24.3 Implementation of a splitter . . . . . . . . . . . . . . . . . . . 230

25.1 Overlapping LL/SC . . . . . . . . . . . . . . . . . . . . . . . . 243

26.1 Obstruction-free 2-process test-and-set . . . . . . . . . . . . . 250

26.2 Obstruction-free deque . . . . . . . . . . . . . . . . . . . . . . 252
26.3 Obstruction-freedom booster from [FLMS05] . . . . . . . . . . 255

27.1 Safe agreement (adapted from [BGLR01]) . . . . . . . . . . . . 268

28.1 Participating set . . . . . . . . . . . . . . . . . . . . . . . . . . 288

LIST OF ALGORITHMS xx

29.1 Approximate agreement . . . . . . . . . . . . . . . . . . . . . . 293

31.1 Dijkstra’s large-state token ring algorithm [Dij74] . . . . . . . 301

35.1 Beeping a maximal independent set (from [AABJ+ 11] . . . . . 338

A.1 Recruiting algorithm for Problem A.1.2. . . . . . . . . . . . . . 344

A.2 Candidate algorithm for asynchronous agreement . . . . . . . . 347
A.3 Committee election using ABD . . . . . . . . . . . . . . . . . . 349
A.4 An alleged counter. Code for process i. . . . . . . . . . . . . . 350
A.5 Implementation of a rock-paper-scissors object . . . . . . . . . 353

B.1 Reporting Alice’s alarming messages . . . . . . . . . . . . . . . 359

B.2 Shutdown mechanism based on Chandy-Lamport . . . . . . . . 362
B.3 Consensus from totally-ordered partial broadcast . . . . . . . . 365
B.4 Peterson’s mutual exclusion algorithm using a counter . . . . . 367
B.5 A 2-bounded counter . . . . . . . . . . . . . . . . . . . . . . . 369

C.1 Computing eccentricity in a tree . . . . . . . . . . . . . . . . . 381

C.2 Rotor array . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384
C.3 Two-process consensus using a rotor array . . . . . . . . . . . . 384
C.4 Max register modified to use a test-and-set bit . . . . . . . . . 387
C.5 Randomized one-shot test-and-set for two processes . . . . . . 390
C.6 Splitter using one-bit registers . . . . . . . . . . . . . . . . . . 396

D.1 Counter algorithm for Problem D.4.2. . . . . . . . . . . . . . . 407

D.2 Two-process consensus using the object from Problem D.5.1 . 408
D.3 Implementation of a rotate register . . . . . . . . . . . . . . . 411
D.4 Randomized two-process test-and-set for D.6.2 . . . . . . . . . 412
D.5 Mutex using a swap object and register . . . . . . . . . . . . . 417

E.1 Resettable fetch-and-increment . . . . . . . . . . . . . . . . . . 428

E.2 Consensus using a lockable register . . . . . . . . . . . . . . . 429
E.3 Timestamps with n ≥ 3 and one Byzantine process . . . . . . . 431
E.4 Counter from set object . . . . . . . . . . . . . . . . . . . . . . 432

G.1 Spambot as an I/O automaton . . . . . . . . . . . . . . . . . . 447

Preface

These are notes for the Fall 2020 semester version of the Yale course CPSC
465/565 Theory of Distributed Systems. This document also incorporates the
lecture schedule and assignments, as well as some sample assignments from
previous semesters. Because this is a work in progress, it will be updated
frequently over the course of the semester.
The most recent version of these notes will be available at http://www.cs.
yale.edu/homes/aspnes/classes/465/notes.pdf. More stable archival
versions may be found at https://arxiv.org/abs/2001.04235.
Not all topics in the notes will be covered during this semester. Some
chapters have not been updated and are marked as possibly out of date.
Much of the structure of the course follows the textbook, Attiya and
Welch’s Distributed Computing [AW04], with some topics based on Lynch’s
Distributed Algorithms [Lyn96] and additional readings from the research
literature. In most cases you’ll find these materials contain much more detail
than what is presented here, so it may be better to consider this document a
supplement to them than to treat it as your primary source of information.

Acknowledgments
Many parts of these notes were improved by feedback from students taking
various versions of this course, as well as others who have kindly pointed
out errors in the notes after reading them online. Many of these suggestions,
sadly, went unrecorded, so I must apologize to the many students who
should be thanked here but whose names I didn’t keep track of in the past.
However, I can thank Mike Marmar and Hao Pan in particular for suggesting
improvements to some of the posted solutions, Guy Laden for suggesting
corrections to Figure 12.1, and Ali Mamdouh for pointing out an error in
the original presentation of Algorithm 5.2.

xxi
Lecture schedule

As always, the future is uncertain, so you should take parts of the schedule
that haven’t happened yet with a grain of salt. Unless otherwise specified,
readings refer to chapters or sections in the course notes.

2020-09-01 What is distributed computing and why do we have a theory of

it? Basic models: message passing, shared memory, local interactions.
Configurations, events, executions, and schedules. The adversary. Basic
message-passing model. A simple flooding protocol. Safety properties
and invariants. Readings: Chapters 1 and 2.

2020-09-03 Liveness properties. Fairness and performance measures. Safety

and liveness. Drawing message-passing executions. Broadcast and con-
vergecast. Synchronous vs. asynchronous message-passing. Readings:
Rest of Chapter 2, Chapter 3.

2020-09-08 Distributed breadth-first search. Start of leader election. Read-

ings: Chapters 4 and 5 through §5.1.

2020-09-10 More leader election. Readings: Rest of Chapter 5.

2020-09-15 Causal ordering, logical clocks, and snapshots. Readings: Chap-

ter 6.

2020-09-17 Synchronizers. Readings: Chapter 7.

2020-09-22 Synchronous agreement with crash failures. Impossibility of

Byzantine agreement with n/3 faults. Readings: Chapter 8 (except
§8.3), Chapter 9, §10.1.2.

2020-09-24 Phase king algorithm for synchronous Byzantine agreement.

Impossibility of asynchronous agreement with one crash failure. Read-
ings: §10.2.2, Chapter 11.

xxii
LECTURE SCHEDULE xxiii

2020-09-29 Paxos. Readings: Chapter 12.

2020-10-01 Failure detectors. Readings: Chapter 13.
2020-10-06 Shared memory: model, linearizability, emulating message
passing using shared memory, the ABD algorithm for distributed
shared memory. Readings: Chapters 15 and 16.
2020-10-08 Start of mutual exclusion: description of the problem, algo-
rithms for various models. Readings: Chapter 17 through §17.4.1.
2020-10-13 More mutual exclusion: Fast mutual exclusion using splitters,
Burns-Lynch space lower bound, RMR complexity. Readings: §§17.4.2,
17.4.4, and 17.5.
2020-10-15 Wait-free computation and the wait-free hierarchy (levels 1, 2,
and ∞). Readings: Chapter 18 except §§18.1.4 and 18.2; [Her91b].
2020-10-20 More wait-free stuff: Intermediate levels of the wait-free hierar-
chy. Herlihy’s universal construction and helping. Readings: Rest of
Chapter 18.
2020-10-22 Atomic snapshots of shared memory. Atomic snapshots in
O(n2 ) individual steps using double collects with helping. Reduction
to lattice agreement. Readings: Chapter 19 through §19.3.4.
2020-10-27 Implementing lattice agreement. The Jayanti-Tan-Toueg lower
bound. Readings: §19.3.5, Chapter 20.
2020-10-29 Restricted-use objects. Readings: Chapter 21.
2020-11-03 Randomized consensus: adopt-commit objects, shared coins,
Chor-Israeli-Li and variants. Readings: Chapter 23 through §23.6.
2020-11-05 Randomized renaming and test-and-set for an adaptive adver-
sary. Randomized test-and-set and consensus for an oblivious adversary.
Deterministic renaming using splitters. Readings: Rest of Chapter 23,
§24.4.3; [AAG+ 10, AA11, GW12a], [MA95].
2020-11-10 BG simulation of n process with f failures by f + 1 pro-
cesses with f failures. Topological methods for testing solvability
of asynchronous decision tasks. Readings: Chapter 27, Chapter 28;
[BG93, BGLR01]; AW §16.1 if you want to see a non-topological proof
of the k-set agreement result, [BG97] for iterated immediate snapshots,
[HS99] for more about the topological approach.
LECTURE SCHEDULE xxiv

2020-11-12 Self-stabilization and local computation: Dijkstra’s token ring

circulation algorithm, self-stabilizing synchronizers and BFS trees,
relation between local algorithms and self-stabilization. Readings:
Chapter 31; [Dij74, AKM+ 93, LSW09].

2020-11-17 Distributed graph algorithms. The LOCAL and CONGEST

models. Local graph coloring. Readings: Chapter 32.

2020-11-19 Population protocols. Readings: Chapter 33.

2020-12-01 CPSC 565 student presentations. See §A.6.

2020-12-03 CPSC 565 student presentations.

Chapter 1

Introduction

Distributed systems are characterized by their structure: a typical dis-

tributed system will consist of some large number of interacting devices that
each run their own programs but that are affected by receiving messages, or
observing shared-memory updates or the states of other devices. Examples
of distributed systems range from simple systems in which a single client
talks to a single server to huge amorphous networks like the Internet as a
whole.
As distributed systems get larger, it becomes harder and harder to
predict or understand their behavior. Part of the reason for this is that
we as programmers have not yet developed a standardized set of tools for
managing complexity (like subroutines or objects with narrow interfaces,
or even simple structured programming mechanisms like loops or if/then
statements) as are found in sequential programming. Part of the reason is
that large distributed systems bring with them large amounts of inherent
nondeterminism—unpredictable events like delays in message arrivals, the
sudden failure of components, or in extreme cases the nefarious actions of
faulty or malicious machines opposed to the goals of the system as a whole.
Because of the unpredictability and scale of large distributed systems, it can
often be difficult to test or simulate them adequately. Thus there is a need
for theoretical tools that allow us to prove properties of these systems that
will let us use them with confidence.
The first task of any theory of distributed systems is modeling: defining
a mathematical structure that abstracts out all relevant properties of a large
distributed system. There are many foundational models in the literature for
distributed systems, but for this class we will follow [AW04] and use simple
automaton-based models.

1
CHAPTER 1. INTRODUCTION 2

What this means is that we model each process in the system as an

automaton that has some sort of local state, and model local computation
as a transition rule that tells us how to update this state in response to
various events. Depending on what kinds of system we are modeling, these
events might correspond to local computation, to delivery of a message by a
network, carrying out some operation on a shared memory, or even something
like a chemical reaction between two molecules. The transition rule for a
system specifies how the states of all processes involved in the event are
updated, based on their previous states. We can think of the transition
rule as an arbitrary mathematical function (or relation if the processes are
nondeterministic); this corresponds in programming terms to implementing
local computation by processes as a gigantic table lookup.
Obviously this is not how we program systems in practice. But what this
approach does is allow us to abstract away completely from how individual
processes work, and emphasize how all of the processes interact with each
other. This can lead to odd results: for example, it’s perfectly consistent
with this model for some process to be able to solve the halting problem, or
carry out arbitrarily complex calculations between receiving a message and
sending its response. A partial justification for this assumption is that in
practice, the multi-millisecond latencies in even reasonably fast networks are
eons in terms of local computation. And as with any assumption, we can
always modify it if it gets us into trouble.

1.1 Models
The global state consisting of all process states is called a configuration,
and we think of the system as a whole as passing from one global state
or configuration to another in response to each event. When this occurs
the processes participating in the event update their states, and the other
processes do nothing. This does not model concurrency directly; instead,
we interleave potentially concurrent events in some arbitrary way. The
advantage of this interleaving approach is that it gives us essentially the
same behavior as we would get if we modeled simultaneous events explicitly,
but still allows us to consider only one event at a time and use induction to
prove various properties of the sequence of configurations we might reach.
We will often use lowercase Greek letters for individual events or sequences
of events. Configurations are typically written as capital Latin letters (often
C). An execution of a schedule is an alternating sequence of configurations
and events C0 σ0 C1 σ1 C2 . . . , where Ci+1 is the configuration that results from
CHAPTER 1. INTRODUCTION 3

applying event σi to configuration C. A schedule is a sequence of events

σ0 σ1 . . . from some execution. We say that an event σ is enabled in C if
this event can be carried out in C; an example would be that the event that
we deliver a particular message in a message-passing system is enabled only
if that message has been sent and not yet delivered. When σ is enabled in
C, it is sometime convenient to write Cσ for the configuration that results
from applying σ to C.
What events are available, and what effects they have, will depend
on what kind of model we are considering. We may also have additional
constraints on what kinds of schedules are admissible, which restricts the
schedules under consideration to those that have certain desirable properties
(say, every message that is sent is eventually delivered). There are many
models in the distributed computing literature, which can be divided into a
handful of broad categories:

• Message passing models (which we will cover in Part I) correspond

to systems where processes communicate by sending messages through
a network. In synchronous message-passing, every process sends
out messages at time t that are delivered at time t + 1, at which point
more messages are sent out that are delivered at time t + 2, and so
on: the whole system runs in lockstep, marching forward in perfect
synchrony.1 Such systems are difficult to build when the components
become too numerous or too widely dispersed, but they are often
easier to analyze than asynchronous systems, where messages are
only delivered eventually after some unknown delay. Variants on these
models include semi-synchronous systems, where message delays are
unpredictable but bounded, and various sorts of timed systems. Further
variations come from restricting which processes can communicate with
which others, by allowing various sorts of failures: crash failures
that stop a process dead, Byzantine failures that turn a process
evil, or omission failures that drop messages in transit. Or—on the
helpful side—we may supply additional tools like failure detectors
(Chapter 13) or randomization (Chapter 23).

• Shared-memory models (Part II) correspond to systems where pro-

cesses communicate by executing operations on shared objects
1
In an interleaving model, these apparently simultaneous events are still recorded one
at a time. What makes the system synchronous is that we demand that, in any admissible
schedule, all n events for time t occur as a sequential block, followed by all n events for
time t + 1, and so on.
CHAPTER 1. INTRODUCTION 4

In the simplest case, the objects are simple memory cells supporting
read and write operations. These are called atomic registers. But
in general, the objects could be more complex hardware primitives
like compare-and-swap (§18.1.3), load-linked/store-conditional
(§18.1.3), atomic queues, or even more exotic objects from the seldom-
visited theoretical depths.
Practical shared-memory systems may be implemented as distributed
shared-memory (Chapter 16) on top of a message-passing system.
This gives an alternative approach to designing message-passing systems
if it turns out that shared memory is easier to use for a particular
problem.
Like message-passing systems, shared-memory systems must also deal
with issues of asynchrony and failures, both in the processes and in the
shared objects.
Realistic shared-memory systems have additional complications, in that
modern CPUs allow out-of-order execution in the absence of special
(and expensive) operations called fences or memory barriers.[AG95]
We will effectively be assuming that our shared-memory code is liberally
sprinkled with these operations so that nothing surprising happens,
but this is not always true of real production code, and indeed there is
work in the theory of distributed computing literature on algorithms
that don’t require unlimited use of memory barriers.

• A third family of models has no communication mechanism indepen-

dent of the processes. Instead, the processes may directly observe
the states of other processes. These models are used in analyzing
self-stabilization, for some biologically inspired systems, and
for computation by population protocols or chemical reaction
networks. We will discuss some of this work in Part III.

• Other specialized models emphasize particular details of distributed

systems, such as the labeled-graph models used for analyzing routing or
the topological models used to give a very high-level picture of various
distributed decision problems (see Chapter 28).

We’ll see many of these at some point in this course, and examine which
of them can simulate each other under various conditions.
CHAPTER 1. INTRODUCTION 5

1.2 Properties
Properties we might want to prove about a system include:

• Safety properties, of the form “nothing bad ever happens” or, more
precisely, “there are no bad reachable configurations.” These include
things like “at most one of the traffic lights at the intersection of Busy
Road and Main Street is ever green” or “every value read from a counter
equals the number of preceding increment operations.” Such properties
are typically proved using an , a property of configurations that is true
initially and that is preserved by all transitions (this is essentially a
disguised induction proof).

• Liveness properties, of the form “something good eventually happens.”

An example might be “my email is eventually either delivered or
returned to me.” These are not properties of particular states (I might
unhappily await the eventual delivery of my email for decades without
violating the liveness property just described), but of executions, where
the property must hold starting at some finite time. Liveness properties
are generally proved either from other liveness properties (e.g., “all
messages in this message-passing system are eventually delivered”)
or from a combination of such properties and some sort of timer
argument where some progress metric improves with every transition
and guarantees the desirable state when it reaches some bound (also a
disguised induction proof).

• Fairness properties are a strong kind of liveness property of the form

“something good eventually happens to everybody.” Such properties
exclude starvation, a situation where most of the kids are happily
chowing down at the orphanage (“some kid eventually eats something”
is a liveness property) but poor Oliver Twist is dying in the corner for
lack of gruel.

• Simulations show how to build one kind of system from another,

such as a reliable message-passing system built on top of an unreliable
system (TCP [Pos81]), a shared-memory system built on top of a
message-passing system (distributed shared memory—see Chapter 16),
or a synchronous system build on top of an asynchronous system
(synchronizers—see Chapter 7).

• Impossibility results describe things we can’t do. For example, the

classic Two Generals impossibility result (Chapter 8) says that it’s
CHAPTER 1. INTRODUCTION 6

impossible to guarantee agreement between two processes across an

unreliable message-passing channel if even a single message can be
lost. Other results characterize what problems can be solved if various
fractions of the processes are unreliable, or if asynchrony makes timing
assumptions impossible. These results, and similar lower bounds that
describe things we can’t do quickly, include some of the most technically
sophisticated results in distributed computing. They stand in contrast
to the situation with sequential computing, where the reliability and
predictability of the underlying hardware makes proving lower bounds
extremely difficult.

There are some basic proof techniques that we will see over and over
again in distributed computing.
For lower bound and impossibility proofs, the main tool is the in-
distinguishability argument. Here we construct two (or more) executions
in which some process has the same input and thus behaves the same way,
regardless of what algorithm it is running. This exploitation of process’s ig-
norance is what makes impossibility results possible in distributed computing
despite being notoriously difficult in most areas of computer science.2
For safety properties, statements that some bad outcome never occurs,
the main proof technique is to construct an invariant. An invariant is
essentially an induction hypothesis on reachable configurations of the system;
an invariant proof shows that the invariant holds in all initial configurations,
and that if it holds in some configuration, it holds in any configuration that
is reachable in one step.
Induction is also useful for proving termination and liveness properties,
statements that some good outcome occurs after a bounded amount of time.
Here we typically structure the induction hypothesis as a progress measure,
showing that some sort of partial progress holds by a particular time, with
the full guarantee implied after the time bound is reached.

2
An exception might be lower bounds for data structures, which also rely on a process’s
ignorance.
 Part I

Message passing

7
Chapter 2

Model

Message passing models simulate networks. Because any interaction between

physically separated processors requires transmitting information from one
place to another, all distributed systems are, at a low enough level, message-
passing systems. We start by defining a formal model of these systems.

2.1 Basic message-passing model

We have a collection of n processes p1 . . . p2 , each of which has a state
consisting of a state from from state set Qi . We think of these processes
as nodes in a directed communication graph or network. The edges in
this graph are a collection of point-to-point channels or buffers bij , one
for each pair of adjacent processes i and j, representing messages that have
been sent but that have not yet been delivered. Implicit in this definition is
that messages are point-to-point, with a single sender and recipient: if you
want broadcast, you have to build it yourself.
A configuration of the system consists of a vector of states, one for each
process and channel. The configuration of the system is updated by an event,
in which (1) zero or more messages in channels bij are delivered to process pj ,
removing them from bij ; (2) pj updates its state in response; and (3) zero or
more messages are added by pj to outgoing channels bji . We generally think
of these events as delivery events when at least one message is delivered,
and as computation events when none are. An execution segment is a
sequence of alternating configurations and events C0 , φ1 , C1 , φ2 , . . . , in which
each triple Ci φi+1 Ci+1 is consistent with the transition rules for the event
φi+1 , and the last element of the sequence (if any) is a configuration. If the
first configuration C0 is an initial configuration of the system, we have an

8
CHAPTER 2. MODEL 9

execution. A schedule is an execution with the configurations removed.

2.1.1 Formal details

Let P be the set of processes, Q the set of process states, and M the set of
possible messages.
Each process pi has a state statei ∈ Q. Each channel bij has a state
bufferij ∈ P(M ). We assume each process has a transition function
δ : Q × P(M ) → Q × P(P × M ) that maps tuples consisting of a state and
a set of incoming messages a new state and a set of recipients and messages
to be sent. An important feature of the transition function is that the
process’s behavior can’t depend on which of its previous messages have been
delivered or not. A delivery event del(i, A), where A = {(jk , mk )}) removes
each message mk from bji , updates statei according to δ(statei , A), and adds
the outgoing messages specified to δ(statei , A) to the appropriate channels.
A computation event comp(i) does the same thing, except that it applies
δ(statej , ∅).
Some implicit features in this definition:
• A process can’t tell when its outgoing messages are delivered, because
the channel states aren’t available as input to δ.

• Processes are deterministic: The next action of each process depends

only on its current state, and not on extrinsic variables like the phase
of the moon, coin-flips, etc. We may wish to relax this condition later
by allowing coin-flips; to do so, we will need to extend the model to
incorporate probabilities.

• It is possible to determine the accessible state of a process by looking

only at events that involve that process. Specifically, given a schedule
S, define the restriction S|i to be the subsequence consisting of all
comp(i) and del(i, A) events (ranging over all possible A). Since these
are the only events that affect the state of i, and only the state of i is
needed to apply the transition function, we can compute the state of i
looking only at S|i. In particular, this means that i will have the same
accessible state after any two schedules S and S 0 where S|i = S 0 |i, and
thus will take the same actions in both schedules. This is the basis for
indistinguishability proofs (§8.2), a central technique in obtaining
lower bounds and impossibility results.
Attiya and Welch [AW04] use a different model in which communication
channels are not modeled separately from processes, but instead are baked
CHAPTER 2. MODEL 10

into processes as outbuf and inbuf variables. This leads to some oddities like
having to distinguish the accessible state of a process (which excludes the
outbufs) from the full state (which doesn’t). Our approach is close to that of
Lynch [Lyn96], in that we have separate automata representing processes and
communication channels. But since the resulting model produces essentially
the same executions, the exact details don’t really matter.1

2.1.2 Network structure

It may be the case that not all processes can communicate directly; if so,
we impose a network structure in the form of a directed graph, where i can
send a message to j if and only if there is an edge from i to j in the graph.
Typically we assume that each process knows the identity of all its neighbors.
For some problems (e.g., in peer-to-peer systems or other overlay net-
works) it may be natural to assume that there is a fully-connected underlying
network but that we have a dynamic network on top of it, where processes
can only send to other processes that they have obtained the addresses of in
some way.

2.2 Asynchronous systems

In an asynchronous model, only minimal restrictions are placed on when
messages are delivered and when local computation occurs. A schedule is
said to be admissible if (a) there are infinitely many computation steps
for each process, and (b) every message is eventually delivered. (These are
fairness conditions.) The first condition (a) assumes that processes do not
explicitly terminate, which is the assumption used in [AW04]; an alternative,
which we will use when convenient, is to assume that every process either
has infinitely many computation steps or reaches an explicit halting state.
1
The late 1970s and early 1980s saw a lot of research on finding the “right” definition
of a distributed system, and some of the disputes from that era were hard fought. But in
the end, all the various proposed models turned out to be more or less equivalent, which
is not surprising since the authors were ultimately trying to represent the same intuitive
understanding of these systems. So most distributed computing papers now just use some
phrasing like “we consider the standard model of an asynchronous message-passing system”
and leave it to the reader to assume that this standard model is their favorite one.
An example of this trick in action is that you will never see del(i, A) or comp(i) again
after you finish reading this footnote.
CHAPTER 2. MODEL 11

2.2.1 Example: client-server computing

Almost every distributed system in practical use is based on client-server
interactions. Here one process, the client, sends a request to a second
process, the server, which in turn sends back a response. We can model
this interaction using our asynchronous message-passing model by describing
what the transition functions for the client and the server look like: see
Algorithms 2.1 and 2.2.

1 initially do
2 send request to server
Algorithm 2.1: Client-server computation: client code

1 upon receiving request do

2 send response to client
Algorithm 2.2: Client-server computation: server code

The interpretation of Algorithm 2.1 is that the client sends request (by
adding it to its outbuf) in its very first computation event (after which it does
nothing). The interpretation of Algorithm 2.2 is that in any computation
event where the server observes request in its inbuf, it sends response.
We want to claim that the client eventually receives response in any
admissible execution. To prove this, observe that:
1. After finitely many steps, the client carries out a computation event.
This computation event puts request in its outbuf.
2. After finitely many more steps, a delivery event occurs that delivers
request to the server. This causes the server to send response.
3. After finitely many more steps, a delivery event delivers response to
the client, causing it to process response (and do nothing, given that
we haven’t included any code to handle this response).
Each step of the proof is justified by the constraints on admissible
executions. If we could run for infinitely many steps without a particular
process doing a computation event or a particular message being delivered,
we’d violate those constraints.
Most of the time we will not attempt to prove the correctness of a
protocol at quite this level of tedious detail. But if you are only interested in
CHAPTER 2. MODEL 12

distributed algorithms that people actually use, you have now seen a proof
of correctness for 99.9% of them, and do not need to read any further.

2.3 Synchronous systems

A synchronous message-passing system is exactly like an asynchronous
system, except we insist that the schedule consists of alternating phases in
which (a) every process executes a computation step, and (b) all messages
are delivered while none are sent.2 The combination of a computation phase
and a delivery phase is called a round. Synchronous systems are effectively
those in which all processes execute in lock-step, and there is no timing
uncertainty. This makes protocols much easier to design, but makes them
less resistant to real-world timing oddities. Sometimes this can be dealt with
by applying a synchronizer (Chapter 7), which transforms synchronous
protocols into asynchronous protocols at a small cost in complexity.

2.4 Drawing message-passing executions

Though formally we can describe an execution in a message-passing system
as a long list of events, this doesn’t help much with visualizing the underlying
communication pattern. So it can sometimes be helpful to use a more visual
representation of a message-passing execution that shows how information
flows through the system.
A typical example is given in Figure 2.1. In this picture, time flows
from left to right, and each process is represented by a horizontal line. This
convention reflects the fact that processes have memory, so any information
available to a process at some time t is also available at all times t0 ≥ t.
Events are represented by marked points on these lines, and messages are
represented by diagonal lines between events. The resulting picture looks like
a collection of world lines as used in physics to illustrate the path taken
by various objects through spacetime.
Pictures like Figure 2.1 can be helpful for illustrating the various con-
straints we might put on message delivery. In Figure 2.1, the system is
completely asynchronous: messages can be delivered in any order, even if
sent between the same processes. If we run the same protocol under stronger
assumptions, we will get different communication patterns.
2
Formally, the delivery phase consists of n separate delivery events, in any order, that
between them clean out all the channels.
CHAPTER 2. MODEL 13

p3

p2

p1

Time →

Figure 2.1: Asynchronous message-passing execution. Time flows left-to-

right. Horizontal lines represent processes. Nodes represent events. Diagonal
edges between events represent messages. In this execution, p1 executes a
computation event that sends messages to p2 and p3 . When p2 receives this
message, it sends messages to p1 and p3 . Later, p2 executes a computation
event that sends a second message to p1 . Because the system is asynchronous,
there is no guarantee that messages arrive in the same order they are sent.

For example, Figure 2.2 shows an execution that is still asynchronous but
that assumes FIFO (first-in first-out) channels. A FIFO channel from some
process p to another process q guarantees that q receives messages in the
same order that p sends them (this can be simulated by a non-FIFO channel
by adding a sequence number to each message, and queuing messages at
the receiver until all previous messages have been processed).
If we go as far as to assume synchrony, we get the execution in Figure 2.3.
Now all messages take exactly one time unit to arrive, and computation
events follow each other in lockstep.

p3

p2

p1

Time →

Figure 2.2: Asynchronous message-passing execution with FIFO channels.

Multiple messages from one process to another are now guaranteed to be
delivered in the order they are sent.
CHAPTER 2. MODEL 14

p3

p2

p1

Time →

Figure 2.3: Synchronous message-passing execution. All messages are now

delivered in exactly one time unit, and computation events occur exactly
one time unit after the previous event.

2.5 Complexity measures

There is no explicit notion of time in the asynchronous model, but we can
define a time measure by adopting the rule that every message is delivered
and processed at most 1 time unit after it is sent. Formally, we assign time
0 to the first event, and assign the largest time we can to each subsequent
event, subject to the constraints that (a) no event is assigned a larger time
than any later event; (b) if a message m from i to j is created by an event at
time t, then the time for the delivery of m from i to j is no greater than t + 1,
and (c) any computation step is assigned a time no later than the previous
event at the same process (or 0 if the process has no previous events). This
is consistent with an assumption that message propagation takes at most 1
time unit and that local computation takes 0 time units.
Another way to look at this is that it is a definition of a time unit in terms
of maximum message delay together with an assumption that message delays
dominate the cost of the computation. This last assumption is pretty much
always true for real-world networks with any non-trivial physical separation
between components, thanks to speed of light limitations.
An example of an execution annotated with times in this way is given in
Figure 2.4.
The time complexity of a protocol (that terminates) is the time of the
last event at any process.
Note that looking at step complexity, the number of computation
events involving either a particular process (individual step complexity)
or all processes (total step complexity) is not useful in the asynchronous
model, because a process may be scheduled to carry out arbitrarily many
computation steps without any of its incoming or outgoing messages being
CHAPTER 2. MODEL 15

p3
1 1
p2
1 1
p1
0 2 2
Time →

Figure 2.4: Asynchronous message-passing execution with times.

delivered, which probably means that it won’t be making any progress. These
complexity measures will be more useful when we look at shared-memory
models (Part II).
For a protocol that terminates, the message complexity is the total
number of messages sent. We can also look at message length in bits, total
bits sent, etc., if these are useful for distinguishing our new improved protocol
from last year’s model.
For synchronous systems, time complexity becomes just the number of
rounds until a protocol finishes. Message complexity is still only loosely
connected to time complexity; for example, there are synchronous leader
election (Chapter 5) algorithms that, by virtue of grossly abusing the
synchrony assumption, have unbounded time complexity but very low message
complexity.
Chapter 3

Broadcast and convergecast

Here we’ll describe protocols for propagating information throughout a

network from some central initiator and gathering information back to that
same initiator. We do this both because the algorithms are actually useful
and because they illustrate some of the issues that come up with keeping
time complexity down in an asynchronous message-passing system.

3.1 Flooding
Flooding is about the simplest of all distributed algorithms. It’s dumb and
expensive, but easy to implement, and gives you both a broadcast mechanism
and a way to build rooted spanning trees.
We’ll give a fairly simple presentation of flooding roughly following
Chapter 2 of [AW04].

3.1.1 Basic algorithm

The basic flooding algorithm is shown in Algorithm 3.1. The idea is that
when a process receives a message M , it forwards it to all of its neighbors
unless it has seen it before, which it tracks using a single bit seen-message.
Theorem 3.1.1. Every process receives M after at most D time and at
most |E| messages, where D is the diameter of the network and E is the set
of (directed) edges in the network.
Proof. Message complexity: Each process only sends M to its neighbors
once, so each edge carries at most one copy of M .
Time complexity: By induction on d(root, v), we’ll show that each v
receives M for the first time no later than time d(root, v) ≤ D. The base

16
CHAPTER 3. BROADCAST AND CONVERGECAST 17

1 initially do
2 if pid = root then
3 seen-message ← true
4 send M to all neighbors
5 else
6 seen-message ← false

7 upon receiving M do
8 if seen-message = false then
9 seen-message ← true
10 send M to all neighbors

Algorithm 3.1: Basic flooding algorithm

case is when v = root, d(root, v) = 0; here root receives message at time

0. For the induction step, Let d(root, v) = k > 0. Then v has a neighbor
u such that d(root, u) = k − 1. By the induction hypothesis, u receives M
for the first time no later than time k − 1. From the code, u then sends
M to all of its neighbors, including v; M arrives at v no later than time
(k − 1) + 1 = k.

Note that the time complexity proof also demonstrates correctness: every
process receives M at least once.
As written, this is a one-shot algorithm: you can’t broadcast a second
message even if you wanted to. The obvious fix is for each process to
remember which messages it has seen and only forward the new ones (which
costs memory) and/or to add a time-to-live (TTL) field on each message
that drops by one each time it is forwarded (which may cost extra messages
and possibly prevents complete broadcast if the initial TTL is too small).
The latter method is what was used for searching in http://en.wikipedia.
org/wiki/Gnutella, an early peer-to-peer system. An interesting property
of Gnutella was that since the application of flooding was to search for huge
(multiple MiB) files using tiny ( 100 byte) query messages, the actual bit
complexity of the flooding algorithm was not especially large relative to the
bit complexity of sending any file that was found.
We can optimize the algorithm slightly by not sending M back to the
node it came from; this will slightly reduce the message complexity in many
cases but makes the proof a sentence or two longer. (It’s all a question of
what you want to optimize.)
CHAPTER 3. BROADCAST AND CONVERGECAST 18

3.1.2 Adding parent pointers

To build a spanning tree, modify Algorithm 3.1 by having each process
remember who it first received M from. The revised code is given as
Algorithm 3.2

1 initially do
2 if pid = root then
3 parent ← root
4 send M to all neighbors
5 else
6 parent ← ⊥

7 upon receiving M from p do

8 if parent = ⊥ then
9 parent ← p
10
11 send M to all neighbors

Algorithm 3.2: Flooding with parent pointers

We can easily prove that Algorithm 3.2 has the same termination proper-
ties as Algorithm 3.1 by observing that if we map parent to seen-message by
the rule ⊥ → false, anything else → true, then we have the same algorithm.
We would like one additional property, which is that when the algorithm
quiesces (has no outstanding messages), the set of parent pointers form a
rooted spanning tree. For this we use induction on time:
Lemma 3.1.2. At any time during the execution of Algorithm 3.2, the
following invariant holds:
1. If u.parent 6= ⊥, then u.parent.parent 6= ⊥ and following parent pointers
gives a path from u to root.
2. If there is a message M in transit from u to v, then u.parent 6= ⊥.
Proof. We have to show that that the invariant is true initially, and that any
event preserves the invariant. We’ll assume that all events are delivery events
for a single message, since we can have the algorithm treat a multi-message
delivery event as a sequence of single-message delivery events.
We’ll treat the initial configuration as the result of the root setting its
parent to itself and sending messages to all its neighbors. It’s not hard to
verify that the invariant holds in the resulting configuration.
CHAPTER 3. BROADCAST AND CONVERGECAST 19

For a delivery event, let v receive M from u. There are two cases: if
v.parent is already non-null, the only state change is that M is no longer in
transit, so we don’t care about u.parent any more. If v.parent is null, then

1. v.parent is set to u. This triggers the first case of the invariant. From
the induction hypothesis we have that u.parent = 6 ⊥ and that there
exists a path from u to the root. Then v.parent.parent = u.parent 6= ⊥
and the path from v → u → root gives the path from v.

2. Message M is sent to all of v’s neighbors. Because M is now in transit

from v, we need v.parent 6= ⊥; but we just set it to u, so we are happy.

At the end of the algorithm, the invariant shows that every process has a
path to the root, i.e., that the graph represented by the parent pointers is
connected. Since this graph has exactly |V | − 1 edges (if we don’t count the
self-loop at the root), it’s a tree.
Though we get a spanning tree at the end, we may not get a very good
spanning tree. For example, suppose our friend the adversary picks some
Hamiltonian path through the network and delivers messages along this
path very quickly while delaying all other messages for the full allowed 1
time unit. Then the resulting spanning tree will have depth |V | − 1, which
might be much worse than D. If we want the shallowest possible spanning
tree, we need to do something more sophisticated: see the discussion of
distributed breadth-first search in Chapter 4. However, we may be
happy with the tree we get from simple flooding: if the message delay on
each link is consistent, then it’s not hard to prove that we in fact get a
shortest-path tree. As a special case, flooding always produces a BFS tree in
the synchronous model.
Note also that while the algorithm works in a directed graph, the parent
pointers may not be very useful if links aren’t two-way.

3.1.3 Termination
See [AW04, Chapter 2] for further modifications that allow the processes to
detect termination. In a sense, each process can terminate as soon as it is
done sending M to all of its neighbors, but this still requires some mechanism
for clearing out the inbuf; by adding acknowledgments as described in [AW04],
we can terminate with the assurance that no further messages will be received.
CHAPTER 3. BROADCAST AND CONVERGECAST 20

3.2 Convergecast
A convergecast is the inverse of broadcast: instead of a message propagating
down from a single root to all nodes, data is collected from outlying nodes
to the root. Typically some function is applied to the incoming data at
each node to summarize it, with the goal being that eventually the root
obtains this function of all the data in the entire system. (Examples would
be counting all the nodes or taking an average of input values at all the
nodes.)
A basic convergecast algorithm is given in Algorithm 3.3; it propagates
information up through a previously-computed spanning tree.

1 initially do
2 if I am a leaf then
3 send input to parent

4 upon receiving M from c do

5 append (c, M ) to buffer
6 if buffer contains messages from all my children then
7 v ← f (buffer, input)
8 if pid = root then
9 return v
10 else
11 send v to parent

Algorithm 3.3: Convergecast

The details of what is being computed depend on the choice of f :

• If input = 1 for all nodes and f is sum, then we count the number of
nodes in the system.

• If input is arbitrary and f is sum, then we get a total of all the input
values.

• Combining the above lets us compute averages, by dividing the total

of all the inputs by the node count.

• If f just concatenates its arguments, the root ends up with a vector of

all the input values.
CHAPTER 3. BROADCAST AND CONVERGECAST 21

Running time is bounded by the depth of the tree: we can prove by

induction that any node at height h (height is length of the longest path from
this node to some leaf) sends a message by time h at the latest. Message
complexity is exactly n − 1, where n is the number of nodes; this is easily
shown by observing that each node except the root sends exactly one message.
Proving that convergecast returns the correct value is similarly done by
induction on depth: if each child of some node computes a correct value, then
that node will compute f applied to these values and its own input. What
the result of this computation is will, of course, depend on f ; it generally
makes the most sense when f represents some associative operation (as in
the examples above).

3.3 Flooding and convergecast together

A natural way to build the spanning tree used by convergecast is to run
flooding first. This also provides a mechanism for letting the leaves know
that they are leaves and initiating the protocol. The combined algorithm is
shown as Algorithm 3.4.
However, this may lead to very bad time complexity for the convergecast
stage. Consider a wheel-shaped network consisting of one central node p0
connected to nodes p1 , p2 , . . . , pn−1 , where each pi is also connected to pi+1 .
By carefully arranging for the pi pi+1 links to run much faster than the p0 pi
links, the adversary can make flooding build a tree that consists of a single
path p0 p1 p2 . . . pn−1 , even though the diameter of the network is only 2.
While it only takes 2 time units to build this tree (because every node is only
one hop away from the initiator), when we run convergecast we suddenly
find that the previously-speedy links are now running only at the guaranteed
≤ 1 time unit per hop rate, meaning that convergecast takes n − 1 time.
This may be less of an issue in real networks, where the latency of links
may be more uniform over time, meaning that a deep tree of fast links is
still likely to be fast when we reach the convergecast step. But in the worst
case we will need to be more clever about building the tree. We show how
to do this in Chapter 4.
CHAPTER 3. BROADCAST AND CONVERGECAST 22

1 initially do
2 children ← ∅
3 nonChildren ← ∅
4 if pid = root then
5 parent ← root
6 send init to all neighbors
7 else
8 parent ← ⊥

9 upon receiving init from p do

10 if parent = ⊥ then
11 parent ← p
12 send init to all neighbors
13 else
14 send nack to p

15 upon receiving nack from p do

16 nonChildren ← nonChildren ∪ {p}
17 as soon as children ∪ nonChildren includes all my neighbors do
18 v ← f (buffer, input)
19 if pid = root then
20 return v
21 else
22 send ack(v) to parent

23 upon receiving ack(v) from k do

24 add (k, v) to buffer
25 add k to children
Algorithm 3.4: Flooding and convergecast combined
Chapter 4

Distributed breadth-first
search

Here we describe some algorithms for building a breadth-first search

(BFS) tree in a network. All assume that there is a designated initiator
node that starts the algorithm. At the end of the execution, each node except
the initiator has a parent pointer and every node has a list of children. These
are consistent and define a BFS tree: nodes at distance k from the initiator
appear at level k of the tree.
In a synchronous network, flooding (§3.1) solves BFS; see [AW04,
Lemma 2.8, page 21] or [Lyn96, §4.2]. So the interesting case is when
the network is asynchronous.
In an asynchronous network, the complication is that we can no longer
rely on synchronous communication to reach all nodes at distance d at the
same time. So instead we need to keep track of distances explicitly, or
possibly enforce some approximation to synchrony in the algorithm. (A
general version of this last approach is to apply a synchronizer to one of the
synchronous algorithms using a synchronizer; see Chapter 7.)
To keep things simple, we’ll drop the requirement that a parent learn
the IDs of its children, since this can be tacked on as a separate notification
protocol, in which each child just sends one message to its parent once it
figures out who its parent is.

4.1 Using explicit distances

This is a translation of the AsynchBFS automaton from [Lyn96, §15.4]. It’s
a very simple algorithm, closely related to Dijkstra’s algorithm for shortest

23
CHAPTER 4. DISTRIBUTED BREADTH-FIRST SEARCH 24

paths, but there is otherwise no particular reason to use it. Not only does
it not detect termination, but it is also dominated by the O(D) time and
O(DE) message complexity synchronizer-based algorithm described in §4.3.
(Here D is the diameter of the network, the maximum distance between
any two nodes.)
The idea is to run flooding with distances attached. Each node sets its
distance to 1 plus the smallest distance sent by its neighbors and its parent
to the neighbor supplying that smallest distance. A node notifies all its
neighbors of its new distance whenever its distance changes.
Pseudocode is given in Algorithm 4.1

1 initially do
2 if pid = initiator then
3 distance ← 0
4 send distance to all neighbors
5 else
6 distance ← ∞

7 upon receiving d from p do

8 if d + 1 < distance then
9 distance ← d + 1
10 parent ← p
11 send distance to all neighbors

Algorithm 4.1: AsynchBFS algorithm (from [Lyn96])

(See [Lyn96] for a precondition-effect description, which also includes

code for buffering outgoing messages.)
The claim is that after at most O(V E) messages and O(D) time, all
distance values are equal to the length of the shortest path from the initiator
to the appropriate node. The proof is by showing the following:

Lemma 4.1.1. The variable distancep is always the length of some path
from initiator to p, and any message sent by p is also the length of some
path from initiator to p.

Proof. The second part follows from the first; any message sent equals p’s
current value of distance. For the first part, suppose p updates its distance;
then it sets it to one more than the length of some path from initiator to p0 ,
which is the length of that same path extended by adding the pp0 edge.
CHAPTER 4. DISTRIBUTED BREADTH-FIRST SEARCH 25

We also need a liveness argument that says that distancep = d(initiator, p)

no later than time d(initiator, p). Note that we can’t detect when distance
stabilizes to the correct value without a lot of additional work.
In [Lyn96], there’s an extra |V | term in the time complexity that comes
from message pile-ups, since the model used there only allows one incoming
message to be processed per time units (the model in [AW04] doesn’t have
this restriction). The trick to arranging this to happen often is to build a
graph where node 1 is connected to nodes 2 and 3, node 2 to 3 and 4, node
3 to 4 and 5, etc. This allows us to quickly generate many paths of distinct
lengths from node 1 to node k, which produces k outgoing messages from
node k. It may be that a more clever analysis can avoid this blowup, by
showing that it only happens in a few places.

4.2 Using layering

This approach is used in the LayeredBFS algorithm in [Lyn96], which is due
to Gallager [Gal82].
Here we run a sequence of up to |V | instances of the simple algorithm
with a distance bound on each: instead of sending out just 0, the initiator
sends out (0, bound), where bound is initially 1 and increases at each phase.
A process only sends out its improved distance if it is less than bound.
Each phase of the algorithm constructs a partial BFS tree that contains
only those nodes within distance bound of the root. This tree is used to
report back to the root when the phase is complete. For the following phase,
notification of the increase in bound increase is distributed only through
the partial BFS tree constructed so far. With some effort, it is possible to
prove that in a bidirectional network that this approach guarantees that
each edge is only probed once with a new distance (since distance-1 nodes
are recruited before distance-2 nodes and so on), and the bound-update and
acknowledgment messages contribute at most |V | messages per phase. So we
get O(E + V D) total messages. But the time complexity is bad: O(D2 ) in
the worst case.

4.3 Using local synchronization

The reason the layering algorithm takes so long is that at each phase we
have to phone all the way back up the tree to the initiator to get permission
to go on to the next phase. We need to do this to make sure that a node
is only recruited into the tree once: otherwise we can get pile-ups on the
CHAPTER 4. DISTRIBUTED BREADTH-FIRST SEARCH 26

channels as in the simple algorithm. But we don’t necessarily need to do this

globally. Instead, we’ll require each node at distance d to delay sending out
a recruiting message until it has confirmed that none of its neighbors will be
sending it a smaller distance. We do this by having two classes of messages:1

• exactly(d): “I know that my distance is d.”

• more-than(d): “I know that my distance is > d.”

The rules for sending these messages for a non-initiator are:

1. I can send exactly(d) as soon as I have received exactly(d − 1) from at

least one neighbor and more-than(d − 2) from all neighbors.

2. I can send more-than(d) if d = 0 or as soon as I have received more-than(d−

1) from all neighbors.

The initiator sends exactly(0) to all neighbors at the start of the protocol
(these are the only messages the initiator sends).
My distance will be the unique distance that I am allowed to send in an
exactly(d) messages. Note that this algorithm terminates in the sense that
every node learns its distance at some finite time.
If you read the discussion of synchronizers in Chapter 7, this algorithm
essentially corresponds to building the alpha synchronizer into the syn-
chronous BFS algorithm, just as the layered model builds in the beta
synchronizer. See [AW04, §11.3.2] for a discussion of BFS using synchro-
nizers. The original approach of applying synchronizers to get BFS is due to
Awerbuch [Awe85].
We now show correctness. Under the assumption that local computation
takes zero time and message delivery takes at most 1 time unit, we’ll show
that if d(initiator, p) = d, (a) p sends more-than(d0 ) for any d0 < d by time
d0 , (b) p sends exactly(d) by time d, (c) p never sends more-than(d0 ) for any
d0 ≥ d, and (d) p never sends exactly(d0 ) for any d0 6= d. For parts (c) and
(d) we use induction on d0 ; for (a) and (b), induction on time. This is not
terribly surprising: (c) and (d) are safety properties, so we don’t need to
talk about time. But (a) and (b) are liveness properties so time comes in.
Let’s start with (c) and (d). The base case is that the initiator never
sends any more-than messages at all, and so never sends more-than(0), and
1
In an earlier version of these notes, these messages where called distance(d) and
not-distance(d); the more self-explanatory exactly and more-than terminology is taken from
[BDLP08].
CHAPTER 4. DISTRIBUTED BREADTH-FIRST SEARCH 27

any non-initiator never sends exactly(0). For larger d0 , observe that if a

non-initiator p sends more-than(d0 ) for d0 ≥ d, it must first have received
more-than(d0 − 1) from all neighbors, including some neighbor p0 at distance
d−1. But the induction hypothesis tells us that p0 can’t send more-than(d0 −1)
for d0 − 1 ≥ d − 1. Similarly, to send exactly(d0 ) for d0 < d, p must first
have received exactly(d0 − 1) from some neighbor p0 , but again p0 must be at
distance at least d − 1 from the initiator and so can’t send this message either.
In the other direction, to send exactly(d0 ) for d0 > d, p must first receive
more-than(d0 − 2) from this closer neighbor p0 , but then d0 − 2 > d − 2 ≥ d − 1
so more-than(d0 − 2) is not sent by p0 .
Now for (a) and (b). The base case is that the initiator sends exactly(0)
to all nodes at time 0, giving (a), and there is no more-than(d0 ) with d0 < 0
for it to send, giving (b) vacuously; and any non-initiator sends more-than(0)
immediately. At time t + 1, we have that (a) more-than(t) was sent by any
node at distance t + 1 or greater by time t and (b) exactly(t) was sent by
any node at distance t by time t; so for any node at distance t + 2 we
send more-than(t + 1) no later than time t + 1 (because we already received
more-than(t) from all our neighbors) and for any node at distance t + 1 we
send exactly(t + 1) no later than time t + 1 (because we received all the
preconditions for doing so by this time).
Message complexity: A node at distance d sends more-than(d0 ) for all
0 < d0 < d and exactly(d) and no other messages. So we have message
complexity bounded by |E| · D in the worst case. Note that this is gives a
bound of O(DE), which is slightly worse than the O(E + DV ) bound for
the layered algorithm.
Time complexity: It’s immediate from (a) and (b) that all messages that
are sent are sent by time D, and indeed that any node p learns its distance
at time d(initiator, p). So we have optimal time complexity, at the cost of
higher message complexity. I don’t know if this trade-off is necessary, or if a
more sophisticated algorithm could optimize both.
Our time proof assumes that messages don’t pile up on edges, or that
such pile-ups don’t affect delivery time (this is the default assumption used
in [AW04]). A more sophisticated proof could remove this assumption.
One downside of this algorithm is that it has to be started simultaneously
at all nodes. Alternatively, we could trigger “time 0” at each node by
a broadcast from the initiator, using the usual asynchronous broadcast
algorithm; this would give us a BFS tree in O(|E| · D) messages (since the
O(|E|) messages of the broadcast disappear into the constant) and 2D time.
The analysis of time goes through as before, except that the starting time 0
becomes the time at which the last node in the system is woken up by the
CHAPTER 4. DISTRIBUTED BREADTH-FIRST SEARCH 28

broadcast. Further optimizations are possible; see, for example, the paper of
Boulinier et al. [BDLP08], which shows how to run the same algorithm with
constant-size messages.
Chapter 5

Leader election

(See also [AW04, Chapter 3] or [Lyn96, Chapter 3].)

The basic idea of leader election is that we want a single process to
declare itself leader and the others to declare themselves non-leaders. The
non-leaders may or may not learn the identity of the leader as part of the
protocol; if not, we can always add an extra phase where the leader broadcasts
its identity to the others. Traditionally, leader election has been used as a
way to study the effects of symmetry, and many leader election algorithms
are designed for networks in the form of a ring.
A classic result of Angluin [Ang80] shows that leader election in a ring is
impossible if the processes do not start with distinct identities. The proof
is that if everybody is in the same state at every step, they all put on the
crown at the same time. We discuss this result in §5.1.
With ordered identities, a simple algorithm due to Le Lann [LL77] and
Chang and Roberts [CR79] solves the problem in O(n) time with O(n2 )
messages: I send out my own id clockwise and forward any id bigger than
mine. If I get my id back, I win. This works with a unidirectional ring,
doesn’t require synchrony, and never produces multiple leaders. See §5.2.1.
On a bidirectional ring we can get O(n log n) messages and O(n) time
with power-of-2 probing, using an algorithm of Hirschberg and Sinclair [HS80].
See §5.2.2.
An evil trick: if we have synchronized starting, known n, and known id
space, we can have process with id i wait until round i · n to start sending
its id around, and have everybody else drop out when they receive it; this
way only one process (the one with smallest id) ever starts a message and
only n messages are sent [FL87]. But the running time can be pretty bad.
For general networks, we can apply the same basic strategy as in Le Lann-

29
CHAPTER 5. LEADER ELECTION 30

Chang-Roberts by having each process initiate a broadcast/convergecast

algorithm that succeeds only if the initiator has the smallest id. See §5.3.
Some additional algorithms for the asynchronous ring are given in §§5.2.3
and 5.2.4. Lower bounds are shown in §5.4.

5.1 Symmetry
A system exhibits symmetry if we can permute the nodes without changing
the behavior of the system. More formally, we can define a symmetry
as an equivalence relation on processes, where we have the additional
properties that all processes in the same equivalence class run the same code;
and whenever p is equivalent to p0 , each neighbor q of p is equivalent to a
corresponding neighbor q 0 of p0 .
An example of a network with a lot of symmetries would be an anony-
mous ring, which is a network in the form of a cycle (the ring part) in
which every process runs the same code (the anonymous part). In this case
all nodes are equivalent. If we have a line, then we might or might not have
any non-trivial symmetries: if each node has a sense of direction that tells
it which neighbor is to the left and which is to the right, then we can identify
each node uniquely by its distance from the left edge. But if the nodes don’t
have a sense of direction, we can flip the line over and pair up nodes that
map to each other.1
Symmetries are convenient for proving impossibility results, as observed by
Angluin [Ang80]. The underlying theme is that without some mechanism for
symmetry breaking, a message-passing system escape from a symmetric
initial configuration. The following lemma holds for deterministic systems,
basically those in which processes can’t flip coins:
Lemma 5.1.1. A symmetric deterministic message-passing system that
starts in an initial configuration in which equivalent processes have the same
state has a synchronous execution in which equivalent processes continue to
have the same state.
Proof. Easy induction on rounds: if in some round p and p0 are equivalent
and have the same state, and all their neighbors are equivalent and have the
same state, then p and p0 receive the same messages from their neighbors
and can proceed to the same state (including outgoing messages) in the next
round.
1
Typically, this does not mean that the nodes can’t tell their neighbors apart. But it
does mean that if we swap the labels for all the neighbors (corresponding to flipping the
entire line from left to right), we get the same executions.
CHAPTER 5. LEADER ELECTION 31

An immediate corollary is that you can’t do leader election in an anony-

mous system with a symmetry that puts each node in a non-trivial equivalence
class, because as soon as I stick my hand up to declare I’m the leader, so do
all my equivalence-class buddies.
With randomization, Lemma 5.1.1 doesn’t directly apply, since we can
break symmetry by having my coin-flips come up differently from yours. It
does show that we can’t guarantee convergence to a single leader in any fixed
amount of time (because otherwise we could just fix all the coin flips to get
a deterministic algorithm). Depending on what the processes know about
the size of the system, it may still be possible to show that a randomized
algorithm necessarily fails in some cases.2
A more direct way to break symmetry is to assume that all processes
have identities; now processes can break symmetry by just declaring that
the one with the smaller or larger identity wins. This approach is taken in
the algorithms in the following sections.

5.2 Leader election in rings

Here we’ll describe some basic leader election algorithms for rings. Histor-
ically, rings were the first networks in which leader election was studied,
because they are the simplest networks whose symmetry makes the problem
difficult, and because of the connection to token-ring networks, a method for
congestion control in local-area networks that is no longer used much.

5.2.1 The Le Lann-Chang-Roberts algorithm

This is about the simplest leader election algorithm there is. It works in
a unidirectional ring, where messages can only travel clockwise.3 The
algorithms works does not require synchrony, but we’ll assume synchrony to
make it easier to follow.
Formally, we’ll let the state space for each process i consist of two variables:
leader, initially 0, which is set to 1 if i decides it’s a leader; and maxId, the
largest id seen so far. We assume that i denotes i’s position rather than its
2
Specifically, if the processes don’t know the size of the ring, we can imagine a ring
of size 2n in which the first n processes happen to get exactly the same coin-flips as the
second n processes for long enough that two matching processes, one in each region, both
think they have won the fight in a ring of size n and declare themself to be the leader.
3
We’ll see later in §5.2.3 that the distinction between unidirectional rings and bidirec-
tional rings is not a big deal, but for now let’s imagine that having a unidirectional ring is
a serious hardship.
CHAPTER 5. LEADER ELECTION 32

id, which we’ll write as idi . We will also treat all positions as values mod n,
to simplify the arithmetic.
The initial version of this algorithm was proposed by Le Lann [LL77]; it
involved sending every id all the way around the ring, and having a node
decide it was a leader if it had the largest id. Chang and Roberts [CR79]
improved on this by having nodes refuse to forward any id smaller than the
maximum id seen so far. This means that only the largest id makes it all the
way around the ring, so a node can declare itself leader the moment it sees
its own id. The resulting algorithm is known as either Chang-Roberts or Le
Lann-Chang-Roberts (LCR). We’ll go with the latter because it is always
polite to be generous with credit.
Code for the LCR algorithm is given in Algorithm 5.1.

1 initially do
2 leader ← 0
3 maxId ← idi
4 send idi to clockwise neighbor
5 upon receiving j do
6 if j = idi then
7 leader ← 1
8 if j > maxId then
9 maxId ← j
10 send j to clockwise neighbor

Algorithm 5.1: LCR leader election

5.2.1.1 Proof of correctness for synchronous executions

By induction on the round number k. The induction hypothesis is that in
round k, each process i’s leader bit is 0, its maxId value is equal to the largest
id in the range (i − k) . . . i, and that it sends idi−k if and only if idi−k is
the largest id in the range (i − k) . . . i. The base case is that when k = 0,
maxId = idi is the largest id in i . . . i, and i sends idi . For the induction step,
observe that in round k − 1, i − 1 sends id(i−1)−(k−1) = idi−k if and only if it
is the largest in the range (i − k) . . . (i − 1), and that i adopts it as the new
value of maxId and sends it just in case it is larger than the previous largest
value in (i − k + 1) . . . (i − 1), i.e., if it is the largest value in (i − k) . . . i.
Finally, in round n − 1, i − 1 sends idi−N = idi if and only if i is the
CHAPTER 5. LEADER ELECTION 33

largest id in (i − n + 1) . . . i, the whole state space. So i receives idi and sets

leaderi = 1 if and only if it has the maximum id.
With some tinkering, this proof can be adapted to show that the protocol
also works in asynchronous executions.

5.2.1.2 Performance
It’s immediate from the correctness proof that the protocols terminates after
exactly n rounds.
To count message traffic, observe that each process sends at most 1
message per round, for a total of O(n2 ) messages. This is a tight bound
since if the ids are in decreasing order n, n − 1, n − 2, . . . 1, then no messages
get eaten until they hit n.

5.2.2 The Hirschberg-Sinclair algorithm

This algorithm improves on Le Lann-Chang-Roberts by reducing the message
complexity. The idea is that instead of having each process send a message all
the way around a ring, each process will first probe locally to see if it has the
largest id within a short distance. If it wins among its immediate neighbors,
it doubles the size of the neighborhood it checks, and continues as long as
it has a winning id. This means that most nodes drop out quickly, giving
a total message complexity of O(n log n). The running time is a constant
factor worse than LCR, but still O(n).
To specify the protocol, it may help to think of messages as mobile agents
and the state of each process as being of the form (local-state, {agents I’m carrying}).
Then the sending rule for a process becomes ship any agents in whatever
direction they want to go and the transition rule is accept any incoming
agents and update their state in terms of their own internal transition rules.
An agent state for LCR will be something like (original-sender, direction,
hop-count, max-seen) where direction is R or L depending on which way the
agent is going, hop-count is initially 2k when the agent is sent and drops
by 1 each time the agent moves, and max-seen is the biggest id of any node
the agent has visited. An agent turns around (switches direction) when
hop-count reaches 0.
To prove this works, we can mostly ignore the early phases (though we
have to show that the max-id node doesn’t drop out early, which is not too
hard). The last phase involves any surviving node probing all the way around
the ring, so it will declare itself leader only when it receives its own agent
from the left. That exactly one node does so is immediate from the same
CHAPTER 5. LEADER ELECTION 34

argument for LCR.

Complexity analysis is mildly painful but basically comes down to the
fact that any node that sends a message 2k hops had to be a winner at phase
2k−1 , which means that it is the largest of some group of 2k−1 ids. Thus
the 2k -hop senders are spaced at least 2k−1 away from each other and there
are at most n/2k−1 of them. Summing up over all dlg ne phases, we get
Pdlg ne k k−1 = O(n log n) messages and Pdlg ne 2k = O(n) time.
k=0 2 n/2 k=0

5.2.3 Peterson’s algorithm for the unidirectional ring

This algorithm is due to Peterson [Pet82] and assumes an asynchronous,
unidirectional ring. It gets O(n log n) message complexity in all executions.
The basic idea (2-way communication version): Start with n candidate
leaders. In each of at most lg n asynchronous phases, each candidate probes
its nearest neighbors to the left and right; if its ID is larger than the IDs of
both neighbors, it survives to the next phase. Non-candidates act as relays
passing messages between candidates. As in Hirschberg and Sinclair (§5.2.2),
the probing operations in each phase take O(n) messages, and at least half
of the candidates drop out in each phase. The last surviving candidate wins
when it finds that it’s its own neighbor.
To make this work in a 1-way ring, we have to simulate 2-way communi-
cation by moving the candidates clockwise around the ring to catch up with
their unsendable counterclockwise messages. Peterson’s algorithm does this
with a two-hop approach that is inspired by the 2-way case above; in each
phase k, a candidate effectively moves two positions to the right, allowing it
to look at the ids of three phase-k candidates before deciding to continue
in phase k + 1 or not. Here is a very high-level description; it assumes that
we can buffer and ignore incoming messages from the later phases until we
get to the right phase, and that we can execute sends immediately upon
receiving messages. Doing this formally in terms of the model of §2.1 means
that we have to build explicit internal buffers into our processes, which we
can easily do but won’t do here (see [Lyn96, pp. 483–484] for the right way
to do this).
We can use a similar trick to transform any bidirectional-ring algorithm
into a unidirectional-ring algorithm: alternate between phases where we send
a message right, then send a virtual process right to pick up any left-going
messages deposited for us. The problem with this trick is that it requires two
messages per process per phase, which gives us a total message complexity of
O(n2 ) if we start with an O(n)-time algorithm. Peterson’s algorithm avoids
this by propagating only the surviving candidates.
CHAPTER 5. LEADER ELECTION 35

Pseudocode for Peterson’s algorithm is given in Algorithm 5.2.

1 procedure candidate()
2 phase ← 0
3 current ← pid
4 while true do
5 send probe(phase, current)
6 wait for probe(phase, x)
7 id2 ← x
8 send probe(phase + 1/2, id2 )
9 wait for probe(phase + 1/2, x)
10 id3 ← x
11 if id2 = current then
12 I am the leader!
13 return
14 else if id2 > current and id2 > id3 do
15 current ← id2
16 phase ← phase + 1
17 else
18 switch to relay()

19 procedure relay()
20 upon receiving probe(p, i) do
21 send probe(p, i)

Algorithm 5.2: Peterson’s leader-election algorithm

Note: The phase arguments in the probe messages are useless if one has
FIFO channels, which is why [Lyn96] doesn’t use them.
Proof of correctness is essentially the same as for the 2-way algorithm.
For any pair of adjacent candidates, at most one of their current IDs survives
to the next phase. So we get a sole survivor after lg n phases. Each process
sends or relays at most 2 messages per phases, so we get at most 2n lg n total
messages.

5.2.4 A simple randomized O(n log n)-message algorithm

An alternative to running a more sophisticated algorithm is to reduce the
average cost of LCR using randomization. The presentation here follows the
average-case analysis done by Chang and Roberts [CR79].
CHAPTER 5. LEADER ELECTION 36

Run LCR where each id is constructed by prepending a long random

bit-string to the real id. This gives uniqueness (since the real id’s act as
tie-breakers) and something very close to a random permutation on the
constructed id’s. When we have unique random id’s, a simple argument
shows that the i-th largest id only propagates an expected n/i hops, giving
a total of O(nHn ) = O(n log n) hops.4 Unique random id’s occur with high
probability provided the range of the random sequence is  n2 .
The downside of this algorithm compared to Peterson’s is that knowledge
of n is required to pick random id’s from a large enough range. It also has
higher bit complexity, since Peterson’s algorithm is sending only IDs (in the
FIFO-channel version) without any random padding. An possible upside is
that if the range of random ids is large enough, we can run it without any
initial ids at all, as long as we are willing to accept a small probability of
accidentally electing two leaders.

5.3 Leader election in general networks

For general networks, a simple approach is to have each node initiate a
breadth-first-search and convergecast, with nodes refusing to participate in
the protocol for any initiator with a lower id. It follows that only the node
with the maximum id can finish its protocol; this node becomes the leader.
If messages from parallel broadcasts are combined, it’s possible to keep the
message complexity of this algorithm down to O(DE).
More sophisticated algorithms reduce the message complexity by coalesc-
ing local neighborhoods similar to what happens in the Hirschberg-Sinclair
and Peterson algorithms. A noteworthy example is an O(n log n) message-
complexity algorithm of Afek and Gafni [AG91], who also show an Ω(n log n)
lower bound on message complexity for any synchronous algorithm in a
complete network.

5.4 Lower bounds

Here we present two classic Ω(log n) lower bounds on message complexity
for leader election in the ring. The first, due to Burns [Bur80], assumes
that the system is asynchronous and that the algorithm is uniform: it
does not depend on the size of the ring. The second, due to Frederickson
4
Alternatively, we could consider the average-case complexity of the algorithm when
we assume all n! orderings of the ids are equally likely; this also gives O(n log n) expected
message complexity [CR79].
CHAPTER 5. LEADER ELECTION 37

and Lynch [FL87], allows a synchronous system and relaxes the uniformity
assumption, but requires that the algorithm can’t do anything to ids but
copy and compare them.

5.4.1 Lower bound on asynchronous message complexity

Here we describe a lower bound for uniform asynchronous leader election in
the ring. The description here is based on [AW04, §3.3.3]; a slightly different
presentation can also be found in [Lyn96, §15.1.4]. The original result is due
to Burns [Bur80]. We assume the system is deterministic.
The basic idea is to construct a bad execution in which n processes send
lots of messages recursively, by first constructing two bad (n/2)-process
executions and pasting them together in a way that generates many extra
messages. If the pasting step produces Θ(n) additional messages, we get a
recurrence T (n) ≥ 2T (n/2) + Θ(n) for the total message traffic, which has
solution T (n) = Ω(n log n).
We’ll assume that all processes are trying to learn the identity of the
process with the smallest id. This is a slightly stronger problem that mere
leader election, but it can be solved with at most an additional 2n messages
once we actually elect a leader. So if we get a lower bound of f (n) messages
on this problem, we immediately get a lower bound of f (n) − 2n on leader
election.
To construct the bad execution, we consider “open executions” on rings
of size n where no message is delivered across some edge (these will be partial
executions, because otherwise the guarantee of eventual delivery kicks in).
Because no message is delivered across this edge, the processes can’t tell if
there is really a single edge there or some enormous unexplored fragment of
a much larger ring. Our induction hypothesis will show that a line of n/2
processes can be made to send at least T (n/2) messages in an open execution
(before seeing any messages across the open edge); we’ll then show that a
linear number of additional messages can be generated by pasting two such
executions together end-to-end, while still getting an open execution with n
processes.
In the base case, we let n = 1. Somebody has to send a message eventually,
giving T (2) ≥ 1.
For larger n, suppose that we have two open executions on n/2 processes
that each send at least T (n/2) messages. Break the open edges in both
executions and paste the resulting lines together to get a ring of size n;
similarly paste the schedules σ1 and σ2 of the two executions together to
get a combined schedule σ1 σ2 with at least 2T (n/2) messages. Note that in
CHAPTER 5. LEADER ELECTION 38

the combined schedule no messages are passed between the two sides, so the
processes continue to behave as they did in their separate executions.
Let e and e0 be the edges we used to past together the two rings. Extend
σ1 σ2 by the longest possible suffix σ3 in which no messages are delivered
across e and e0 . Since σ3 is as long as possible, after σ1 σ2 σ3 , there are no
messages waiting to be delivered across any edge except e and e0 and all
processes are quiescent—they will send no additional messages until they
receive one.
We now consider some suffix σ4 such causes the protocol to finish. While
executing σ4 , construct two sets of processes S and S 0 by the following rules:
1. If a process is not yet in S or S 0 and receives a message delivered across
e, put it in S; similarly if it receives a message delivered across e0 , put
it in S 0 .
2. If a process is not yet in S or S 0 and receives a message that was sent
by a process in S, put it in S; similarly for S 0 .
triggered by a delivery across e
Observe that S ∪ S 0 includes every process on the half of the ring with
the larger minimum id, because any such process that doesn’t receive a
message in σ4 doesn’t learn the global minimum. So |S ∪ S 0 | ≥ n/2 and thus
min(|S|, |S 0 |) ≥ n/4.
Assume without loss of generality that it is |S| that is at least n/4.
Except for the two processes incident to e, every process that is added to S
is added in response to a message sent in σ4 . So there are at least n/4 − 2
such messages. We can also argue that all of these messages are sent in
the subschedule τ of σ4 that contains only messages that do not depend on
messages delivered across e0 . It follows that σ1 σ2 σ3 τ is an open execution
on n processes with at least 2T (n/2) + n/4 − 2 sent messages. This gives
T (n) ≥ 2T (n/2) + n/4 − 2 = 2T (n/2) + Ω(n) as claimed.

5.4.2 Lower bound for comparison-based algorithms

Here we give an Ω(n log n) lower bound on messages for synchronous-start
comparison-based algorithms in bidirectional synchronous rings. For full
details see [Lyn96, §3.6], [AW04, §3.4.2], or the original JACM paper by
Frederickson and Lynch [FL87].
Basic ideas:

• Two fragments i . . . i + k and j . . . j + k of a ring are order-equivalent

provided idi+a > idi+b if and only if idj+a > idj+b for b = 0 . . . k.
CHAPTER 5. LEADER ELECTION 39

• An algorithm is comparison-based if it can’t do anything to IDs but

copy them and test for <. The state of such an algorithm is modeled
by some non-ID state together with a big bag of IDs, messages have a
pile of IDs attached to them, etc. Two states/messages are equivalent
under some mapping of IDs if you can translate the first to the second
by running all IDs through the mapping.
An equivalent version uses an explicit equivalence relation between
processes. Let executions of p1 and p2 be similar if both processes
send messages in the same direction(s) in the same rounds and both
processes declare themselves leader (or not) at the same round. Then
an algorithm is comparison-based based if order-equivalent rings yield
similar executions for corresponding processes. This can be turned
into the explicit-copying-IDs model by replacing the original protocol
with a full-information protocol in which each message is replaced
by the ID and a complete history of the sending process (including all
messages it has every received).

• Define an active round as a round in which at least one message

is sent. Claim: Actions of i after k active rounds depends, up to an
order-equivalent mapping of ids, only on the order-equivalence class of
ids in i − k . . . i + k, the k-neighborhood of i. Proof: by induction on
k. Suppose i and j have order-equivalent (k − 1)-neighborhoods; then
after k − 1 active rounds they have equivalent states by the induction
hypothesis. In inactive rounds, i and j both receive no messages and
update their states in the same way. In active rounds, i and j receive
order-equivalent messages and update their states in an order-equivalent
way.

• If we have an order of ids with a lot of order-equivalent k-neighborhoods,

then after k active rounds if one process sends a message, so do a lot
of other ones.

Now we just need to build a ring with a lot of order-equivalent neighbor-

hoods. For n a power of 2 we can use the bit-reversal ring, e.g., id sequence
000, 100, 010, 110, 001, 101, 011, 111 (in binary) when n = 8. Figure 5.1 gives
a picture of what this looks like for n = 32.
For n not a power of 2 we look up Frederickson and Lynch [FL87] or
Attiya et al. [ASW88]. In either case we get Ω(n/k) order-equivalent members
of each equivalence class after k active rounds, giving Ω(n/k) messages per
active round, which sums to Ω(n log n).
CHAPTER 5. LEADER ELECTION 40

35

30

25

20

15

10

0
-5 0 5 10 15 20 25 30 35

Figure 5.1: Labels in the bit-reversal ring with n = 32

For non-comparison-based algorithms we can still prove Ω(n log n) mes-

sages for time-bounded algorithms, but it requires techniques from Ram-
sey theory, the branch of combinatorics that studies when large enough
structures inevitably contain substructures with certain properties.5 Here
“time-bounded” means that the running time can’t depend on the size of the
ID space. See [AW04, §3.4.2] or [Lyn96, §3.7] for the textbook version, or
[FL87, §7] for the original result.
The intuition is that for any fixed protocol, if the ID space is large
enough, then there exists a subset of the ID space where the protocol
acts like a comparison-based protocol. So the existence of an O(f (n))-
message time-bounded protocol implies the existence of an O(f (n))-message
comparison-based protocol, and from the previous lower bound we know
f (n) is Ω(n log n). Note that time-boundedness is necessary: we can’t prove
the lower bound for non-time-bounded algorithms because of the i · n trick.

5
The classic example is Ramsey’s Theorem, which says that if you color the edges of
a complete graph red or blue, while trying to avoid having any subsets of k vertices with
all edges between them the same color, you will no longer be able to once the graph is large
enough (for any fixed k). See [GRS90] for much more on the subject of Ramsey theory.
Chapter 6

Logical clocks

Logical clocks assign a timestamp to all events in an asynchronous message-

passing system that simulates real time, thereby allowing timing-based algo-
rithms to run despite asynchrony. In general, they don’t have anything to
do with clock synchronization or wall-clock time; instead, they provide nu-
merical values that increase over time and are consistent with the observable
behavior of the system. This means that local events on a single process
have increasing times, and messages are never delivered before they are sent,
when time is measured using the logical clock.

6.1 Causal ordering

The underlying notion of a logical clock is causal ordering, a partial order
on events that describes when one event e provably occurs before some other
event e0 .
For the purpose of defining casual ordering and logical clocks, we will
assume that a schedule consists of send events and receive events, which
correspond to some process sending a single message or receiving a single
message, respectively.
Given two schedules S and S 0 , call S and S 0 similar if S|p = S 0 |p for all
processes p; in other words, S and S 0 are similar if they are indistinguishable
by all participants. We can define a causal ordering on the events of some
schedule S implicitly by considering all schedules S 0 similar to S, and declare
that e < e0 if e precedes e0 in all such S. But it is usually more useful to
make this ordering explicit.
Following [AW04, §6.1.1] (and ultimately [Lam78]), define the happens-
before relation ⇒ on a schedule S to consist of:
S

41
CHAPTER 6. LOGICAL CLOCKS 42

1. All pairs (e, e0 ) where e precedes e0 in S and e and e0 are events of the
same process.

2. All pairs (e, e0 ) where e is a send event and e0 is the receive event for
the same message.

3. All pairs (e, e0 ) where there exists a third event e00 such that e ⇒ e00
S
and e00 ⇒ e0 . (In other words, we take the transitive closure of the
S
relation defined by the previous two cases.)

It is not terribly hard to show that this gives a partial order; the main
observation is that if e ⇒ e0 , then e precedes e0 in S. So ⇒ is a subset of the
S S
total order <S given by the order of events in S.
A causal shuffle S 0 of a schedule S is a permutation of S that is
consistent with the happens-before relation on S; that is, if e happens-before
e0 in S, then e precedes e0 in S 0 . The importance of the happens-before
relation follows from the following lemma, which says that the causal shuffles
of S are precisely the schedules S 0 that are similar to S.

Lemma 6.1.1. Let S 0 be a permutation of the events in S. Then the

following two statements are equivalent:

1. S 0 is a causal shuffle of S.

2. S 0 is the schedule of an execution fragment of a message-passing system

with S|p = S 0 |p for all S 0 .

Proof. (1 ⇒ 2). We need to show both similarity and that S 0 corresponds to

some execution fragment. We’ll show similarity first. Pick some p; then every
event at p in S also occurs in S 0 , and they must occur in the same order by
the first case of the definition of the happens-before relation. This gets us
halfway to showing S 0 is the schedule of some execution fragment, since it
says that any events initiated by p are consistent with p’s programming. To
get the rest of the way, observe that any other events are receive events. For
each receive event e0 in S, there must be some matching send event e also in
S; thus e and e0 are both in S 0 and occur in the right order by the second
case of the definition of happens-before.
(2 ⇒ 1). First observe that since every event e in S 0 occurs at some
process p, if S 0 |p = S|p for all p, then there is a one-to-one correspondence
between events in S 0 and S, and thus S 0 is a permutation of S. Now we need
to show that S 0 is consistent with ⇒. Let e ⇒ e0 . There are three cases.
S S
CHAPTER 6. LOGICAL CLOCKS 43

1. e and e0 are events of the same process p and e <S e0 . But then e <S 0 e0
because S|p = S 0 |p.

2. e is a send event and e0 is the corresponding receive event. Then

e <S 0 e0 because S 0 is the schedule of an execution fragment.

3. e ⇒ e0 by transitivity. Then each step in the chain connecting e to e0

S
uses one of the previous cases, and e <S 0 e0 by transitivity of <S 0 .

There are two main applications for causal shuffles:

1. We can prove upper bounds by using a causal shuffle to turn some

arbitrary S into a nice S 0 , and argue that the niceness of S 0 means
that S at least looks nice to the processes. An example of this can be
found in Lemma 7.1.1.

2. We can prove lower bounds by using a causal shuffle to turn some

specific S into a nasty S 0 , and argue that the existence of S 0 tells us
that there exist bad schedules for some particular problem. An example
of this can be found in §7.4.2. This works particularly well because ⇒
S
includes enough information to determine the latest possible time of
any event in either S or S 0 , so rearranging schedules like this doesn’t
change the worst-case time.

In both cases, we are using the fact that if I tell you ⇒, then you know
S
everything there is to know about the order of events in S that you can
deduce from reports from each process together with the fact that messages
don’t travel back in time.
In the case that we want to use this information inside an algorithm, we
run into the issue that ⇒ is a pretty big relation (Θ(|S|2 ) bits with a naive
S
encoding), and seems to require global knowledge of <S to compute. So we
can ask if there is some simpler, easily computable description that works
almost as well. This is where logical clocks come in.

6.2 Implementations
The basic idea of a logical clock is to compute a timestamp for each event,
so that comparing timestamps gives information about ⇒. Note that these
S
timestamps need not be totally ordered. In general, we will have a relation
CHAPTER 6. LOGICAL CLOCKS 44

<L between timestamps such that e ⇒ e0 implies e <L e0 , but it may be that
S
there are some pairs of events that are ordered by the logical clock despite
being incomparable in the happens-before relation.
Examples of logical clocks that use small timestamps but add extra
ordering are Lamport clocks [Lam78], discussed in §6.2.1; and Neiger-Toueg-
Welch clocks [NT87, Wel87], discussed in §6.2.2. These both assign integer
timestamps to events and may order events that are not causally related.
The main difference between them is that Lamport clocks do not alter the
underlying execution, but may allow arbitrarily large jumps in the logical
clock values; while Neiger-Toueg-Welch clocks guarantee small increments at
the cost of possibly delaying parts of the system.1
More informative are vector clocks [Fid91, Mat93], discussed in §6.2.3.
These use n-dimensional vectors of integers to capture ⇒ exactly, at the cost
S
of much higher overhead.

6.2.1 Lamport clock

Lamport’s logical clock [Lam78] runs on top of any other message-passing
protocol, adding additional state at each process and additional content to
the messages (which is invisible to the underlying protocol). Every process
maintains a local variable clock. When a process sends a message or executes
an internal step, it sets clock ← clock + 1 and assigns the resulting value
as the clock value of the event. If it sends a message, it piggybacks the
resulting clock value on the message. When a process receives a message with
timestamp t, it sets clock ← max(clock, t) + 1; the resulting clock value is
taken as the time of receipt of the message. (To make life easier, we assume
messages are received one at a time.)

Theorem 6.2.1. If we order all events by clock value, we get an execution

of the underlying protocol that is locally indistinguishable from the original
execution.

Proof. Let e <L e0 if e has a lower clock value than e0 . If e and e0 are two
events of the same process, then e <L e0 . If e and e0 are send and receive
events of the same message, then again e <L e0 . So for any events e, e0 , if
e ⇒ e0 , then e <L e0 . Now apply Lemma 6.1.1.
S
1
This makes them similar to synchronizers, which we will discuss in Chapter 7.
CHAPTER 6. LOGICAL CLOCKS 45

6.2.2 Neiger-Toueg-Welch clock

Lamport’s clock has the advantage of requiring no changes in the behavior
of the underlying protocol, but has the disadvantage that clocks are entirely
under the control of the logical-clock protocol and may as a result make
huge jumps when a message is received. If this is unacceptable—perhaps the
protocol needs to do some unskippable maintenance task every 1000 clock
ticks—then an alternative approach due to Neiger and Toueg [NT87] and
Welch [Wel87] can be used.
Method: Each process maintains its own variable clock, which it in-
crements whenever it feels like it. To break ties, the process extends the
clock value to hclock, id, eventCounti where eventCount is a count of send and
receive events (and possibly local computation steps). As in Lamport’s clock,
each message in the underlying protocol is timestamped with the current
extended clock value. Because the protocol can’t change the clock values on
its own, when a message is received with a timestamp later than the current
extended clock value, its delivery is delayed until clock exceeds the message
timestamp, at which point the receive event is assigned the extended clock
value of the time of delivery.

Theorem 6.2.2. If we order all events by clock value, we get an execution

of the underlying protocol that is locally indistinguishable from the original
execution.

Proof. Again, we have that (a) all events at the same process occur in
increasing order (since the event count rises even if the clock value doesn’t,
and we assume that the clock value doesn’t drop) and (b) all receive events
occur later than the corresponding send event (since we force them to). So
Lemma 6.1.1 applies.

The advantage of the Neiger-Toueg-Welch clock is that it doesn’t impose

any assumptions on the clock values, so it is possible to make clock be a
real-time clock at each process and nonetheless have a causally-consistent
ordering of timestamps even if the local clocks are not perfectly synchronized.
If some process’s clock is too far off, it will have trouble getting its messages
delivered quickly (if its clock is ahead) or receiving messages (if its clock is
behind)—the net effect is to add a round-trip delay to that process equal
to the difference between its clock and the clock of its correspondent. But
the protocol works well when the processes’ clocks are closely synchronized,
which is a reasonable assumption in many systems thanks to the Network
CHAPTER 6. LOGICAL CLOCKS 46

Time Protocol, cheap GPS receivers, and clock synchronization mechanisms

built into most cellular phone networks.2

6.2.3 Vector clocks

Logical clocks give a superset of the happens-before relation: if e ⇒ e0 , then
S
e <L e0 (or conversely, if e 6<L e0 , then it is not the case that e ⇒ e0 ). This
S
is good enough for most applications, but what if we want to compute ⇒
S
exactly?
Here we can use a vector clock, invented independently by Fidge [Fid91]
and Mattern [Mat93]. Instead of a single clock value, each event is stamped
with a vector of values, one for each process. When a process executes a
local event or a send event, it increments only its own component xp of the
vector. When it receives a message, it increments xp and sets each xq to the
max of its previous value and the value of xq piggybacked on the message.
We define VC(e) ≤ VC(e0 ), where VC(e) is the value of the vector clock for e,
if VC(e)i ≤ VC(e0 )i for all i.
Theorem 6.2.3. Fix a schedule S; then for any e, e0 , V C(e) < V C(e0 ) if
and only if e ⇒ e0 .
S

Proof. The if part follows immediately from the update rules for the vector
clock. For the only if part, suppose e does not happen-before e0 . Then e and
e0 are events of distinct processes p and p0 . For VC(e) < VC(e0 ) to hold, we
must have VC(e)p < VC(e0 )p ; but this can occur only if the value of VC(e)p
is propagated to p0 by some sequence of messages starting at p and ending
at p0 at or before e0 occurs. In this case we have e ⇒ e0 .
S

6.3 Consistent snapshots

A consistent snapshot of a message-passing computation is a description
of the states of the processes (and possibly messages in transit, but we
can reduce this down to just states by keeping logs of messages sent and
received) that gives the global configuration at some instant of a schedule
that is a consistent reordering of the real schedule (a consistent cut in
the terminology of [AW04, §6.1.2]. Without shutting down the protocol
2
As I write this, my computer reports that its clock is an estimated 289 microseconds
off from the timeserver it is synchronized to, which is less than a tenth of the round-trip
delay to machines on the same local-area network and a tiny fraction of the round-trip
delay to machines elsewhere, including the timeserver machine.
CHAPTER 6. LOGICAL CLOCKS 47

before taking a snapshot this is the about the best we can hope for in a
message-passing system.
Logical clocks can be used to obtain consistent snapshots: pick some
logical clock time and have each process record its state at this time (i.e.,
immediately after its last step before the time or immediately before its first
step after the time). We have already argued that the logical clock gives a
consistent reordering of the original schedule, so the set of values recorded is
just the configuration at the end of an appropriate prefix of this reordering.
In other words, it’s a consistent snapshot.
If we aren’t building logical clocks anyway, there is a simpler consistent
snapshot algorithm due to Chandy and Lamport [CL85]. Here some central
initiator broadcasts a snap message, and each process records its state and
immediately forwards the snap message to all neighbors when it first receives
a snap message. To show that the resulting configuration is a configuration
of some consistent reordering, observe that (with FIFO channels) no process
receives a message before receiving snap that was sent after the sender sent
snap: thus causality is not violated by lining up all the pre-snap operations
before all the post-snap ones.3
The full Chandy-Lamport algorithm adds a second marker message that is
used to sweep messages in transit out of the communications channels, which
avoids the need to keep logs if we want to reconstruct what messages are in
transit (this can also be done with the logical clock version). The idea is that
when a process records its state after receiving the snap message, it issues
a marker message on each outgoing channel. For incoming channels, the
process all records all messages received between the snapshot and receiving
a marker message on that channel (or nothing if it receives marker before
receiving snap). A process only reports its value when it has received a
marker on each channel. The marker and snap messages can also be combined
if the broadcast algorithm for snap resends it on all channels anyway, and a
further optimization is often to piggyback both on messages of the underlying
protocol if the underlying protocol is chatty enough.
Note that Chandy-Lamport is equivalent to the logical-time snapshot
using Lamport clocks, if the snap message is treated as a message with a
very large timestamp. For Neiger-Toueg-Welch clocks, we get an algorithm
where processes spontaneously decide to take snapshots (since Neiger-Toueg-
Welch clocks aren’t under the control of the snapshot algorithm) and delay
3
If FIFO channels are not available, they can be simulated in the absence of failures by
adding a sequence number to each outgoing message on a given channel, and processing
messages at the recipient only when all previous messages have been processed.
CHAPTER 6. LOGICAL CLOCKS 48

post-snapshot messages until the local snapshot has been taken. This can be
implemented as in Chandy-Lamport by separating pre-snapshot messages
from post-snapshot messages with a marker message, and essentially turns
into Chandy-Lamport if we insist that a process advance its clock to the
snapshot time when it receives a marker.

6.3.1 Property testing

Consistent snapshots are in principle useful for debugging (since one can
gather a consistent state of the system without being able to talk to every
process simultaneously), and in practice are mostly used for detecting stable
properties of the system. Here a stable property is some predicate on
global configurations that remains true in any successor to a configuration
in which it is true, or (bending the notion of properties a bit) functions
on configurations whose values don’t change as the protocol runs. Typical
examples are quiescence and its evil twin, deadlock. More exotic examples
include total money supply in a banking system that cannot create or destroy
money, or the fact that every process has cast an irrevocable vote in favor
of some proposal or advanced its Neiger-Toueg-Welch-style clock past some
threshold.
The reason we can test such properties using consistent snapshot is
that when the snapshot terminates with value C in some configuration C 0 ,
even though C may never have occurred during the actual execution of the
protocol, there is an execution which leads from C to C 0 . So if P holds in
C, stability means that it holds in C 0 .
Naturally, if P doesn’t hold in C, we can’t say much. So in this case we
re-run the snapshot protocol and hope we win next time. If P eventually
holds, we will eventually start the snapshot protocol after it holds and obtain
a configuration (which again may not correspond to any global configuration
that actually occurs) in which P holds.
Chapter 7

Synchronizers

Synchronizers simulate an execution of a failure-free synchronous system

in a failure-free asynchronous system. See [AW04, Chapter 11] or [Lyn96,
Chapter 16] for a detailed (and rigorous) presentation.

7.1 Definitions
Formally, a synchronizer sits between the underlying network and the pro-
cesses and does one of two things:

• A global synchronizer guarantees that no process receives a message

from round r until all processes have sent their messages for round r.

• A local synchronizer guarantees that no process receives a mes-

sage from round r until all of that process’s neighbors have sent their
messages for round r.

In both cases, the synchronizer packages all the incoming round r mes-
sages m for a single process together and delivers them as a single action
recv(p, m, r). Similarly, a process is required to hand over all of its outgoing
round-r messages to the synchronizer as a single action send(p, m, r)—this
prevents a process from changing its mind and sending an extra round-r
message or two. It is easy to see that the global synchronizer produces
executions that are effectively indistinguishable from synchronous executions,
assuming that a synchronous execution is allowed to have some variability in
exactly when within a given round each process does its thing. The local
synchronizer only guarantees an execution that is locally indistinguishable
from an execution of the global synchronizer: an individual process can’t

49
CHAPTER 7. SYNCHRONIZERS 50

tell the difference, but comparing actions at different (especially widely sepa-
rated) processes may reveal some process finishing round r + 1 while others
are still stuck in round r or earlier. Whether this is good enough depends on
what you want: it’s bad for coordinating simultaneous missile launches, but
may be just fine for adapting a synchronous message-passing algorithm (as
with distributed breadth-first search as described in §4.3) to an asynchronous
system, if we only care about the final states of the processes and not when
precisely those states are reached.
Formally, the relation between global and local synchronization is de-
scribed by the following lemma:

Lemma 7.1.1. For any schedule S of a locally synchronous execution, there

is a schedule S 0 of a globally synchronous execution such that S|p = S 0 |p for
all processes p.

Proof. Essentially, we use the same happens-before relation as in Chapter 6,

and the fact that if a schedule S 0 is a causal shuffle of another schedule S
(i.e., a permutation of T that preserves causality), then S 0 |p = S|p for all p
(Lemma 6.1.1).
Given a schedule S, consider a schedule S 0 in which the events are ordered
first by increasing round and then by putting all sends before receives. This
ordering is consistent with ⇒, so it’s a causal shuffle of S and S 0 |p = S|p.
S
But it is globally synchronized, because no round r operation ever happens
before a round (r − 1) operation.

7.2 Implementations
Here we describe several implementations of synchronizers. All of them give
at least local synchrony. One of them, the beta synchronizer (§7.2.2), also
gives global synchrony.
The names were chosen by their inventor, Baruch Awerbuch [Awe85].
The main difference between them is the mechanism used to determine when
round-r messages have been delivered.
In the alpha synchronizer, every node sends a message to every neigh-
bor in every round (possibly a dummy message if the underlying protocol
doesn’t send a message); this allows the receiver to detect when it’s gotten
all its round-r messages (because it expects to get a message from every
neighbor) but may produce huge blow-ups in message complexity in a dense
graph.
CHAPTER 7. SYNCHRONIZERS 51

In the beta synchronizer, messages are acknowledged by their receivers

(doubling the message complexity), so the senders can detect when all of
their messages are delivered. But now we need a centralized mechanism to
collect this information from the senders and distribute it to the receivers,
since any particular receiver doesn’t know which potential senders to wait for.
This blows up time complexity, as we essentially end up building a global
synchronizer with a central leader.
The gamma synchronizer combines the two approaches at different
levels to obtain a trade-off between messages and time that depends on the
structure of the graph and how the protocol is organized.
Details of each synchronizer are given below.

7.2.1 The alpha synchronizer

The alpha synchronizer uses local information to construct a local synchro-
nizer. In round r, the synchronizer at p sends p’s message (tagged with the
round number) to each neighbor p0 or noMsg(r) if it has no messages. When
it collects a message or noMsg from each neighbor for round r, it delivers
all the messages. It’s easy to see that this satisfies the local synchronization
specification.
This produces no change in time but may drastically increase message
complexity because of all the extra noMsg messages flying around. For a
synchronous protocol that runs in T rounds with M messages, the same
protocol running with the alpha synchronizer will still run in T time units,
but the message complexity will go up to T · |E| messages, or worse if the
original algorithm doesn’t detect termination.

7.2.2 The beta synchronizer

The beta synchronizer centralizes detection of message delivery using a rooted
directed spanning tree (previously constructed). When p0 receives a round-r
message from p, it responds with ack(r). When p collects an ack for all the
messages it sent plus an OK from all of its children, it sends OK to its parent.
When the root has all the ack and OK messages it is expecting, it broadcasts
go. Receiving go makes p deliver the queued round-r messages.
This works because in order for the root to issue go, every round-r
message has to have gotten an acknowledgment, which means that all round-
r messages are waiting in the receivers’ buffers to be delivered. For the beta
synchronizer, message complexity for one round increases slightly from M to
CHAPTER 7. SYNCHRONIZERS 52

2M + 2(n − 1), but time complexity goes up by a factor proportional to the

depth of the tree.

7.2.3 The gamma synchronizer

The gamma synchronizer combines the alpha and beta synchronizers to try
to get low blowups on both time complexity and message complexity. The
essential idea is to cover the graph with a spanning forest and run beta
within each tree and alpha between trees. Specifically:

• Every message in the underlying protocol gets acked (including mes-

sages that pass between trees).

• When a process has collected all of its outstanding round-r acks, it

sends OK up its tree.

• When the root of a tree gets all acks and OK, it sends ready to the
roots of all adjacent trees (and itself). Two trees are adjacent if any of
their members are adjacent.

• When the root collects ready from itself and all adjacent roots, it
broadcasts go through its own tree.

As in the alpha synchronizer, we can show that no root issues go unless it

and all its neighbors issue ready, which happens only after both all nodes in
the root’s tree and all their neighbors (some of whom might be in adjacent
trees) have received acks for all messages. This means that when a node
receives go it can safely deliver its bucket of messages.
Message complexity is comparable to the beta synchronizer assuming
there aren’t too many adjacent trees: 2M messages for sends and acks,
plus O(n) messages for in-tree communication, plus O(Eroots ) messages for
root-to-root communication. Time complexity per synchronous round is
proportional to the depth of the trees: this includes both the time for in-tree
communication, and the time for root-to-root communication, which might
need to be routed through leaves.
In a particularly nice graph, the gamma synchronizer can give costs
comparable to the costs of the original synchronous algorithm. An example
in [Lyn96] is a ring of k-cliques, where we build a tree in each clique and get
O(1) time blowup and O(n) added messages. This is compared to O(n/k)
time blowup for the beta synchronizer and O(k) message blowup (or worse)
for the alpha synchronizer. Other graphs may favor tuning the size of the
CHAPTER 7. SYNCHRONIZERS 53

trees in the forest toward the alpha or beta ends of the spectrum, e.g., if
the whole graph is a clique (and we didn’t worry about contention issues),
we might as well just use beta and get O(1) time blowup and O(n) added
messages.

7.3 Applications
See [AW04, §11.3.2] or [Lyn96, §16.5]. The one we have seen is distributed
breadth-first search, where the two asynchronous algorithms we described in
Chapter 4 were essentially the synchronous algorithms with the beta and
alpha synchronizers embedded in them. But what synchronizers give us in
general is the ability to forget about problems resulting from asynchrony
provided we can assume no failures (which may be a very strong assumption)
and are willing to accept a bit of overhead.

7.4 Limitations of synchronizers

Here we show some lower bounds on synchronizers, justifying our previous
claim that failures are trouble and showing that global synchronizers are
necessarily slow in a high-diameter network.

7.4.1 Impossibility with crash failures

These synchronizers all fail badly if some process crashes. In the α synchro-
nizer, the system slowly shuts down as a wave of waiting propagates out
from the dead process. In the β synchronizer, the root never gives the green
light for the next round. The γ synchronizer, true to its hybrid nature, fails
in a way that is a hybrid of these two disasters.
This is unavoidable in the basic asynchronous model, although we don’t
have all the results we need to prove this yet. The idea is that if we are in a
synchronous system with crash failures, it’s possible to solve agreement, the
problem of getting all the processes to agree on a bit (see Chapter 9). But
it’s not possible to solve this problem in an asynchronous system with even
one crash failure (see Chapter 11). Since a synchronous-with-crash-failure
agreement protocol on top of a fault-tolerant synchronizer would give a
solution to an unsolvable problem, the element of this stack that we don’t
know an algorithm for must be the one we can’t do. Hence there are no
fault-tolerant synchronizers.
CHAPTER 7. SYNCHRONIZERS 54

We’ll see more examples of this trick of showing that a particular simula-
tion is impossible because it would allow us to violate impossibility results
later, especially when we start looking at the strength of shared-memory
objects in Chapter 18.

7.4.2 Unavoidable slowdown with global synchronization

The session problem [AFL83] gives a lower bound on the speed of a global
synchronizer, or more generally on any protocol that tries to approximate
synchrony in a certain sense. Recall that in a global synchronizer, our goal is
to produce a simulation that looks synchronous from the outside; that is, that
looks synchronous to an observer that can see the entire schedule. In contrast,
a local synchronizer produces a simulation that looks synchronous from
the inside—the resulting execution is indistinguishable from a synchronous
execution to any of the processes, but an outside observer can see that
different processes execute different rounds at different times. The global
synchronizer we’ve seen takes more time than a local synchronizer; the session
problem shows that this is necessary.
In our description, we will mostly follow [AW04, §6.2.2].
A solution to the session problem is an asynchronous protocol in which
each process repeatedly executes some special action. Our goal is to
guarantee that these special actions group into s sessions, where a session
is an interval of time in which every process executes at least one special
action. We also want the protocol to terminate: this means that in every
execution, every process executes a finite number of special actions.
A synchronous system can solve this problem trivially in s rounds: each
process executes one special action per round. For an asynchronous system, a
lower bound of Attiya and Mavronicolas [AM94] (based on an earlier bound
of Arjomandi, Fischer, and Lynch [AFL83], who defined the problem in
a slightly different communication model), shows that if the diameter of
the network is D, there is no solution to the s-session problem that takes
(s − 1)D time or less in the worst case. The argument is based on reordering
events in any such execution to produce fewer than s sessions, using the
happens-before relation described in Chapter 6.
We now give an outline of the proof that this is expensive. (See [AW04,
§6.2.2] for the real proof.)
Fix some algorithm A for solving the s-session problem, and suppose that
its worst-case time complexity is (s − 1)D or less. Consider some synchronous
execution of A (that is, one where the adversary scheduler happens to arrange
the schedule to be synchronous) that takes (s − 1)D rounds or less. Divide
CHAPTER 7. SYNCHRONIZERS 55

this execution into two segments: an initial segment β that includes all
rounds with special actions, and a suffix δ that includes any extra rounds
where the algorithm is still floundering around. We will mostly ignore δ, but
we have to leave it in to allow for the possibility that whatever is happening
there is important for the algorithm to work (say, to detect termination).
We now want to perform a causal shuffle on β that leaves it with only
s − 1 sessions. Because causal shuffles don’t affect time complexity, this will
give us a new bad execution β 0 δ that has only s − 1 sessions despite taking
(s − 1)D time.
The first step is to chop β into at most s − 1 segments β1 , β2 , . . . of at
most D rounds each. Because a message sent in round i is not delivered until
round i + 1, if we have a chain of k messages, each of which triggers the next,
then if the first message is sent in round i, the last message is not delivered
until round i + k. If the chain has length D, its events (including the initial
send and the final delivery) span D + 1 rounds i, i + 1, . . . , i + D. In this
case the initial send and final delivery are necessarily in different segments
βi and βi+1 .
Now pick processes p0 and p1 at distance D from each other. Then any
chain of messages starting at p0 within some segment reaches p1 after the
end of the segment. It follows that for any events e0 of p0 and e1 of p1 in
the same segment βi , e0 6⇒ e1 . So there exists a causal shuffle of βi that
βδ
puts all events of p0 after all events of p1 .1 By a symmetrical argument, we
can similarly put all events of p1 after all events of p0 . In both cases the
resulting schedule is indistinguishable by all processes from the original.
So now we apply these shuffles to each of the segments βi in alternating
order: p0 goes first in the even-numbered segments and p1 goes first in the
odd-numbered segments, yielding a sequence of shuffled segments βi0 . This
has the effect of putting the p0 events together, as in this example with
(s − 1) = 4:

βδ|(p0 , p1 ) = β1 β2 β3 β4 δ|(p0 , p1 )
= β10 β20 β30 β40 δ|(p0 , p1 )
= (p1 p0 )(p0 p1 )(p1 p0 )(p0 p1 )δ
= p1 (p0 p0 )(p1 p1 )(p0 p0 )p1 δ

(here each p0 , p1 stands in for a sequence of events of each process).

1
Proof: Because e0 6⇒ e1 , we can add e1 < e0 for all events e1 and e0 in βi and still
βδ
have a partial order consistent with ⇒. Now apply topological sort to get the shuffle.
βδ
CHAPTER 7. SYNCHRONIZERS 56

Now let’s count sessions. We can’t end a session until we reach a point
where both processes have taken at least one step since the end of the last
session. If we mark with a slash the earliest places where this can happen,
we get a picture like this:

p1 p0 /p0 p1 /p1 p0 /p0 p1 /p1 δ.

We have at most s − 1 sessions! This concludes the proof.

Chapter 8

Coordinated attack

(See also [Lyn96, §5.1].)

The Two Generals problem was the first widely-known distributed con-
sensus problem, described in 1978 by Jim Gray [Gra78, §5.8.3.3.1], although
the same problem previously appeared under a different name [AEH75].
The setup of the problem is that we have two generals on opposite sides
of an enemy army, who must choose whether to attack the army or retreat.
If only one general attacks, his troops will be slaughtered. So the generals
need to reach agreement on their strategy.
To complicate matters, the generals can only communicate by sending
messages by (unreliable) carrier pigeon. We also suppose that at some point
each general must make an irrevocable decision to attack or retreat. The
interesting property of the problem is that if carrier pigeons can become
lost, there is no protocol that guarantees agreement in all cases unless the
outcome is predetermined (e.g., the generals always attack no matter what
happens). The essential idea of the proof is that any protocol that does
guarantee agreement can be shortened by deleting the last message; iterating
this process eventually leaves a protocol with no messages.
Adding more generals turns this into the coordinated attack problem,
a variant of consensus; but it doesn’t make things any easier.

8.1 Formal description

To formalize this intuition, suppose that we have n ≥ 2 generals in a
synchronous system with unreliable channels—the set of messages received
in round i + 1 is always a subset of the set sent in round i, but it may be
a proper subset (even the empty set). Each general starts with an input 0

57
CHAPTER 8. COORDINATED ATTACK 58

(retreat) or 1 (attack) and must output 0 or 1 after some bounded number

of rounds. The requirements for the protocol are that, in all executions:

Agreement All processes output the same decision (0 or 1).

Validity If all processes have the same input x, and no messages are lost,
all processes produce output x. (If processes start with different inputs
or one or more messages are lost, processes can output 0 or 1 as long
as they all agree.)

Termination All processes terminate in a bounded number of rounds.1

Sadly, there is not protocol that satisfies all three conditions. We show
this in the next section.

8.2 Impossibility proof

To show coordinated attack is impossible,2 we use an indistinguishability
proof .
The basic idea of an indistinguishability proof is this:

• Execution A is indistinguishable from execution B for some process

p if p sees the same things (messages or operation results) in both
executions.

• If A is indistinguishable from B for p, then because p can’t tell which

of these two possible worlds it is in, it returns the same output in both.

So far, pretty dull. But now let’s consider a chain of hypothetical

executions A = A0 A1 . . . Ak = B, where each Ai is indistinguishable from
Ai+1 for some process pi . Suppose also that we are trying to solve an
agreement task, where every process must output the same value. Then since
pi outputs the same value in Ai and Ai+1 , every process outputs the same
1
Bounded means that there is a fixed upper bound on the length of any execution.
We could also demand merely that all processes terminate in a finite number of rounds.
In general, finite is a weaker requirement than bounded, but if the number of possible
outcomes at each step is finite (as they are in this case), they’re equivalent. The reason
is that if we build a tree of all configurations, each configuration has only finitely many
successors, and the length of each path is finite, then König’s lemma (see http://en.
wikipedia.org/wiki/Konig’s_lemma) says that there are only finitely many paths. So we
can take the length of the longest of these paths as our fixed bound. [BG97, Lemma 3.1]
2
Without making additional assumptions, always a caveat when discussing impossibility.
CHAPTER 8. COORDINATED ATTACK 59

value in Ai and Ai+1 . By induction on k, every process outputs the same

value in A and B, even though A and B may be very different executions.
This gives us a tool for proving impossibility results for agreement: show
that there is a path of indistinguishable executions between two executions
that are supposed to produce different output. Another way to picture this:
consider a graph whose nodes are all possible executions with an edge between
any two indistinguishable executions; then the set of output-0 executions
can’t be adjacent to the set of output-1 executions. If we prove the graph is
connected, we prove the output is the same for all executions.
For coordinated attack, we will show that no protocol satisfies all of
agreement, validity, and termination using an indistinguishability argument.
The key idea is to construct a path between the all-0-input and all-1-input
executions with no message loss via intermediate executions that are indis-
tinguishable to at least one process.
Let’s start with A = A0 being an execution in which all inputs are 1 and
all messages are delivered. We’ll build executions A1 , A2 , etc., by pruning
messages. Consider Ai and let m be some message that is delivered in
the last round in which any message is delivered. Construct Ai+1 by not
delivering m. Observe that while Ai is distinguishable from Ai+1 by the
recipient of m, on the assumption that n ≥ 2 there is some other process
that can’t tell whether m was delivered or not (the recipient can’t let that
other process know, because no subsequent message it sends are delivered
in either execution). Continue until we reach an execution Ak in which all
inputs are 1 and no messages are sent. Next, let Ak+1 through Ak+n be
obtained by changing one input at a time from 1 to 0; each such execution
is indistinguishable from its predecessor by any process whose input didn’t
change. Finally, construct Ak+n through Ak+n+k0 by adding back messages
in the reverse process used for A0 through Ak ; note that this might not
result in exactly k new messages, because the number of messages might
depend on the inputs. This gets us to an execution Ak+n+k0 in which all
processes have input 0 and no messages are lost. If agreement holds, then
the indistinguishability of adjacent executions to some process means that
the common output in A0 is the same as in Ak+n+k0 . But validity requires
that A0 outputs 1 and Ak+n+k0 outputs 0: so either agreement or validity is
violated in some execution.
CHAPTER 8. COORDINATED ATTACK 60

8.3 Randomized coordinated attack

So we now know that we can’t solve the coordinated attack problem. But
maybe we want to solve it anyway. The solution is to change the problem.
Randomized coordinated attack is like standard coordinated attack,
but with less coordination. Specifically, we’ll allow the processes to flip
coins to decide what to do, and assume that the communication pattern
(which messages get delivered in each round) is fixed and independent of
the coin-flips. This corresponds to assuming an oblivious adversary that
can’t see what is going on at all or perhaps a content-oblivious adversary
that can only see where messages are being sent but not the contents of the
messages. We’ll also relax the agreement property to only hold with some
high probability:

Randomized agreement For any adversary A, the probability that some

process decides 0 and some other process decides 1 given A is at most
.

Validity and termination are as before.

8.3.1 An algorithm
Here’s an algorithm that gives  = 1/r. (See [Lyn96, §5.2.2] for details
or [VL92] for the original version.) A simplifying assumption is that network
is complete, although a strongly-connected network with r greater than or
equal to the diameter also works.

• First part: tracking information levels

– Each process tracks its “information level,” initially 0. The state

of a process consists of a vector of (input, information-level) pairs
for all processes in the system. Initially this is (my-input, 0) for
itself and (⊥, −1) for everybody else.
– Every process sends its entire state to every other process in every
round.
– Upon receiving a message m, process i stores any inputs carried in
m and, for each process j, sets leveli [j] to max(leveli [j], levelm [j]).
It then sets its own information level to minj (leveli [j]) + 1.

• Second part: deciding the output

– Process 1 chooses a random key value uniformly in the range [1, r].
CHAPTER 8. COORDINATED ATTACK 61

– This key is distributed along with leveli [1], so that every process
with leveli [1] ≥ 0 knows the key.
– A process decides 1 at round r if and only if it knows the key,
its information level is greater than or equal to the key, and all
inputs are 1.

8.3.2 Why it works

Termination Immediate from the algorithm.

Validity • If all inputs are 0, no process sees all 1 inputs (technically

requires an invariant that processes’ non-null views are consistent
with the inputs, but that’s not hard to prove.)
• If all inputs are 1 and no messages are lost, then the information
level of each process after k rounds is k (prove by induction) and
all processes learn the key and all inputs (immediate from first
round). So all processes decide 1.

Randomized Agreement • First prove a lemma: Define levelti [k] to

be the value of leveli [k] after t rounds. Then for all i, j, k, t, (1)
leveli [j]t ≤ levelj [j]t−1 and (2) leveli [k]t − levelj [k]t ≤ 1. As
always, the proof is by induction on rounds. Part (1) is easy and
boring so we’ll skip it. For part (2), we have:
– After 0 rounds, level0i [k] = level0j [k] = −1 if neither i nor j
equals k; if one of them is k, we have level0k [k] = 0, which is
still close enough.
– After t rounds, consider levelti [k] − levelt−1 i [k] and similarly
leveltj [k]−leveljt−1 [k]. It’s not hard to show that each can jump
by at most 1. If both deltas are +1 or both are 0, there’s
no change in the difference in views and we win from the
induction hypothesis. So the interesting case is when leveli [k]
stays the same and levelj [k] increases or vice versa.
– There are two ways for levelj [k] to increase:
∗ If j 6= k, then j received a message from some j 0 with
levelt−1 t−1
j 0 [k] > levelj [k]. From the induction hypothesis,
levelt−1 t−1 t
j 0 [k] ≤ leveli [k] + 1 = leveli [k]. So we are happy.
∗ If j = k, then j has leveltj [j] = 1 + mink6=j leveltj [k] ≤
1 + leveltj [i] ≤ 1 + levelti [i]. Again we are happy.
CHAPTER 8. COORDINATED ATTACK 62

• Note that in the preceding, the key value didn’t figure in; so
everybody’s level at round r is independent of the key.
• So now we have that levelri [i] is in {`, ` + 1}, where ` is some fixed
value uncorrelated with the key. The only way to get some process
to decide 1 while others decide 0 is if ` + 1 ≥ key but ` < key. (If
` = 0, a process at this level doesn’t know key, but it can still
reason that 0 < key since key is in [1, r].) This can only occur if
key = ` + 1, which occurs with probability at most 1/r since key
was chosen uniformly.

8.3.3 Almost-matching lower bound

The bound on the probability of disagreement in the previous algorithm is
almost tight. Varghese and Lynch [VL92] show that no synchronous algorithm
1
can get a probability of disagreement less than r+1 , using a stronger validity
condition that requires that the processes output 0 if any input is 0. This is
a natural assumption for database commit, where we don’t want to commit
if any process wants to abort. We restate their result below:
Theorem 8.3.1. For any synchronous algorithm for randomized coordinated
attack that runs in r rounds that satisfies the additional condition that all
non-faulty processes decide 0 if any input is 0, Pr[disagreement] ≥ 1/(r + 1).
Proof. Let  be the bound on the probability of disagreement. Define levelti [k]
as in the previous algorithm (whatever the real algorithm is doing). We’ll
show Pr[i decides 1] ≤  · (levelri [i] + 1), by induction on levelri [i].
• If levelri [i] = 0, the real execution is indistinguishable (to i) from an
execution in which some other process j starts with 0 and receives no
messages at all. In that execution, j must decide 0 or risk violating
the strong validity assumption. So i decides 1 with probability at most
 (from the disagreement bound).
• If levelri [i] = k > 0, the real execution is indistinguishable (to i) from
an execution in which some other process j only reaches level k − 1
and thereafter receives no messages. From the induction hypothesis,
Pr[j decides 1] ≤ k in that pruned execution, and so Pr[i decides 1] ≤
(k + 1) in the pruned execution. But by indistinguishability, we also
have Pr[i decides 1] ≤ (k + 1) in the original execution.
Now observe that in the all-1 input execution with no messages lost,
levelri [i] = r and Pr[i decides 1] = 1 (by validity). So 1 ≤ (r + 1), which
implies  ≥ 1/(r + 1).
Chapter 9

Synchronous agreement

Here we’ll consider synchronous agreement algorithm with stopping failures,

where a process stops dead at some point, sending and receiving no further
messages. We’ll also consider Byzantine failures, where a process deviates
from its programming by sending arbitrary messages, but mostly just to see
how crash-failure algorithms hold up; for algorithms designed specifically for
a Byzantine model, see Chapter 10.
If the model has communication failures instead, we have the coordinated
attack problem from Chapter 8.

9.1 Problem definition

We use the usual synchronous model with n processes with binary inputs and
binary outputs. Up to f processes may fail at some point; when a process
fails, one or one or more of its outgoing messages are lost in the round of
failure and all outgoing messages are lost thereafter.
There are two variants on the problem, depending on whether we want
a useful algorithm (and so want strong conditions to make our algorithm
more useful) or a lower bound (and so want weak conditions to make our
lower bound more general). For algorithms, we will ask for these conditions
to hold:

Agreement All non-faulty processes decide the same value.

Validity If all processes start with the same input, all non-faulty processes
decide it.

Termination All non-faulty processes eventually decide.

63
CHAPTER 9. SYNCHRONOUS AGREEMENT 64

For lower bounds, we’ll replace validity with non-triviality (often called
validity in the literature):

Non-triviality There exist failure-free executions A and B that produce

different outputs.

Non-triviality follows from validity but doesn’t imply validity; for example,
a non-trivial algorithm might have the property that if all non-faulty processes
start with the same input, they all decide something else.
In §9.2, we’ll show that a simple algorithm gives agreement, termination,
and validity with f failures using f + 1 rounds. We’ll then show in §9.3 that
non-triviality, agreement, and termination imply that f + 1 rounds is the
best possible. In Chapter 10, we’ll show that the agreement is still possible
in f + 1 rounds even if faulty processes can send arbitrary messages instead
of just crashing, but only if the number of faulty processes is strictly less
than n/3.

9.2 Solution using flooding

The flooding algorithm, due to Dolev and Strong [DS83] gives a straightfor-
ward solution to synchronous agreement for the crash failure case. It runs in
f + 1 rounds assuming f crash failures. (This algorithm is also described in
more detail in [AW04, §5.1.3] or [Lyn96, §6.2.1].)
Each process keeps a set of (process, input) pairs, initially just {(myId, myInput)}.
At round r, I broadcast my set to everybody and take the union of my set
and all sets I receive. At round f + 1, I decide on f (S), where f is some
fixed function from sets of process-input pairs to outputs that picks some
input in S: for example, f might take the input with the smallest process-id
attached to it, take the max of all known input values, or take the majority
of all known input values.

Lemma 9.2.1. After f + 1 rounds, all non-faulty processes have the same
set.

Proof. Let Sir be the set stored by process i after r rounds. What we’ll really
show is that if there are no failures in round k, then Sir = Sjr = Sik+1 for all
i, j, and r > k. To show this, observe that no faults in round k means that
all processes that are still alive at the start of round k send their message
to all other processes. Let L be the set of live processes in round k. At the
end of round k, for i in L we have Sik+1 = j∈L Sjk = S. Now we’ll consider
S

some round r = k + 1 + m and show by induction on m that Sik+m = S; we

CHAPTER 9. SYNCHRONOUS AGREEMENT 65

already did m = 0, so for larger m notice that all messages are equal to S
and so Sik+1+m is the union of a whole bunch of S’s. So in particular we
have Sif +1 = S (since some failure-free round occurred in the preceding f + 1
rounds) and everybody decides the same value f (S).

Flooding depends on being able to trust second-hand descriptions of

values; it may be that process 1 fails in round 0 so that only process 2 learns
its input. If process 2 can suddenly tell 3 (but nobody else) about the input
in round f + 1—or worse, tell a different value to 3 and 4—then we may
get disagreement. This remains true even if Byzantine processes can’t fake
inputs (e.g., because an input value is really a triple (i, v, signature(v)) using
an unforgeable digital signature)—the reason is that a Byzantine process
could horde some input (i, v, signature(v)) until the very last round and then
deliver it to only some of the non-faulty processes.

9.3 Lower bound on rounds

Here we show that synchronous agreement requires at least f + 1 rounds
if f processes can fail. This proof is modeled on the one in [Lyn96, §6.7]
and works backwards from the final state; for a proof of the same result
that works in the opposite direction, see [AW04, §5.1.4]. The original result
(stated for Byzantine failures) is due to Dolev and Strong [DS83], based on
a more complicated proof due to Fischer and Lynch [FL82]; see the chapter
notes for Chapter 5 of [AW04] for more discussion of the history.
Note that unlike the algorithms in the preceding and following sections,
which provide validity, the lower bound applies even if we only demand
non-triviality.
Like the similar proof for coordinated attack (§8.2), the proof uses an
indistinguishability argument. But we have to construct a more complicated
chain of intermediate executions.
A crash failure at process i means that (a) in some round r, some or
all of the messages sent by i are not delivered, and (b) in subsequent rounds,
no messages sent by i are delivered. The intuition is that i keels over dead
in the middle of generating its outgoing messages for a round. Otherwise i
behaves perfectly correctly. A process that crashes at some point during an
execution is called faulty
We will show that if up to f processes can crash, and there are at least
f + 2 processes,1 then at least f + 1 rounds are needed (in some execution)
1
With only f + 1 processes, we can solve agreement in f rounds using flooding. The
CHAPTER 9. SYNCHRONOUS AGREEMENT 66

for any algorithm that satisfies agreement, termination, and non-triviality.

In particular, we will show that if all executions run in f or fewer rounds,
then the indistinguishability graph is connected; this implies non-triviality
doesn’t hold, because (as in §8.2), two adjacent states must decide the same
value because of the agreement property.2
Now for the proof. To simplify the argument, let’s assume that all
executions terminate in exactly f rounds (we can always have processes send
pointless chitchat to pad out short executions) and that every processes sends
a message to every other process in every round where it has not crashed
(more pointless chitchat). Formally, this means we have a sequence of rounds
0, 1, 2, . . . , f − 1 where each process sends a message to every other process
(assuming no crashes), and a final round f where all processes decide on a
value (without sending any additional messages).
We now want to take any two executions A and B and show that both
produce the same output. To do this, we’ll transform A’s inputs into B’s
inputs one process at a time, crashing processes to hide the changes. The
problem is that just crashing the process whose input changed might change
the decision value—so we have to crash later witnesses carefully to maintain
indistinguishability all the way across the chain.
Let’s say that a process p crashes fully in round r if it crashes in round
r and no round-r messages from p are delivered. The communication
pattern of an execution describes which messages are delivered between
processes without considering their contents—in particular, it tells us which
processes crash and what other processes they manage to talk to in the round
in which they crash.
With these definitions, we can state and prove a rather complicated
induction hypothesis:

Lemma 9.3.1. For any f -round protocol with n ≥ f + 2 processes permitting

up to f crash failures; any process p; and any execution A in which at
most one process crashes per round in rounds 0 . . . r − 1, p crashes fully in
round r + 1, and no other processes crash; there is a sequence of executions
A = A0 A1 . . . Ak such that each Ai is indistinguishable from Ai+1 by some
idea is that either (a) at most f − 1 processes crash, in which case the flooding algorithm
guarantees agreement; or (b) exactly f processes crash, in which case the one remaining
non-faulty process agrees with itself. So f + 2 processes are needed for the lower bound
to work, and we should be suspicious of any lower bound proof that does not use this
assumption.
2
The same argument works with even a weaker version of non-triviality that omits the
requirement that A and B are failure-free, but we’ll keep things simple.
CHAPTER 9. SYNCHRONOUS AGREEMENT 67

process, each Ai has at most one crash per round, and the communication
pattern in Ak is identical to A except that p crashes fully in round r.

Proof. By induction on f − r. If r = f , we just crash p in round r and

nobody else notices. For r < f , first crash p in round r instead of r + 1, but
deliver all of its round-r messages anyway (this is needed to make space for
some other process to crash in round r + 1). Then choose some message m
sent by p in round r, and let p0 be the recipient of m. We will show that we
can produce a chain of indistinguishable executions between any execution
in which m is delivered and the corresponding execution in which it is not.
If r = f − 1, this is easy; only p0 knows whether m has been delivered,
and since n ≥ f + 2, there exists another non-faulty p00 that can’t distinguish
between these two executions, since p0 sends no messages in round f or later.
If r < f − 1, we have to make sure p0 doesn’t tell anybody about the missing
message.
By the induction hypothesis, there is a sequence of executions starting
with A and ending with p0 crashing fully in round r + 1, such that each
execution is indistinguishable from its predecessor. Now construct the
sequence

A → (A with p0 crashing fully in r + 1)

→ (A with p0 crashing fully in r + 1 and m lost)
→ (A with m lost and p0 not crashing).

The first and last step apply the induction hypothesis; the middle one yields
indistinguishable executions since only p0 can tell the difference between m
arriving or not and its lips are sealed.
We’ve shown that we can remove one message through a sequence of
executions where each pair of adjacent executions is indistinguishable to
some process. Now paste together n − 1 such sequences (one per message)
to prove the lemma.

The rest of the proof: Crash some process fully in round 0 and then
change its input. Repeat until all inputs are changed.

9.4 Variants
So far we have described binary consensus, since all inputs are 0 or 1. We
can also allow larger input sets. With crash failures, this allows a stronger
validity condition: the output must be equal to some non-faulty process’s
CHAPTER 9. SYNCHRONOUS AGREEMENT 68

input. It’s not hard to see that Dolev-Strong (§9.2) gives this stronger
condition.
Chapter 10

Byzantine agreement

Like synchronous agreement (as in Chapter 9) except that we replace crash

failures with Byzantine failures, where a faulty process can ignore its
programming and send any messages it likes. Since we are operating under
a universal quantifier, this includes the case where the Byzantine processes
appear to be colluding with each other under the control of a centralized
adversary.

10.1 Lower bounds

We’ll start by looking at lower bounds.

10.1.1 Minimum number of rounds

We’ve already seen an f +1 lower bound on rounds for crash failures (see §9.3).
This lower bound applies a fortiori to Byzantine failures, since Byzantine
failures can simulate crash failures.

10.1.2 Minimum number of processes

We can also show that we need n > 3f processes. For n = 3 and f = 1 the
intuition is that Byzantine B can play non-faulty A and C off against each
other, telling A that C is Byzantine and C that A is Byzantine. Since A is
telling C the same thing about B that B is saying about A, C can’t tell the
difference and doesn’t know who to believe. Unfortunately, this tragic soap
opera is not a real proof, since we haven’t actually shown that B can say
exactly the right thing to keep A and C from guessing that B is evil.

69
CHAPTER 10. BYZANTINE AGREEMENT 70

A0 B0 A0 B0

C1 C0

Č B1 A1

Figure 10.1: Three-process vs. six-process execution in Byzantine agreement

lower bound. Processes A0 and B0 in right-hand execution receive same
messages as in left-hand three-process execution with Byzantine Č simulation
C0 through C1 . So validity forces them to decide 0. A similar argument
using Byzantine Ǎ shows the same for C0 .

The real proof:1 Consider an artificial execution where (non-Byzantine)

A, B, and C are duplicated and then placed in a ring A0 B0 C0 A1 B1 C1 , where
the digits indicate inputs. We’ll still keep the same code for n = 3 on A0 , B0 ,
etc., but when A0 tries to send a message to what it thinks of as just C we’ll
send it to C1 while messages from B0 will instead go to C0 . For any adjacent
pair of processes (e.g. A0 and B0 ), the behavior of the rest of the ring could
be simulated by a single Byzantine process (e.g. C), so each process in the
6-process ring behaves just as it does in some 3-process execution with 1
Byzantine process. It follows that all of the processes terminate and decide
in the unholy 6-process Frankenexecution2 the same value that they would in
the corresponding 3-process Byzantine execution. So what do they decide?
Given two processes with the same input, say, A0 and B0 , the giant
execution is indistinguishable from an A0 B0 Č execution where Č is Byzantine
(see Figure 10.1. Validity says A0 and B0 must both decide 0. Since this
works for any pair of processes with the same input, we have each process
deciding its input. But now consider the execution of C0 A1 B̌, where B̌ is
Byzantine. In the big execution, we just proved that C0 decides 0 and A1
decides 1, but since the C0 A1 B execution is indistinguishable from the big
execution to C0 and A1 , they do the same thing here and violate agreement.
This shows that with n = 3 and f = 1, we can’t win. We can generalize
this to n = 3f . Suppose that there were an algorithm that solved Byzantine
agreement with n = 3f processes. Group the processes into groups of size f ,
and let each of the n = 3 processes simulate one group, with everybody in
1
The presentation here is based on [AW04, §5.2.3]. The original impossibility result
is due to Pease, Shostak, and Lamport [PSL80]. This particular proof is due to Fischer,
Lynch, and Merritt [FLM86].
2
Not a real word.
CHAPTER 10. BYZANTINE AGREEMENT 71

B0

B0 A0 D0

A0 D0 C0
C1
Č A1
B1
D1

Figure 10.2: Four-process vs. eight-process execution in Byzantine agreement

connectivity lower bound. Because Byzantine Č can simulate C0 , D1 , B1 , A1 ,
and C1 , good processes A0 , B0 and D0 must all decide 0 or risk violating
validity.

the group getting the same input, which can only make things easier. Then
we get a protocol for n = 3 and f = 1, an impossibility.

10.1.3 Minimum connectivity

So far, we’ve been assuming a complete communication graph. If the graph is
not complete, we may not be able to tolerate as many failures. In particular,
we need the connectivity of the graph (minimum number of nodes that must
be removed to split it into two components) to be at least 2f + 1. See [Lyn96,
§6.5] for the full proof. The essential idea is that if we have an arbitrary
graph with a vertex cut of size k < 2f + 1, we can simulate it on a 4-process
graph where A is connected to B and C (but not D), B and C are connected
to each other, and D is connected only to B and C. Here B and C each
simulate half the processes in the size-k cut, A simulates all the processes
on one side of the cut and D all the processes on the other side. We then
construct an 8-process artificial execution with two non-faulty copies of each
of A, B, C, and D and argue that if one of B or C can be Byzantine then
the 8-process execution is indistinguishable to the remaining processes from
a normal 4-process execution. (See Figure 10.1.)
An argument similar to the n > 3f proof then shows we violate one of
validity or agreement: if we replacing C0 , C1 , and all the nodes on one side of
the C0 + C1 cut with a single Byzantine Č, we force the remaining non-faulty
nodes to decide their inputs or violate validity. But then doing the same
CHAPTER 10. BYZANTINE AGREEMENT 72

thing with B0 and B1 yields an execution that violates agreement.

Conversely, if we have connectivity 2f +1, then the processes can simulate
a general graph by sending each other messages along 2f + 1 predetermined
vertex-disjoint paths and taking the majority value as the correct message.
Since the f Byzantine processes can only corrupt one path each (assuming
the non-faulty processes are careful about who they forward messages from),
we get at least f +1 good copies overwhelming the f bad copies. This reduces
the problem on a general graph with sufficiently high connectivity to the
problem on a complete graph, allowing Byzantine agreement to be solved if
the other lower bounds are met.

10.1.4 Weak Byzantine agreement

(Here we are following [Lyn96, §6.6]. The original result is due to Lam-
port [Lam83].)
Weak Byzantine agreement is like regular Byzantine agreement, but
validity is only required to hold if there are no faulty processes at all. If
there is a single faulty process, the non-faulty processes can output any value
regardless of their inputs (as long as they agree on it). Sadly, this weakening
doesn’t improve things much: even weak Byzantine agreement can be solved
only if n ≥ 3f + 1.
Proof: As in the strong Byzantine agreement case, we’ll construct a many-
process Frankenexecution to figure out a strategy for a single Byzantine
process in a 3-process execution. The difference is that now the number of
processes in our synthetic execution is much larger, since we want to build
an execution where at least some of our test subjects think they are in a non-
Byzantine environment. The trick is to build a very big, highly-symmetric
ring so that at least some of the processes are so far away from the few
points of asymmetry that might clue them in to their odd condition that the
protocol terminates before they notice.
Fix some protocol that allegedly solves weak Byzantine agreement, and
let r be the number of rounds for the protocol. Construct a ring of 6r pro-
cesses A01 B01 C01 A02 B02 C02 . . . A0r B0r C0r A10 B10 C10 . . . A1r B1r C1r , where
each Xij runs the code for process X in the 3-process protocol with in-
put i. For each adjacent pair of processes, there is a 3-process Byzantine
execution which is indistinguishable from the 6r-process execution for that
pair: since agreement holds in all Byzantine executions, each adjacent pair
decides the same value in the big execution and so either everybody decides
0 or everybody decides 1 in the big execution.
Now we’ll show that means that validity is violated in some no-failures
CHAPTER 10. BYZANTINE AGREEMENT 73

3-process execution. We’ll extract this execution by looking at the execution

of processes A0,r/2 B0,r/2 C0,r/2 . The argument is that up to round r, any
input-0 process that is at least r steps in the ring away from the nearest
1-input process acts like the corresponding process in the all-0 no-failures
3-process execution. Since A0,r/2 is 3r/2 > r hops away from A1r and
similarly for C0,r/2 , our 3 stooges all decide 0 by validity. But now repeat
the same argument for A1,r/2 B1,r/2 C1,r/2 and get 3 new stooges that all
decide 1. This means that somewhere in between we have two adjacent
processes where one decides 0 and one decides 1, violating agreement in the
corresponding 3-process execution where the rest of the ring is replaced by a
single Byzantine process. This concludes the proof.
This result is a little surprising: we might expect that weak Byzantine
agreement could be solved by allowing a process to return a default value if
it notices anything that might hint at a fault somewhere. But this would
allow a Byzantine process to create disagreement revealing its bad behavior
to just one other process in the very last round of an execution otherwise
headed for agreement on the non-default value. The chosen victim decides the
default value, but since it’s the last round, nobody else finds out. Even if the
algorithm is doing something more sophisticated, examining the 6r-process
execution will tell the Byzantine process exactly when and how to start
acting badly.

10.2 Upper bounds

Here we describe two upper bounds for Byzantine agreement, one of which
gets an optimal number of rounds at the cost of many large messages, and
the other of which gets smaller messages at the cost of more rounds. (We
are following §§5.2.4–5.2.5 of [AW04] in choosing these algorithms.) Neither
of these algorithms is state-of-the-art, but they demonstrate some of the
issues in solving Byzantine agreement without the sometimes-complicated
optimizations needed to get all the parameters of the algorithm down simul-
taneously.

10.2.1 Exponential information gathering gets n = 3f + 1

The idea of exponential information gathering is that each process will
do a lot of gossiping, but now its state is no longer just a flat set of inputs,
but a tree describing who it heard what from. We build this tree out of pairs
of the form (path, input) where path is a sequence of intermediaries with no
repetitions and input is some input. A process’s state at each round is just a
CHAPTER 10. BYZANTINE AGREEMENT 74

set of such pairs. At the end of f + 1 rounds of communication (necessary

because of the lower bound for crash failures), each non-faulty process
attempts to untangle the complex web of hearsay and second-hand lies to
compute the same decision value as the other processes. This technique was
used by Pease, Shostak, and Lamport [PSL80] to show that their impossibility
result is tight: there exists an algorithm for Byzantine agreement that runs
in f + 1 synchronous rounds and guarantees agreement and validity as long
as n ≥ 3f + 1.

1 S ← {hhi , inputi}
2 for round ← 0 . . . f do
3 Send {hxi, vi | hx, vi ∈ S ∧ |x| = round ∧ i 6∈ x} to all processes
4 upon receiving S 0 from j do
// Filter out obviously bogus tuples
5 if ∀ hxj 0 , vi ∈ S 0 : |x| = round ∧ j 0 = j then
6 S ← S ∪ S0

// Compute decision value

7 for each path w of length f + 1 with no repeats do
8 if hw, vi ∈ S for some v then
9 Let val0 (w, i) = v
10 else
11 Let val0 (w, i) = 0

12 for each path w of length f or less with no repeats do

13 Let val0 (w, i) = majorityj6∈w val(wj, i)
14 Decide val0 (hi , i)
Algorithm 10.1: Exponential information gathering. Code for pro-
cess i.

The algorithm is given in Algorithm 10.1. The communication phase is

just gossiping, where each process starts with its only its input and forwards
any values it hears about along with their provenance to all of the other
processes. At the end of this phase, each process has a set of pairs of the
form (path, value) where path spans all sequences of 0 to f + 1 distinct ids
and value is the input value forwarded along that path.
We write val(w, i) for the value stored in i’s list at the end of the protocol
that is associated with path w. Because we can’t trust these val(w, i) values
to be an accurate description of any process’s input if there is a Byzantine
process in w, each process computes for itself replacement values val0 (w, i)
CHAPTER 10. BYZANTINE AGREEMENT 75

that use majority voting to try to get a more trustworthy picture of the
original inputs.
Formally, we think of the set of paths as a tree where w is the parent of
wj for each path w and each id j not in w. To apply EIG in the Byzantine
model, ill-formed messages received from j are treated as missing messages,
but otherwise the data-collecting part of EIG proceeds as in the crash failure
model. However, we compute the decision value from the last-round values
recursively as follows. First replace any missing pair involving a path w with
|w| = f + 1 with (w, 0). Then for each path w, define val0 (w, i) to be the
majority value among val0 (wj, i) for all j, or val(w, i) if |w| = f + 1. Finally,
have process i decide val0 (hi , i), which it can compute locally from its own
stored values val(w, i).
Each val0 is a reconstruction of older values from later ones. As we move
up the tree from wj to w we are moving backwards in time, until in the
end we get the decision value val0 (hi, i) as a majority of reconstructed inputs
val0 (j, i). One way to think about this is that I don’t trust j to give me the
right value for wj—even when w = hi and j is claiming to report its own
input—so instead I take a majority of values of wj that j allegedly reported
to other people. But since I don’t trust those other people either, I use the
same process recursively to construct those reports.

10.2.1.1 Proof of correctness

This is just a sketch of the proof from [Lyn96, §6.3.2]; essentially the same
argument appears in [AW04, §5.2.4].
We start with a basic observation that good processes send and record
values correctly:

Lemma 10.2.1. If i, j, and k are all non-faulty then for all w, val(wk, i) =
val(wk, j) = val(w, k).

Proof. Trivial: k announces the same value val(w, k) to both i and j.

More involved is this lemma, which says that when we reconstruct a

value for a trustworthy process at some level, we get the same value that it
sent us. In particular this will be used to show that the reconstructed inputs
val0 (j, i) are all equal to the real inputs for good processes.

Lemma 10.2.2. If j is non-faulty then val0 (wj, i) = val(w, j) for all non-
faulty i and all w.
CHAPTER 10. BYZANTINE AGREEMENT 76

Proof. By induction on f + 1 − |wj|. If |wj| = f + 1, then val0 (wj, i) =

val(wj, i) = val(w, j). If |wj| < f + 1, then val(wj, k) = val(w, j) for all non-
faulty k. It follows that val(wjk, i) = val(w, j) for all non-faulty i and k (that
do not appear in w). The bad guys report at most f bad values val(wj, k 0 ),
but the good guys report at least n − f − |wj| good values val(wj, k). Since
n ≥ 3f + 1 and |wj| ≤ f , we have n − f − |wj| ≥ 3f + 1 − f − f ≥ f + 1
good values, which are a majority.
We call a node w common if val0 (w, i) = val0 (w, j) for all non-faulty i, j.
Lemma 10.2.2 implies that wk is common if k is non-faulty. We can also
show that any node whose children are all common is also common, whether
or not the last process in its label is faulty.
Lemma 10.2.3. Let wk be common for all k. Then w is common.
Proof. Recall that, for |w| < f + 1, val0 (w, i) is the majority value among
all val0 (wk, i). If all wk are common, then val0 (wk, i) = val0 (wk, j) for all
non-faulty i and j, so i and j compute the same majority values and get
val0 (w, i) = val0 (w, j).
We can now prove the full result.
Theorem 10.2.4. Exponential information gathering using f + 1 rounds
in a synchronous Byzantine system with at most f faulty processes satisfies
validity and agreement, provided n ≥ 3f + 1.
Proof. Validity: Immediate application of Lemmas 10.2.1 and 10.2.2 when
w = hi. We have val0 (j, i) = val(j, i) = val(hi, j) for all non-faulty j and i,
which means that a majority of the val0 (j, i) values equal the common input
and thus so does val0 (hi, i).
Agreement: Observe that every path has a common node on it, since a
path travels through f + 1 nodes and one of them is good. If we then suppose
that the root is not common: by Lemma 10.2.3, it must have a not-common
child, that node must have a not-common child, etc. But this constructs
a path from the root to a leaf with no not-common nodes, which we just
proved can’t happen.

10.2.2 Phase king gets constant-size messages

The following algorithm, based on work of Berman, Garay, and Perry [BGP89],
achieves Byzantine agreement in 2(f +1) rounds using constant-size messages,
provided n ≥ 4f + 1. The description here is drawn from [AW04, §5.2.5].
The original Berman-Garay-Perry paper gives somewhat better bounds, but
the algorithm and its analysis are more complicated.
CHAPTER 10. BYZANTINE AGREEMENT 77

10.2.2.1 The algorithm

The basic idea of the algorithm is that we avoid the recursive majority voting
of EIG by running a vote in each of f + 1 phases through a phase king,
some process chosen in advance to run the phase. Since the number of phases
exceeds the number of faults, we eventually get a non-faulty phase king.
The algorithm is structured so that one non-faulty phase king is enough
to generate agreement and subsequent faulty phase kings can’t undo the
agreement.
Pseudocode appears in Algorithm 10.2. Each processes i maintains an
array pref i [j], where j ranges over all process ids. There are also utility
values majority, kingMajority and multiplicity for each process that are used
to keep track of what it hears from the other processes. Initially, pref i [i] is
just i’s input and pref i [j] = 0 for j 6= i.

1 pref i [i] = input

2 for j 6= i do pref i [j] = 0
3 for k ← 1 to f + 1 do
// First round of phase k
4 send pref i [i] to all processes (including myself)
5 pref i [j] ← vj , where vj is the value received from process j
6 majority ← majority value in pref i
7 multiplicity ← number of times majority appears in pref i
// Second round of phase k
8 if i = k then
// I am the phase king
9 send majority to all processes
10 if received m from phase king then
11 kingMajority ← m
12 else
13 kingMajority ← 0
14 if multiplicity > n/2 + f then
15 pref i [i] = majority
16 else
17 pref i [i] = kingMajority

18 return pref i [i]

Algorithm 10.2: Byzantine agreement: phase king

The idea of the algorithm is that in each phase, everybody announces their
CHAPTER 10. BYZANTINE AGREEMENT 78

current preference (initially the inputs). If the majority of these preferences

is large enough (e.g., all inputs are the same), everybody adopts the majority
preference. Otherwise everybody adopts the preference of the phase king.
The majority rule means that once the processes agree, they continue to
agree despite bad phase kings. The phase king rule allows a good phase king
to end disagreement. By choosing a different king in each phase, after f + 1
phases, some king must be good. This intuitive description is justified below.

10.2.2.2 Proof of correctness

Termination is immediate from the algorithm.
For validity, suppose all inputs are v. We’ll show that all non-faulty i
have pref i [i] = v after every phase. In the first round of each phase, process
i receives at least n − f messages containing v; since n ≥ 4f + 1, we have
n − f ≥ 3f + 1 and n/2 + f ≤ (4f + 1)/2 + f = 3f + 1/2, and thus these
n − f messages exceed the n/2 + f threshold for adopting them as the new
preference. So all non-faulty processes ignore the phase king and stick with
v, eventually deciding v after round 2(f + 1).
For agreement, we’ll ignore all phases up to the first phase with a non-
faulty phase king. Let k be the first such phase, and assume that the pref
values are set arbitrarily at the start of this phase. We want to argue that
at the end of the phase, all non-faulty processes have the same preference.
There are two ways that a process can set its new preference in the second
round of the phase:

1. The process i observes a majority of more than n/2 + f identical values

v and ignores the phase king. Of these values, more than n/2 of them
were sent by non-faulty processes. So the phase king also receives these
values (even if the faulty processes change their stories) and chooses
v as its majority value. Similarly, if any other process j observes a
majority of n/2 + f identical values, the two > n/2 non-faulty parts of
the majorities overlap, and so j also chooses v.

2. The process i takes its value from the phase king. We’ve already shown
that i then agrees with any j that sees a big majority; but since the
phase king is non-faulty, process i will agree with any process j that
also takes its new preference from the phase king.

This shows that after any phase with a non-faulty king, all processes
agree. The proof that the non-faulty processes continue to agree is the same
as for validity.
CHAPTER 10. BYZANTINE AGREEMENT 79

10.2.2.3 Performance of phase king

It’s not hard to see that this algorithm sends exactly (f + 1)(n2 + n) messages
of 1 bit each (assuming 1-bit inputs). The cost is doubling the minimum
number of rounds and reducing the tolerance for Byzantine processes. As
mentioned earlier, a variant of phase-king with 3-round phases gets optimal
fault-tolerance with 3(f + 1) rounds (but 2-bit messages). Still better is
a rather complicated descendant of the EIG algorithm due to Garay and
Moses [GM98], which gets f + 1 rounds with n ≥ 3f + 1 while still having
polynomial message traffic.
Chapter 11

Impossibility of
asynchronous agreement

There’s an easy argument that says that you can’t do most things in an
asynchronous message-passing system with n/2 crash failures: partition the
processes into two subsets S and T of size n/2 each, and allow no messages
between the two sides of the partition for some long period of time. Since
the processes in each side can’t distinguish between the other side being
slow and being dead, eventually each has to take action on their own. For
many problems, we can show that this leads to a bad configuration. For
example, for agreement, we can supply each side of the partition with a
different common input value, forcing disagreement because of validity. We
can then satisfy the fairness condition that says all messages are eventually
delivered by delivering the delayed messages across the partition, but it’s
too late for the protocol.
The Fischer-Lynch-Paterson (FLP) result [FLP85] says something much
stronger: you can’t do agreement in an asynchronous message-passing system
if even one crash failure is allowed.1 After its initial publication, it was quickly
generalized to other models including asynchronous shared memory [LAA87],
and indeed the presentation of the result in [Lyn96, §12.2] is given for shared-
memory first, with the original result appearing in [Lyn96, §17.2.3] as a
corollary of the ability of message passing to simulate shared memory. In
these notes, I’ll present the original result; the dependence on the model is
surprisingly limited, and so most of the proof is the same for both shared
memory (even strong versions of shared memory that support operations
1
Unless you augment the basic model in some way, say by adding randomization
(Chapter 23) or failure detectors (Chapter 13).

80
CHAPTER 11. IMPOSSIBILITY OF ASYNCHRONOUS AGREEMENT81

like atomic snapshots2 ) and message passing.

Section 5.3 of [AW04] gives a very different version of the proof, where
it is shown first for two processes in shared memory, then generalized to n
processes in shared memory by adapting the classic Borowsky-Gafni simu-
lation [BG93] to show that two processes with one failure can simulate n
processes with one failure. This is worth looking at (it’s an excellent example
of the power of simulation arguments, and BG simulation is useful in many
other contexts) but we will stick with the original argument, which is simpler.
We will look at this again when we consider BG simulation in Chapter 27.

11.1 Agreement
Usual rules: agreement (all non-faulty processes decide the same value),
termination (all non-faulty processes eventually decide some value), valid-
ity (for each possible decision value, there an execution in which that value
is chosen). Validity can be tinkered with without affecting the proof much.
To keep things simple, we assume the only two decision values are 0 and
1.

11.2 Failures
A failure is an internal action after which all send operations are disabled.
The adversary is allowed one failure per execution. Effectively, this means
that any group of n − 1 processes must eventually decide without waiting
for the n-th, because it might have failed.

11.3 Steps
The FLP paper uses a notion of steps that is slightly different from the
send and receive actions of the asynchronous message-passing model we’ve
been using. Essentially a step consists of receiving zero or more messages
followed by doing a finite number of sends. To fit it into the model we’ve been
using, we’ll define a step as either a pair (p, m), where p receives message
m and performs zero or more sends in response, or (p, ⊥), where p receives
nothing and performs zero or more sends. We assume that the processes are
deterministic, so the messages sent (if any) are determined by p’s previous
state and the message received. Note that these steps do not correspond
2
Chapter 19.
CHAPTER 11. IMPOSSIBILITY OF ASYNCHRONOUS AGREEMENT82

precisely to delivery and send events or even pairs of delivery and send events,
because what message gets sent in response to a particular delivery may
change as the result of delivering some other message; but this won’t affect
the proof.
The fairness condition essentially says that if (p, m) or (p, ⊥) is continu-
ously enabled it eventually happens. Since messages are not lost, once (p, m)
is enabled in some configuration C, it is enabled in all successor configurations
until it occurs; similarly (p, ⊥) is always enabled. So to ensure fairness, we
have to ensure that any non-faulty process eventually performs any enabled
step.
Comment on notation: I like writing the new configuration reached by
applying a step e to C like this: Ce. The FLP paper uses e(C).

11.4 Bivalence and univalence

The core of the FLP argument is a strategy allowing the adversary (who
controls scheduling) to steer the execution away from any configuration in
which the processes reach agreement. The guidepost for this strategy is the
notion of bivalence, where a configuration C is bivalent if there exist traces
T0 and T1 starting from C that lead to configurations CT0 and CT1 where all
processes decide 0 and 1 respectively. A configuration that is not bivalent is
univalent, or more specifically 0-valent or 1-valent depending on whether
all executions starting in the configuration produce 0 or 1 as the decision
value. (Note that bivalence or univalence are the only possibilities because of
termination.) The important fact we will use about univalent configurations
is that any successor to an x-valent configuration is also x-valent.
It’s clear that any configuration where some process has decided is not
bivalent, so if the adversary can keep the protocol in a bivalent configuration
forever, it can prevent the processes from ever deciding. The adversary’s
strategy is to start in an initial bivalent configuration C0 (which we must
prove exists) and then choose only bivalent successor configurations (which
we must prove is possible). A complication is that if the adversary is only
allowed one failure, it must eventually allow any message in transit to a
non-faulty process to be received and any non-faulty process to send its
outgoing messages, so we have to show that the policy of avoiding univalent
configurations doesn’t cause problems here.
CHAPTER 11. IMPOSSIBILITY OF ASYNCHRONOUS AGREEMENT83

11.5 Existence of an initial bivalent configuration

We can specify an initial configuration by specifying the inputs to all processes.
If one of these initial configurations is bivalent, we are done. Otherwise,
let C and C 0 be two initial configurations that differ only in the input of
one process p; by assumption, both C and C 0 are univalent. Consider two
executions starting with C and C 0 in which process p is faulty; we can arrange
for these executions to be indistinguishable to all the other processes, so
both decide the same value x. It follows that both C and C 0 are x-valent.
But since any two initial configurations can be connected by some chain of
such indistinguishable configurations, we have that all initial configurations
are x-valent, which violations validity.

11.6 Staying in a bivalent configuration

Now start in a failure-free bivalent configuration C with some step e = (p, m)
or e = (p, ⊥) enabled in C. Let S be the set of configurations reachable
from C without doing e or failing any processes, and let e(S) be the set
of configurations of the form C 0 e where C 0 is in S. (Note that e is always
enabled in S, since once enabled the only way to get rid of it is to deliver
the message.) We want to show that e(S) contains a failure-free bivalent
configuration.
The proof is by contradiction: suppose that C 0 e is univalent for all C 0
in S. We will show first that there are C0 and C1 in S such that each Ci e
is i-valent. To do so, consider any pair of i-valent Ai reachable from C; if
Ai is in S, let Ci = Ai . If Ai is not in S, let Ci be the last configuration
before executing e on the path from C to Ai (Ci e is univalent in this case by
assumption).
So now we have C0 e and C1 e with Ci e i-valent in each case. We’ll now go
hunting for some configuration D in S and step e0 such that De is 0-valent
but De0 e is 1-valent (or vice versa); such a pair exists because S is connected
and so some step e0 crosses the boundary between the C 0 e = 0-valent and
the C 0 e = 1-valent regions.
By a case analysis on e and e0 we derive a contradiction:

1. Suppose e and e0 are steps of different processes p and p0 . Let both

steps go through in either order. Then Dee0 = De0 e, since in an
asynchronous system we can’t tell which process received its message
first. But De is 0-valent, which implies Dee0 is also 0-valent, which
contradicts De0 e being 1-valent.
CHAPTER 11. IMPOSSIBILITY OF ASYNCHRONOUS AGREEMENT84

2. Now suppose e and e0 are steps of the same process p. Again we let both
go through in either order. It is not the case now that Dee0 = De0 e,
since p knows which step happened first (and may have sent messages
telling the other processes). But now we consider some finite sequence
of steps e1 e2 . . . ek in which no message sent by p is delivered and some
process decides in Dee1 . . . ek (this occurs since the other processes
can’t distinguish Dee0 from the configuration in which p died in D, and
so have to decide without waiting for messages from p). This execution
fragment is indistinguishable to all processes except p from De0 ee1 . . . ek ,
so the deciding process decides the same value i in both executions.
But Dee0 is 0-valent and De0 e is 1-valent, giving a contradiction.

It follows that our assumption was false, and there is some reachable
bivalent configuration C 0 e.
Now to construct a fair execution that never decides, we start with a
bivalent configuration, choose the oldest enabled action and use the above
to make it happen while staying in a bivalent configuration, and repeat.

11.7 Generalization to other models

To apply the argument to another model, the main thing is to replace the
definition of a step and the resulting case analysis of 0-valent De0 e vs 1-valent
Dee0 to whatever steps are available in the other model. For example, in
asynchronous shared memory, if e and e0 are operations on different memory
locations, they commute (just like steps of different processes), and if they
are operations on the same location, either they commute (e.g., two reads)
or only one process can tell whether both happened (e.g., with a write and
a read, only the reader knows, and with two writes, only the first writer
knows). Killing the witness yields two indistinguishable configurations with
different valencies, a contradiction.
We are omitting a lot of details here. See [Lyn96, §12.2] for the real proof,
or Loui and Abu-Amara [LAA87] for the generalization to shared memory, or
Herlihy [Her91b] for similar arguments for a wide variety of shared-memory
primitives. We will see many of these latter arguments in Chapter 18.
Chapter 12

Paxos

The Paxos algorithm for consensus in a message-passing system was first

described by Lamport in 1990 in a tech report that was widely considered
to be a joke (see http://research.microsoft.com/users/lamport/pubs/
pubs.html#lamport-paxos for Lamport’s description of the history). The
algorithm was finally published in 1998 [Lam98], and after the algorithm
continued to be ignored, Lamport finally gave up and translated the results
into readable English [Lam01]. It is now understood to be one of the most
efficient practical algorithms for achieving consensus in a message-passing
system with failure detectors, mechanisms that allow processes to give up on
other stalled processes after some amount of time (which can’t be done in
a normal asynchronous system because giving up can be made to happen
immediately by the adversary).
We will describe the basic Paxos algorithm in §12.1. This is a one-shot
version of Paxos that solves a single agreement problem. The version that is
more typically used, called multi-Paxos, uses repeated executions of the
basic Paxos algorithm to implement a replicated state machine; we’ll describe
this in §12.3.
There are many more variants of Paxos in use. The WikiPedia article
on Paxos (http://en.wikipedia.org/wiki/Paxos_(computer_science))
gives a reasonably good survey of subsequent developments and applications.

12.1 The Paxos algorithm

The algorithm runs in a message-passing model with asynchrony and fewer
than n/2 crash failures (but not Byzantine failures, at least in the original
algorithm). As always, we want to get agreement, validity, and termination.

85
CHAPTER 12. PAXOS 86

The Paxos algorithm itself is mostly concerned with guaranteeing agree-

ment and validity, while allowing for the possibility of termination if there is a
long enough interval in which no process restarts the protocol. A noteworthy
feature of Paxos is that it is robust even to omission failures, in the sense
that lost messages can prevent termination, but if new messages start being
delivered again, the protocol can recover.
Processes are classified as proposers, accepters, and learners (a single
process may have all three roles). The idea is that a proposer attempts to
ratify a proposed decision value (from an arbitrary input set) by collecting
acceptances from a majority of the accepters, and this ratification is observed
by the learners. Agreement is enforced by guaranteeing that only one proposal
can get the votes of a majority of accepters, and validity follows from only
allowing input values to be proposed. The tricky part is ensuring that we
don’t get deadlock when there are more than two proposals or when some of
the processes fail. The intuition behind how this works is that any proposer
can effectively restart the protocol by issuing a new proposal (thus dealing
with lockups), and there is a procedure to release accepters from their old
votes if we can prove that the old votes were for a value that won’t be getting
a majority any time soon.
To organize this vote-release process, we attach a distinct proposal number
to each proposal. The safety properties of the algorithm don’t depend on
anything but the proposal numbers being distinct, but since higher numbers
override lower numbers, to make progress we’ll need them to increase over
time. The simplest way to do this in practice is to make the proposal number
be a timestamp with the proposer’s id appended to break ties. We could
also have the proposer poll the other processes for the most recent proposal
number they’ve seen and add 1 to it.
The revoting mechanism now works like this: before taking a vote, a
proposer tests the waters by sending a prepare(n) message to all accepters,
where n is the proposal number. An accepter responds to this with a promise
never to accept any proposal with a number less than n (so that old proposals
don’t suddenly get ratified) together with the highest-numbered proposal
that the accepter has accepted (so that the proposer can substitute this value
for its own, in case the previous value was in fact ratified). If the proposer
receives a response from a majority of the accepters, the proposer then does
a second phase of voting where it sends accept(n, v) to all accepters and wins
if receives a majority of votes.
So for each proposal, the algorithm proceeds as follows:

1. The proposer sends a message prepare(n) to all accepters. (Sending

CHAPTER 12. PAXOS 87

to only a majority of the accepters is enough, assuming they will all

respond.)

2. Each accepter compares n to the highest-numbered proposal for which it

has responded to a prepare message and the highest-numbered proposal
it has accepted. If n is greater than both, it responds with ack(n, v, nv ),
where v is the highest-numbered proposal it has accepted and nv is the
number of that proposal (or ⊥ and −∞ if there is no such proposal).
An optimization at this point is to allow the accepter to send back
nack(n, n0 ) where n0 is some higher number to let the proposer know
that it’s doomed and should back off and try again with a higher
proposal number. (This keeps a confused proposer who thinks it’s the
future from locking up the protocol until 2087.)

3. The proposer waits (possibly forever) to receive ack from a majority

of accepters. If any ack contained a value, it sets v to the most
recent (in proposal number ordering) value that it received. It then
sends accept(n, v) to all accepters (or just a majority). You should
think of accept as a demand (“Accept!”) rather than acquiescence (“I
accept”)—the accepters still need to choose whether to accept or not.

4. Upon receiving accept(n, v), an accepter accepts v unless it has already

received prepare(n0 ) for some n0 > n. If a majority of accepters accept
the value of a given proposal, that value becomes the decision value of
the protocol.

Implementing these rules require only that each accepter track na , the
highest number of any proposal for which it sent an ack, and hv, nv i, the last
proposal that it accepted. Pseudocode showing the behavior of proposer and
accepters in the core Paxos protocol is given in Algorithm 12.1.
Note that acceptance is a purely local phenomenon; additional messages
are needed to detect which if any proposals have been accepted by a majority
of accepters. Typically this involves a fourth round, where accepters send
accepted(n, v) to all learners.
There is no requirement that only a single proposal is sent out (indeed,
if proposers can fail we will need to send out more to jump-start the proto-
col). The protocol guarantees agreement and validity no matter how many
proposers there are and no matter how often they start.
CHAPTER 12. PAXOS 88

1 procedure Propose(n, v)
// Issue proposal number n with value v
// Assumes n is unique
2 send prepare(n, v) to all accepters
3 wait to receive ack(n, v 0 , nv0 ) from a majority of accepters
4 if some v 0 is not ⊥ then
5 v ← v 0 with maximum nv0
6 send accept(n, v) to all accepters
7 procedure accepter()
8 initially do
9 na ← −∞
10 v←⊥
11 nv ← −∞
12 upon receiving prepare(n) from p do
13 if n > max(na , nv ) then
// Respond to proposal
14 send ack(n, v, nv ) to p
15 na ← n

16 upon receiving accept(n, v 0 ) do

17 if n ≥ max(na , nv ) then
// Accept proposal
18 send accepted(n, v 0 ) to all learners
19 if n > nv then
// Update highest accepted proposal
20 hv, nv i ← hv 0 , ni

Algorithm 12.1: Paxos

CHAPTER 12. PAXOS 89

12.2 Informal analysis: how information flows be-

tween rounds
Call a round the collection of all messages labeled with some particular
proposal n. The structure of the algorithm simulates a sequential execution
in which higher-numbered rounds follow lower-numbered ones, even though
there is no guarantee that this is actually the case in a real execution.
When an accepter sends ack(n, v, nv ), it is telling the round-n proposer
the last value preceding round n that it accepted. The rule that an accepter
only acknowledges a proposal higher than any proposal it has previously
accepted prevents it from sending information “back in time”—the round
nv in an acknowledgment is always less than n. The rule that an accepter
doesn’t accept any proposal earlier than a round it has acknowledged means
that the value v in an ack(n, v, nv ) message never goes out of date—there is
no possibility that an accepter might retroactively accept some later value in
round n0 with nv < n0 < n. So the ack message values tell a consistent story
about the history of the protocol, even if the rounds execute out of order.
The second trick is to use overlapping majorities to make sure that any
value that is accepted is not lost. If the only way to decide on a value in round
n is to get a majority of accepters to accept it, and the only way to make
progress in round n0 is to get acknowledgments from a majority of accepters,
these two majorities overlap. So in particular the overlapping process reports
the round-n proposal value to the proposer in round n0 , and we can show
by induction on n0 that this round-n proposal value becomes the proposal
value in all subsequent rounds that proceed past the acknowledgment stage.
So even though it may not be possible to detect that a decision has been
reached in round n (say, because some of the accepters in the accepting
majority die without telling anybody what they did), no later round will be
able to choose a different value. This ultimately guarantees agreement.

12.2.1 Example execution

For Paxos to work well, proposal numbers should increase over time. But
there is no requirement that proposal numbers are increasing or even that
proposals with different proposal numbers don’t overlap. When thinking
about Paxos, it is easy to make the mistake of ignore cases where proposals
are processed concurrently or out of order. In Figure 12.1, we give an example
of an execution with three proposals running concurrently.
CHAPTER 12. PAXOS 90

p1 p2 p3 a1 a2 a3
prepare(3)
prepare(2)
prepare(1)
ack(3, ⊥, 0)
ack(1, ⊥, 0)
ack(1, ⊥, 0)
accept(1, 1)
accepted(1, 1)
nack(1, 3)
ack(2, 1, 1)
ack(2, ⊥, 0)
accept(2, 1)
nack(2, 3)
accepted(2, 1)
ack(3, 1, 2)
accept(3, 1)
accepted(3, 1)
accepted(3, 1)

Figure 12.1: Example execution of Paxos. There are three proposers p1 , p2 ,

and p3 and three accepters a1 , a2 , and a3 . Proposer p1 ’s proposed value 1 is
not accepted by a majority of processes in round 1, but it is picked up by
proposer p2 in round 2, and is eventually adopted and accepted in round 3.
CHAPTER 12. PAXOS 91

12.2.2 Safety properties

We now present a more formal analysis of the Paxos protocol. We consider
only the safety properties of the protocol, corresponding to validity and
agreement. Without additional assumptions, Paxos does not guarantee
termination.
Call a value chosen if it is accepted by a majority of accepters. The
safety properties of Paxos are:

• No value is chosen unless it is first proposed. (This gives validity.)

• No two distinct values are both chosen. (This gives agreement.)

The first property is immediate from examination of the algorithm:

every value propagated through the algorithm is ultimately a copy of some
proposer’s original input. We can formalize this observation by checking that,
for any set of values S, the property that all values contained in messages or
processes’ internal state are in S is an invariant.
For the second property, we’ll show by induction on proposal number
that a value v chosen with proposal number n is the value chosen by any
proposer pn0 with proposal number n0 . There are two things that make this
true:

1. Any ack(n0 , v 0 , nv0 ) message received by pn0 has nv0 < n0 . Proof: Im-
mediate from the code.

2. If a majority of accepters accept a proposal with number n at some

point during the execution, and pn0 receives ack(n0 , −, −) messages from
a majority of accepters, then pn0 receives at least one ack(n0 , v 0 , nv0 )
message with n0 ≥ n. Proof: Let S be the set of processes that issue
accepted(n, v) and let T be the set of processes that send ack(n0 , −, −)
to p0 . Because S and T are both majorities, there is at least one
accepter a in S ∩ T . Suppose pn0 receives ack(n, v 00 , n00 ) from a. If
n00 < n, then at the time a sends its ack(n, v 00 , n00 ) message, it has not
yet accepted a proposal with number n. But then when it does receive
accept(n, v), it rejects it. This contradicts a ∈ S.

These two properties together imply that pn0 receives at least one
ack(n, v 00 , n00 ) with n ≤ n00 < n0 and no such messages with n00 < n. So
the maximum proposal number it sees is n00 where n ≤ n00 < n. By the
induction hypothesis, the corresponding value is v. It follows that pn0 also
chooses v.
CHAPTER 12. PAXOS 92

12.2.3 Learning the results

Somebody has to find out that a majority accepted a proposal in order to get
a decision value out. The usual way to do this is to have a fourth round of
messages where the accepters send accepted(v, n) to some designated learners.
These are often the processes that need to implement whatever decision was
made by the agreement protocol, but in principle could be any processes
that care about the outcome.

12.2.4 Liveness properties

We’d like the protocol to terminate eventually. Suppose there is a single
proposer, and that it survives long enough to collect a majority of acks and to
send out accepts to a majority of the accepters. If everybody else cooperates,
we get termination in 4 message delays, including the time for the learners
to detect acceptance.
If there are multiple proposers, then they can step on each other. For
example, it’s enough to have two carefully-synchronized proposers alternate
sending out prepare messages to prevent any accepter from every accepting
(since an accepter promises not to accept accept(n, v) once it has responded
to prepare(n + 1)). The solution is to ensure that there is eventually some
interval during which there is exactly one proposer who doesn’t fail. One way
to do this is to use exponential random backoff (as popularized by Ethernet):
when a proposer decides it’s not going to win a round (e.g., by receiving a
nack or by waiting long enough to realize it won’t be getting any more acks
soon), it picks some increasingly large random delay before starting a new
round. Unless something strange is going on, new rounds will eventually
start far enough apart in time that one will get done without interference.
A more abstract solution is to assume some sort of weak leader election
mechanism, which tells each accepter who the “legitimate” proposer is at
each time. The accepters then discard messages from illegitimate proposers,
which prevents conflict at the cost of possibly preventing progress. Progress
is however obtained if the mechanism eventually reaches a state where a
majority of the accepters bow to the same non-faulty proposer long enough
for the proposal to go through.
Such a weak leader election method is an example of a more general class
of mechanisms known as failure detectors, in which each process gets hints
about what other processes are faulty that eventually converge to reality.
The weak-leader-election failure detector needed for Paxos is called the Ω
failure detector. There are other still weaker failure detectors that can also
CHAPTER 12. PAXOS 93

be used to solve consensus. We will discuss failure detectors in detail in

Chapter 13.

12.3 Replicated state machines and multi-Paxos

The most common practical use of Paxos is to implement a replicated
state machine [Lam78]. The idea is to maintain many copies of some data
structure, each on a separate machine, and guarantee that each copy (or
replica) stays in sync with all the others as new operations are applied
to them. This requires some mechanism to ensure that all the different
replicas apply the same sequence of operations, or in other words that the
machines that hold the replicas solve a sequence of agreement problems to
agree on these operations. The payoff is that the state of the data structure
survives the failure of some of the machines, without having to copy the
entire structure every time it changes.
Making all copies consistent requires solving a new version of agreement
every time we want to add another operation. Paxos works well for this
because we can have to proposer simply issue a new proposal without taking
into account any lower-numbered values, assuming that it has verified that
lower-numbered values have in fact been accepted. The round-number
mechanism means that all of the accepters will switch to working on the new
proposal without any modifications to their code.
Typically for this application, we’ll have a single active proposer that is
responsible for serializing any incoming operations to the replicated state
machine. If the proposer doesn’t change very often, a further optimization
allows skipping the prepare and ack messages in between agreement protocols
for consecutive operations. This reduces the time to certify each operation
to a single round-trip for the accept and accepted messages, which is about
the best we can reasonably hope for.
One detail is that to make this work, we need to distinguish between
consecutive proposals by the same proposer, and “new” proposals that change
the proposer in addition to reaching agreement on some value. This is done
by splitting the proposal number into a major and minor number, with
proposals ordered lexicographically. A proposer that wins hx, 0i is allowed to
make further proposals numbered hx, 1i , hx, 2i , etc. But a different proposer
will need to increment x.
Lamport calls this optimization Paxos in [Lam01]; other authors have
called it multi-Paxos to distinguish it from the basic Paxos algorithm.
Chapter 13

Failure detectors

Failure detectors were proposed by Chandra and Toueg [CT96] as a mech-

anism for solving consensus in an asynchronous message-passing system with
crash failures by distinguishing between slow processes and dead processes.
The basic idea is that each process has attached to it a failure detector module
that continuously outputs an estimate of which processes in the system have
failed. The output need not be correct; indeed, the main contribution of
Chandra and Toueg’s paper (and a companion paper by Chandra, Hadzilacos,
and Toueg [CHT96]) is characterizing just how bogus the output of a failure
detector can be and still be useful.
We will mostly follow Chandra and Toueg in these notes; see the paper
for the full technical details.
To emphasize that the output of a failure detector is merely a hint at
the actual state of the world, a failure detector (or the process it’s attached
to) is said to suspect a process at time t if it outputs failed at that time.
Failure detectors can then be classified based on when their suspicions are
correct.
We use the usual asynchronous message-passing model, and in particular
assume that non-faulty processes execute infinitely often, get all their mes-
sages delivered, etc. From time to time we will need to talk about time, and
unless we are clearly talking about real time this just means any steadily
increasing count (e.g., of total events), and will be used only to describe the
ordering of events.

94
CHAPTER 13. FAILURE DETECTORS 95

13.1 How to build a failure detector

Failure detectors are only interesting if you can actually build them. In
a fully asynchronous system, you can’t (this follows from the FLP result
and the existence of failure-detector-based consensus protocols). But with
timeouts, it’s not hard: have each process ping each other process from
time to time, and suspect the other process if it doesn’t respond to the ping
within twice the maximum round-trip time for any previous ping. Assuming
that ping packets are never lost and there is an (unknown) upper bound on
message delay, this gives what is known as an eventually perfect failure
detector: once the max round-trip times rise enough and enough time has
elapsed for the live processes to give up on the dead ones, all and only dead
processes are suspected.

13.2 Classification of failure detectors

Chandra and Toueg define eight classes of failure detectors, based on when
they suspect faulty processes and non-faulty processes. Suspicion of faulty
processes comes under the heading of completeness; of non-faulty processes,
accuracy.

13.2.1 Degrees of completeness

Strong completeness Every faulty process is eventually permanently sus-
pected by every non-faulty process.

Weak completeness Every faulty process is eventually permanently sus-

pected by some non-faulty process.

There are two temporal logic operators embedded in these statements:

“eventually permanently” means that there is some time t0 such that for
all times t ≥ t0 , the process is suspected. Note that completeness says
nothing about suspecting non-faulty processes: a paranoid failure detector
that permanently suspects everybody has strong completeness.

13.2.2 Degrees of accuracy

These describe what happens with non-faulty processes, and with faulty
processes that haven’t crashed yet.

Strong accuracy No process is suspected (by anybody) before it crashes.

CHAPTER 13. FAILURE DETECTORS 96

Weak accuracy Some non-faulty process is never suspected.

Eventual strong accuracy After some initial period of confusion, no pro-

cess is suspected before it crashes. This can be simplified to say that
no non-faulty process is suspected after some time, since we can take
end of the initial period of chaos as the time at which the last crash
occurs.

Eventual weak accuracy After some initial period of confusion, some

non-faulty process is never suspected.

Note that “strong” and “weak” mean different things for accuracy vs
completeness: for accuracy, we are quantifying over suspects, and for com-
pleteness, we are quantifying over suspectors. Even a weakly-accurate failure
detector guarantees that all processes trust the one visibly good process.

13.2.3 Boosting completeness

It turns out that any weakly-complete failure detector can be boosted to give
strong completeness. Recall that the difference between weak completeness
and strong completeness is that with weak completeness, somebody suspects
a dead process, while with strong completeness, everybody suspects it. So to
boost completeness we need to spread the suspicion around a bit. On the
other hand, we don’t want to break accuracy in the process, so there needs to
be some way to undo a premature rumor of somebody’s death. The simplest
way to do this is to let the alleged corpse speak for itself: I will suspect you
from the moment somebody else reports you dead until the moment you tell
me otherwise.
Pseudocode is given in Algorithm 13.1.

1 initially do
2 suspects ← ∅
3 while true do
4 Let S be the set of all processes my weak detector suspects.
5 Send S to all processes.
6 upon receiving S from q do
7 suspects ← (suspects ∪ p) \ {q}
Algorithm 13.1: Boosting completeness
CHAPTER 13. FAILURE DETECTORS 97

It’s not hard to see that this boosts completeness: if p crashes, somebody’s
weak detector eventually suspects it, this process tells everybody else, and p
never contradicts it. So eventually everybody suspects p.
What is slightly trickier is showing that it preserves accuracy. The
essential idea is this: if there is some good-guy process p that everybody trusts
forever (as in weak accuracy), then nobody ever reports p as suspect—this
also covers strong accuracy since the only difference is that now every non-
faulty process falls into this category. For eventual weak accuracy, wait
for everybody to stop suspecting p, wait for every message ratting out p
to be delivered, and then wait for p to send a message to everybody. Now
everybody trusts p, and nobody every suspects p again. Eventual strong
accuracy is again similar.
This will justify ignoring the weakly-complete classes.

13.2.4 Failure detector classes

Two degrees of completeness times four degrees of accuracy gives eight classes
of failure detectors, each of which gets its own name. But since we can boost
weak completeness to strong completeness, we can use this as an excuse to
consider only the strongly-complete classes.

P (perfect) Strongly complete and strongly accurate: non-faulty processes

are never suspected; faulty processes are eventually suspected by ev-
erybody. Easily achieved in synchronous systems.

S (strong) Strongly complete and weakly accurate. The name is misleading

if we’ve already forgotten about weak completeness, but the correspond-
ing W (weak) class is only weakly complete and weakly accurate, so
it’s the strong completeness that the S is referring to.

♦P (eventually perfect) Strongly complete and eventually strongly accu-

rate.

♦S (eventually strong) Strongly complete and eventually weakly accu-

rate.

Jumping to the punch line: P can simulate any of the others, S and
♦P can both simulate ♦S but can’t simulate P or each other, and ♦S can’t
simulate any of the others (See Figure 13.1—we’ll prove all of this later.)
Thus ♦S is the weakest class of failure detectors in this list. However, ♦S is
strong enough to solve consensus, and in fact any failure detector (whatever
CHAPTER 13. FAILURE DETECTORS 98

S ♦P

♦S

Figure 13.1: Partial order of failure detector classes. Higher classes can
simulate lower classes but not vice versa.

its properties) that can solve consensus is strong enough to simulate ♦S

(this is the result in the Chandra-Hadzilacos-Toueg paper [CHT96])—this
makes ♦S the “weakest failure detector for solving consensus” as advertised.
Continuing our tour through Chandra and Toueg [CT96], we’ll show the
simulation results and that ♦S can solve consensus, but we’ll skip the rather
involved proof of ♦S’s special role from Chandra-Hadzilacos-Toueg.

13.3 Consensus with S

With the strong failure detector S, we can solve consensus for any number
of failures.
In this model, the failure detectors as applied to most processes are
completely useless. However, there is some non-faulty process c that nobody
every suspects, and this is enough to solve consensus with as many as n − 1
failures.
The basic idea of the protocol: There are three phases. In the first phase,
the processes gossip about input values for n − 1 asynchronous rounds. In
the second, they exchange all the values they’ve seen and prune out any
that are not universally known. In the third, each process decides on the
lowest-id input that hasn’t been pruned (minimum input also works since at
this point everybody has the same view of the inputs).
Pseudocode is given in Algorithm 13.2
In Phase 1, each process p maintains two partial functions Vp and δp ,
where Vp lists all the input values hq, vq i that p has ever seen and δp lists
only those input values seen in the most recent of n − 1 asynchronous rounds.
Both Vp and δp are initialized to {hp, vp i}. In round i, p sends (i, δp ) to all
processes. It then collects hi, δq i from each q that it doesn’t suspect and sets
δp to q δq \ Vp (where q ranges over the processes from which p received a
S
CHAPTER 13. FAILURE DETECTORS 99

1 Vp ← {hp, vp i}
2 δp ← {hp, vp i}
// Phase 1
3 for i ← 1 to n − 1 do
4 Send hi, δp i to all processes.
5 Wait to receive hi, δq i from all q I do not suspect.
S 
6 δp ← q δq \ Vp
S 
7 Vp ← q δq ∪ Vp
// Phase 2
8 Send hn, Vp i to all processes.
9 Wait to receive hn, Vq i from all q I do not suspect.
T 
10 Vp ← q Vq ∩ Vp
// Phase 3
11 return some input from Vp chosen via a consistent rule.
Algorithm 13.2: Consensus with a strong failure detector

message in round i) and sets Vp to Vp ∪ δp . In the next round, it repeats the

process. Note that each pair hq, vq i is only sent by a particular process p the
first round after p learns it: so any value that is still kicking around in round
n − 1 had to go through n − 1 processes.
In Phase 2, each process p sends hn, Vp i, waits to receive hn, Vq i from
every process it does not suspect, and sets Vp to the intersection of Vp and
all received Vq . At the end of this phase all Vp values will in fact be equal,
as we will show.
In Phase 3, everybody picks some input from their Vp vector according
to a consistent rule.

13.3.1 Proof of correctness

Let c be a non-faulty process that nobody every suspects.
The first observation is that the protocol satisfies validity, since every
Vp contains vc after round 1 and each Vp can only contain input values by
examination of the protocol. Whatever it may do to the other values, taking
intersections in Phase 2 still leaves vc , so all processes pick some input value
from a nonempty list in Phase 3.
To get termination we have to prove that nobody ever waits forever
for a message it wants; this basically comes down to showing that the first
CHAPTER 13. FAILURE DETECTORS 100

non-faulty process that gets stuck eventually is informed by the S-detector

that the process it is waiting for is dead.
For agreement, we must show that in Phase 3, every Vp is equal; in
particular, we’ll show that every Vp = Vc . First it is necessary to show that
at the end of Phase 1, Vc ⊆ Vp for all p. This is done by considering two
cases:

1. If hq, vq i ∈ Vc and c learns hq, vq i before round n − 1, then c sends

hq, vq i to p no later than round n − 1, p waits for it (since nobody ever
suspects c), and adds it to Vp .

2. If hq, vq i ∈ Vc and c learns hq, vq i only in round n − 1, then hq, vq i was

previously sent through n − 1 other processes, i.e., all of them. Each
process p 6= c thus added hq, vq i to Vp before sending it and again
hq, vq i is in Vp .

(The missing case where hq, vq i isn’t in Vc we don’t care about.)

But now Phase 2 knocks out any extra elements in Vp , since Vp gets set to
Vp ∩ Vc ∩ (some other Vq ’s that are supersets of Vc ). It follows that, at the
end of Phase 2, Vp = Vc for all p. Finally, in Phase 3, everybody applies the
same selection rule to these identical sets and we get agreement.

13.4 Consensus with ♦S and f < n/2

The consensus protocol for S depends on some process c never being suspected;
if c is suspected during the entire (finite) execution of the protocol—as can
happen with ♦S—then it is possible that no process will wait to hear from
c (or anybody else) and the processes will all decide their own inputs. So
to solve consensus with ♦S we will need to assume fewer than n/2 failures,
allowing any process to wait to hear from a majority no matter what lies its
failure detector is telling it.
The resulting protocol, known as the Chandra-Toueg consensus pro-
tocol, is structurally similar to the consensus protocol in Paxos.1 The
difference is that instead of proposers blindly showing up, the protocol is
divided into rounds with a rotating coordinator pi in each round r with
r = i (mod n). The termination proof is based on showing that in any round
where the coordinator is not faulty and nobody suspects it, the protocol
finishes.
1
See Chapter 12.
CHAPTER 13. FAILURE DETECTORS 101

The consensus protocol uses as a subroutine a protocol for reliable

broadcast, which guarantees that any message that is sent is either received
by no non-faulty processes or exactly once by all non-faulty processes. Pseu-
docode for reliable broadcast is given as Algorithm 13.3. It’s easy to see that
if a process p is non-faulty and receives m, then the fact that p is non-faulty
means that is successfully sends m to everybody else, and that the other
non-faulty processes also receive the message at least once and deliver it.

1 procedure broadcast(m)
2 send m to all processes.
3 upon receiving m do
4 if I haven’t seen m before then
5 send m to all processes
6 deliver m to myself

Algorithm 13.3: Reliable broadcast

Here’s a sketch of the actual consensus protocol:

• Each process keeps track of a preference (initially its own input) and a
timestamp, the round number in which it last updated its preference.

• The processes go through a sequence of asynchronous rounds, each

divided into four phases:

1. All processes send (round, preference, timestamp) to the coordi-

nator for the round.
2. The coordinator waits to hear from a majority of the processes
(possibly including itself). The coordinator sets its own preference
to some preference with the largest timestamp of those it receives
and sends (round, preference) to all processes.
3. Each process waits for the new proposal from the coordinator or
for the failure detector to suspect the coordinator. If it receives
a new preference, it adopts it as its own, sets timestamp to
the current round, and sends (round, ack) to the coordinator.
Otherwise, it sends (round, nack) to the coordinator.
4. The coordinator waits to receive ack or nack from a majority of
processes. If it receives ack from a majority, it announces the
current preference as the protocol decision value using reliable
broadcast.
CHAPTER 13. FAILURE DETECTORS 102

• Any process that receives a value in a reliable broadcast decides on it

immediately.

Pseudocode is in Algorithm 19.

1 preference ← input
2 timestamp ← 0
3 for round ← 1 . . . ∞ do
4 Send hround, preference, timestampi to coordinator
5 if I am the coordinator then
6 Wait to receive hround, preference, timestampi from majority
of processes.
7 Set preference to value with largest timestamp.
8 Send hround, preferencei to all processes.
9 Wait to receive round, preference0 from coordinator or to suspect
coordinator.
10 if I received round, preference0 then
11 preference ← preference0
12 timestamp ← round
13 Send ack(round) to coordinator.
14 else
15 Send nack(round) to coordinator.
16 if I am the coordinator then
17 Wait to receive ack(round) or nack(round) from a majority of
processes.
18 if I received no nack(round) messages then
19 Broadcast preference using reliable broadcast.

13.4.1 Proof of correctness

For validity, observe that the decision value is an estimate and all estimates
start out as inputs.
For termination, observe that no process gets stuck in Phase 1, 2, or 4,
because either it isn’t waiting or it is waiting for a majority of non-faulty
processes who all sent messages unless they have already decided (this is
why we need the nacks in Phase 3). The loophole here is that processes that
decide stop participating in the protocol; but because any non-faulty process
CHAPTER 13. FAILURE DETECTORS 103

retransmits the decision value in the reliable broadcast, if a process is waiting

for a response from a non-faulty process that already terminated, eventually
it will get the reliable broadcast instead and terminate itself. In Phase 3,
a process might get stuck waiting for a dead coordinator, but the strong
completeness of ♦S means that it suspects the dead coordinator eventually
and escapes. So at worst we do finitely many rounds.
Now suppose that after some time t there is a process c that is never
suspected by any process. Then in the next round in which c is the coordi-
nator, in Phase 3 all surviving processes wait for c and respond with ack, c
decides on the current estimate, and triggers the reliable broadcast protocol
to ensure everybody else decides on the same value. Since reliable broadcast
guarantees that everybody receives the message, everybody decides this value
or some value previously broadcast—but in either case everybody decides.
Agreement is the tricky part. It’s possible that two coordinators both
initiate a reliable broadcast and some processes choose the value from the first
and some the value from the second. But in this case the first coordinator
collected acks from a majority of processes in some round r, and all subsequent
coordinators collected estimates from an overlapping majority of processes in
some round r0 > r. By applying the same induction argument as for Paxos,
we get that all subsequent coordinators choose the same estimate as the first
coordinator, and so we get agreement.

13.5 f < n/2 is still required even with ♦P

We can show that with a majority of failures, we’re in trouble with just ♦P
(and thus with ♦S, which is trivially simulated by ♦P ). The reason is that
♦P can lie to us for some long initial interval of the protocol, and consensus
is required to terminate eventually despite these lies. So the usual partition
argument works: start half of the processes with input 0, half with 1, and
run both halves independently with ♦P suspecting the other half until the
processes in both halves decide on their common inputs. We can now make
♦P happy by letting it stop suspecting the processes, but it’s too late.

13.6 Relationships among the classes

It’s easy to see that P simulates S and ♦P simulates ♦S without modification.
It’s also immediate that P simulates ♦P and S simulates ♦S (make “even-
tually” be “now”), which gives a diamond-shaped lattice structure between
the classes. What is trickier is to show that this structure doesn’t collapse:
CHAPTER 13. FAILURE DETECTORS 104

♦P can’t simulate S, S can’t simulate ♦P , and ♦S can’t simulate any of the

other classes.
First let’s observe that ♦P can’t simulate S: if it could, we would get a
consensus protocol for f ≥ n/2 failures, which we can’t do. It follows that
♦P also can’t simulate P (because P can simulate S).
To show that S can’t simulate ♦P , choose some non-faulty victim process
v and consider an execution in which S periodically suspects v (which it
is allowed to do as long as there is some other non-faulty process it never
suspects). If the ♦P -simulator ever responds to this by refusing to suspect v,
there is an execution in which v really is dead, and the simulator violates
strong completeness. But if not, we violate eventual strong accuracy. Note
that this also implies S can’t simulate P , since P can simulate ♦P . It also
shows that ♦S can’t simulate either of ♦P or P .
We are left with showing ♦S can’t simulate S. Consider a system where
p’s ♦S detector suspects q but not r from the start of the execution, and
similarly r’s ♦S detector also suspects q but not p. Run p and r in isolation
until they give up and decide that q is in fact dead (which they must do
eventually by strong completeness, since this run is indistinguishable from
one in which q is faulty). Then wake up q and crash p and r. Since q is the
only non-faulty process, we’ve violated weak accuracy.
Chandra and Toueg [CT96] give as an example of a natural problem
that can be solved only with P the problem of terminating reliable
broadcast, in which a single leader process attempts to send a message and
all other processes eventually agree on the message if the leader is non-faulty
but must terminate after finite time with a default no message return value
if the leader is faulty.2 The process is solvable using P by just having each
process either wait for the message or for P to suspect the leader, which can
only occur if the leader does in fact crash. If the leader is dead, the processes
must eventually decide on no message; this separates P from ♦S and ♦P
since we can then wake up the leader and let it send its message. But it also
separates P from S, since we can have the S-detector only be accurate for
non-leaders. For other similar problems see the paper.

2
This is a slight weakening of the problem, which however still separates P from the
other classes. For the real problem see Chandra and Toueg [CT96].
Chapter 14

Quorum systems

Last updated 2014. Some material may be out of date.

14.1 Basics
In the past few chapters, we’ve seen many protocols that depend on the
fact that if I talk to more than n/2 processes and you talk to more than
n/2 processes, the two groups overlap. This is a special case of a quorum
system, a family of subsets of the set of processes with the property that
any two subsets in the family overlap. By choosing an appropriate family, we
may be able to achieve lower load on each system member, higher availability,
defense against Byzantine faults, etc.
The exciting thing from a theoretical perspective is that these turn
a systems problem into a combinatorial problem: this means we can ask
combinatorialists how to solve it.

14.2 Simple quorum systems

• Majority and weighted majorities

• Specialized read/write systems where write quorum is a column and

read quorum a row of some grid.

• Dynamic quorum systems: get more than half of the most recent copy.

• Crumbling walls [PW97b, PW97a]: optimal small-quorum system for

good choice of wall sizes.

105
CHAPTER 14. QUORUM SYSTEMS 106

14.3 Goals
• Minimize quorum size.

• Minimize load, defined as the minimum over all access strategies

(probability distributions on quorums) of the maximum over all servers
of probability it gets hit.

• Maximize capacity, defined as the maximum number of quorum ac-

cesses per time unit in the limit if each quorum access ties up a quorum
member for 1 time unit (but we are allowed to stagger a quorum access
over multiple time units).

• Maximize fault-tolerance: minimum number of server failures that

blocks all quorums. Note that for standard quorum systems this is
directly opposed to minimizing quorum size, since killing the smallest
quorum stops us dead.

• Minimize failure probability = probability that every quorum con-

tains at least one bad server, assuming each server fails with independent
probability.

Naor and Wool [NW98] describe trade-offs between these goals (some of
these were previously known, see the paper for citations):

• capacity = 1/load; this is obtained by selecting the quorums indepen-

dently at random according to the load-minimizing distribution. In
particular this means we can forget about capacity and just concentrate
on minimizing load.

• load ≥ max(c/n, 1/c) where c is the minimum quorum size. The first
case is obvious: if every access hits c nodes, spreading them out as
evenly as possible still hits each node c/n of the time. The second is
trickier: Naor and Wool prove it using LP duality, but the argument
essentially says that if we have some quorum Q of size c, then since
every other quorum Q0 intersects Q in at least one place, we can show
that every Q0 adds at least 1 unit of load in total to the c members of
Q. So if we pick a random quorum Q0 , the average load added to all of
Q is at least 1, so the average load added to some particular element
of Q is at least 1/|Q| = 1/c. Combining the two cases, we can’t hope
√
to get load better than 1/ n, and to get this load we need quorums of
√
size at least n.
CHAPTER 14. QUORUM SYSTEMS 107

Figure 14.1: Figure 2 from [NW98]. Solid lines are G(3); dashed lines are
G∗ (3).

• failure probability is at least p when p > 1/2 (and optimal system is to

just pick a single leader in this case), failure probability can be made
exponentially small in size of smallest quorum when p < 1/2 (with
many quorums). These results are due to Peleg and Wool [PW95].

14.4 Paths system

This is an optimal-load system from Naor and Wool [NW98] with exponen-
tially low failure probability, based on percolation theory.
The basic idea is to build a d × d mesh-like graph where a quorum consists
of the union of a top-to-bottom path (TB path) and a left-to-right path (LR
√ √
path); this gives quorum size O( n) and load O(1/ n). Note that the TB
and LR paths are not necessarily direct: they may wander around for a while
in order to get where they are going, especially if there are a lot of failures
√
to avoid. But the smallest quorums will have size 2d + 1 = O( n).
The actual mesh is a little more complicated. Figure 14.1 reproduces the
picture of the d = 3 case from the Naor and Wool paper.
Each server corresponds to a pair of intersecting edges, one from the
CHAPTER 14. QUORUM SYSTEMS 108

G(d) grid and one from the G∗ (d) grid (the star indicates that G∗ (d) is the
dual graph1 of G(d). A quorum consists of a set of servers that produce an
LR path in G(d) and a TB path in G∗ (d). Quorums intersect, because any
LR path in G(d) must cross some TB path in G∗ (d) at some server (in fact,
each pair of quorums intersects in at least two places). The total number of
√
elements n is (d + 1)2 and the minimum size of a quorum is 2d + 1 = Θ( n).
The symmetry of the mesh gives that there exists a LR path in the
mesh if and only if there does not exist a TB path in its complement, the
graph that has an edge only if the mesh doesn’t. For a mesh with failure
probability p < 1/2, the complement is a mesh with failure probability
q = 1 − p > 1/2. Using results in percolation theory, it can be shown that for
failure probability q > 1/2, the probability that there exists a left-to-right
path is exponentially small in d (formally, for each p there is a constant φ(p)
such that Pr[∃LR path] ≤ exp(−φ(p)d)). We then have

Pr[∃(live quorum)] = Pr[∃(TB path) ∧ ∃(LR path)]

= Pr[¬∃(LR path in complement) ∨ ¬∃(TB path in complement)]
≤ Pr[¬∃(LR path in complement)] + Pr[¬∃(TB path in complement)]
≤ 2 exp(−φ(1 − p)d)
√
= 2 exp(−Θ( n)).

So the failure probability of this system is exponentially small for any fixed
p < 1/2.
See the paper [NW98] for more details.

14.5 Byzantine quorum systems

Standard quorum systems are great when you only have crash failures, but
with Byzantine failures you have to worry about finding a quorum that
includes a Byzantine serve who lies about the data. For this purpose you
need something stronger. Following Malkhi and Reiter [MR98] and Malkhi et
al. [MRWW01], one can define:

• A b-disseminating quorum system guarantees |Q1 ∩ Q2 | ≥ b +

1 for all quorums Q1 and Q2 . This guarantees that if I update a
1
See http://en.wikipedia.org/wiki/Dual_graph; the basic idea is that the dual of a
graph G embedded in the plane has a vertex for each region of G, and an edge connecting
each pair of vertices corresponding to adjacent regions, where a region is a subset of the
plane that is bounded by edges of G.
CHAPTER 14. QUORUM SYSTEMS 109

quorum Q1 and you update a quorum Q2 , and there are at most

b Byzantine processes, then there is some non-Byzantine process in
both our quorums. Mostly useful if data is “self-verifying,” that is,
signed with digital signatures that the Byzantine processes can’t forge.
Otherwise, I can’t tell which of the allegedly most recent data values is
the right one since the Byzantine processes lie.

• A b-masking quorum system guarantees |Q1 ∩ Q2 | ≥ 2b + 1 for

all quorums Q1 and Q2 . (In other words, it’s the same as a 2b-
disseminating quorum system.) This allows me to defeat the Byzantine
processes through voting: given 2b + 1 overlapping servers, if I want
the most recent value of the data I take the one with the most recent
timestamp that appears on at least b + 1 servers, which the Byzantine
guys can’t fake.

An additional requirement in both cases is that for any set of servers B

with |B| ≤ b, there is some quorum Q such that Q ∩ B = ∅. This prevents
the Byzantine processes from stopping the system by simply refusing to
participate.
Note: these definitions are based on the assumption that there is some
fixed bound on the number of Byzantine processes. Malkhi and Reiter [MR98]
give more complicated definitions for the case where one has an arbitrary
family {B} of potential Byzantine sets. The definitions above are actually
simplified versions from [MRWW01].
The simplest way to build a b-disseminating quorum system is to use
supermajorities of size at least (n + b + 1)/2; the overlap between any two
such supermajorities is at least (n + b + 1) − n = b + 1. This gives a load of
1
substantially more
p than 2 . There are better constructions that knock the
load down to Θ( b/n); see [MRWW01].
For more on this topic in general, see the survey by by Merideth and
Reiter [MR10].

14.6 Probabilistic quorum systems

The problem with all standard (or strict) quorum systems is that we need big
quorums to get high fault tolerance, since the adversary can always stop us
by knocking out our smallest quorum. A probabilistic quorum system or
more specifically an -intersecting quorum system [MRWW01] improves
the fault-tolerance by relaxing the requirements. For such a system we have
not only a set system Q, but also a probability distribution w supplied by
CHAPTER 14. QUORUM SYSTEMS 110

the quorum system designer, with the property that Pr[Q1 ∩ Q2 = ∅] ≤ 

when Q1 and Q2 are chosen independently according to their weights.

14.6.1 Example
√
Let a quorum be any set of size k n for some k and let all quorums be
chosen uniformly at random. Pick some quorum Q1 ; what is the probability
that a random Q2 does not intersect Q1 ? Imagine we choose the elements
of Q2 one at a time. The chance that the first element x1 of Q2 misses Q1
√ √
is exactly (n − k n)/n = 1 − k/ n, and conditioning on x1 through xi−1
√
missing Q1 the probability that xi also misses it is (n − k n − i + 1)/(n −
√ √
i + 1) ≤ (n − k n)/n = 1 − k/√ n. So taking the √product over all i gives
√ √
Pr[all miss Q1 ] ≤ (1 − k/ n)k n ≤ exp(−k n)k/ n) = exp(−k 2 ). So by
setting k = Θ(ln 1/), we can get our desired -intersecting system.

14.6.2 Performance
Failure probabilities, if naively defined, can be made arbitrarily small: add
low-probability singleton quorums that are hardly ever picked unless massive
failures occur. But the resulting system is still -intersecting.
One way to look at this is that it points out a flaw in the -intersecting
definition: -intersecting quorums may cease to be -intersecting conditioned
on a particular failure pattern (e.g., when all the non-singleton quorums are
knocked out by massive failures). But Malkhi et al. [MRWW01] address the
problem in a different way, by considering only survival of high quality
quorums, where a particular quorum Q is δ-high-quality if Pr[Q1 ∩ Q2 =
√
∅|Q1 = Q] ≤ δ and high quality if it’s -high-quality. It’s not hard to show
that a random quorum is δ-high-quality with probability at least /δ, so a
high quality quorum is one that fails to intersect a random quorum with
√
probability at most  and a high quality quorum is picked with probability
√
at least 1 − .
We can also consider load; Malkhi et al. [MRWW01] show that essentially
the same bounds on load for strict quorum systems also hold for -intersecting
√
quorum systems: load(S) ≥ max((E(|Q|)/n, (1− )2 / E(|Q|)), where E(|Q|)
is the expected size of a quorum. The left-hand branch of the max is just
the average load applied to a uniformly-chosen server. For the right-hand
side, pick some high quality quorum Q0 with size less than or equal to
√
(1 − ) E(|Q|) and consider the load applied to its most loaded member by
√
its nonempty intersection (which occurs with probability at least 1 − )
with a random quorum.
CHAPTER 14. QUORUM SYSTEMS 111

14.7 Signed quorum systems

A further generalization of probabilistic quorum systems gives signed quo-
rum systems [Yu06]. In these systems, a quorum consists of some set of
positive members (servers you reached) and negative members (servers you
tried to reach but couldn’t). These allow O(1)-sized quorums while tolerating
n − O(1) failures, under certain natural probabilistic assumptions. Because
the quorums are small, the load on some servers may be very high: so these
are most useful for fault-tolerance rather than load-balancing. See the paper
for more details.
 Part II

Shared memory

112
Chapter 15

Model

Shared memory models describe the behavior of processes in a multiprocessing

system. These processes might correspond to actual physical processors or
processor cores, or might use time-sharing on a single physical processor. In
either case the assumption is that communication is through some sort of
shared data structures.
Here we describe the basic shared-memory model. See also [AW04, §4.1].
The idea of shared memory is that instead of sending messages to each
other, processes communicate through a pool of shared objects. These are
typically registers supporting read and write operations, but fancier objects
corresponding to more sophisticated data structures or synchronization
primitives may also be included in the model.
It is usually assumed that the shared objects do not experience faults.
This means that the shared memory can be used as a tool to prevent partitions
and other problems that can arise in message passing if the number of faults
get too high. As a result, for large numbers of processor failures, shared
memory is a more powerful model than message passing, although we will see
in Chapter 16 that both models can simulate each other provided a majority
of processes are non-faulty.

15.1 Atomic registers

An atomic register supports read and write operations; we think of these
as happening instantaneously, and think of operations of different processes
as interleaved in some sequence. Each read operation on a particular reg-
ister returns the value written by the last previous write operation. Write
operations return nothing.

113
CHAPTER 15. MODEL 114

A process is defined by giving, for each state, the operation that it would
like to do next, together with a transition function that specifies how the
state will be updated in response to the return value of that operation. A
configuration of the system consists of a vector of states for the processes
and a vector of value for the registers. A sequential execution consists of a
sequence of alternating configurations and operations C0 , π1 , C1 , π2 , C2 . . . ,
where in each triple Ci , πi+1 , Ci+1 , the configuration Ci+1 is the result of
applying πi+1 to configuration Ci . For read operations, this means that the
state of the reading process is updated according to its transition function.
For write operations, the state of the writing process is updated, and the
state of the written register is also updated.
Pseudocode for shared-memory protocols is usually written using stan-
dard pseudocode conventions, with the register operations appearing either
as explicit subroutine calls or implicitly as references to shared variables.
Sometimes this can lead to ambiguity; for example, in the code fragment

done ← leftDone ∧ rightDone,

it is clear that the operation write(done, −) happens after read(leftDone) and

read(rightDone), but it is not clear which of read(leftDone and read(rightDone)
happens first. When the order is important, we’ll write the sequence out
explicitly:

1 leftIsDone ← read(leftDone)
2 rightIsDone ← read(rightDone)
3 write(done, leftIsDone ∧ rightIsDone)

Here leftIsDone and rightIsDone are internal variables of the process, so

using them does not require read or write operations to the shared memory.

15.2 Single-writer versus multi-writer registers

One variation that does come up even with atomic registers is what processes
are allowed to read or write a particular register. A typical assumption is that
registers are single-writer multi-reader—there is only one process that
can write to the register (which simplifies implementation since we don’t have
to arbitrate which of two near-simultaneous writes gets in last and thus leaves
the long-term value), although it’s also common to assume multi-writer
multi-reader registers, which if not otherwise available can be built from
CHAPTER 15. MODEL 115

single-writer multi-reader registers using atomic snapshot (see Chapter 19).

Less common are single-writer single-reader registers, which act much
like message-passing channels except that the receiver has to make an explicit
effort to pick up its mail.

15.3 Fairness and crashes

From the perspective of a schedule, the fairness condition says that every
processes gets to perform an operation infinitely often, unless it enters either
a crashed or halting state where it invokes no further operations. (Note
that unlike in asynchronous message-passing, there is no way to wake up a
process once it stops doing operations, since the only way to detect that any
activity is happening is to read a register and notice it changed.) Because the
registers (at least in in multi-reader models) provide a permanent fault-free
record of past history, shared-memory systems are much less vulnerable to
crash failures than message-passing systems (though FLP1 still applies); so
in extreme cases, we may assume as many as n − 1 crash failures, which
makes the fairness condition very weak. The n − 1 crash failures case is
called the wait-free case—since no process can wait for any other process
to do anything—and has been extensively studied in the literature.
For historical reasons, work on shared-memory systems has tended to
assume crash failures rather than Byzantine failures—possibly because Byzan-
tine failures are easier to prevent when you have several processes sitting in
the same machine than when they are spread across the network, or possibly
because in multi-writer situations a Byzantine process can do much more
damage. But the model by itself doesn’t put any constraints on the kinds of
process failures that might occur.

15.4 Concurrent executions

Often, the operations on our shared objects will be implemented using lower-
level operations. When this happens, it no longer makes sense to assume that
the high-level operations occur one at a time—although an implementation
may try to give that impression to its users. To model to possibility of
concurrency between operations, we split an operation into an invocation
and response, corresponding roughly to a procedure call and its return. The
user is responsible for invoking the object; the object’s implementation (or
1
See Chapter 11.
CHAPTER 15. MODEL 116

the shared memory system, if the object is taken as a primitive) is responsible

for responding. Typically we will imagine that an operation is invoked at
the moment it becomes pending, but there may be executions in which that
does not occur. The time between the invocation and the response for an
operation is the interval of the operation.
A concurrent execution is a sequence of invocations and responses,
where after any prefix of the execution, every response corresponds to some
preceding invocation, and there is at most one invocation for each pro-
cess—always the last—that does not have a corresponding response. How
a concurrent execution may or may not relate to a sequential execution
depends on the consistency properties of the implementation, as described
below.

15.5 Consistency properties

Different shared-memory systems may provide various consistency proper-
ties, which describe how views of an object by different processes mesh with
each other. The strongest consistency property generally used is lineariz-
ability [HW90], where an implementation of an object is linearizable if, for
any concurrent execution of the object, there is a sequential execution of the
object with the same operations and return values, where the (total) order
of operations in the sequential execution is a linearization of the (partial)
order of operations in the concurrent execution.
Less formally, this means that if operation a finishes before operation
b starts in the concurrent execution, then a must come before b in the
sequential execution. An equivalent definition is that we can assign each
operation a linearization point somewhere between when its invocation
and response, and the sequential execution obtained by assuming that all
operations occur atomically at their linearization points is consistent with
the specification of the object. Using either definition, we are given a fair bit
of flexibility in how to order overlapping operations, which can sometimes
be exploited by clever implementations (or lower bounds).
A weaker condition is sequential consistency [Lam79]. This says that
for any concurrent execution of the object, there exists some sequential
execution that is indistinguishable to all processes; however, this sequential
execution might include operations that occur out of order from a global
perspective. For example, we could have an execution of an atomic register
where you write to it, then I read from it, but I get the initial value that
precedes your write. This is sequentially consistent but not linearizable.
CHAPTER 15. MODEL 117

Mostly we will ask any implementations we consider to be linearizable.

However, both linearizability and sequential consistency are much stronger
than the consistency conditions provided by real multiprocessors. For some
examples of weaker memory consistency rules, a good place to start might
be the dissertation of Jalal Y. Kawash [Kaw00].

15.6 Complexity measures

There are several complexity measures for shared-memory systems.
Time Assume that no process takes more than 1 time unit between opera-
tions (but some fast processes may take less). Assign the first operation
in the schedule time 1 and each subsequent operation the largest time
consistent with the bound. The time of the last operation is the time
complexity. This is also known as the big-step or round measure
because the time increases by 1 precisely when every non-faulty process
has taken at least one step, and a minimum interval during which this
occurs counts as a big step or a round.
Total work The total work or total step complexity is just the length
of the schedule, i.e., the number of operations. This doesn’t consider
how the work is divided among the processes, e.g., an O(n2 ) total
work protocol might dump all O(n2 ) operations on a single process
and leave the rest with almost nothing to do. There is usually not
much of a direct correspondence between total work and time. For
example, any algorithm that involves busy-waiting—where a process
repeatedly reads a register until it changes—may have unbounded total
work (because the busy-waiter might spin very fast) even though it
runs in bounded time (because the register gets written to as soon as
some slower process gets around to it). However, it is trivially the case
that the time complexity is never greater than the total work.
Per-process work The per-process work, individual work, per-process
step complexity, or individual step complexity measures the
maximum number of operations performed by any single process. Op-
timizing for per-process work produces more equitably distributed
workloads (or reveals inequitably distributed workloads). Like total
work, per-process work gives an upper bound on time, since each time
unit includes at least one operation from the longest-running process,
but time complexity might be much less than per-process work (e.g.,
in the busy-waiting case above).
CHAPTER 15. MODEL 118

Remote memory references As we’ve seen, step complexity doesn’t make

much sense for processes that busy-wait. An alternative measure is
remote memory reference complexity or RMR complexity. This
measure charges one unit for write operations and the first read op-
eration by each process following a write, but charges nothing for
subsequent read operations if there are no intervening writes (see §17.5
for details). In this measure, a busy-waiting operation is only charged
one unit. RMR complexity can be justified to a certain extent by the
cost structure of multi-processor caching [MCS91, And90].

Contention In multi-writer or multi-reader situations, it may be bad to

have too many processes pounding on the same register at once. The
contention measures the maximum number of pending operations on
any single register during the schedule (this is the simplest of several
definitions out there). A single-reader single-writer algorithm always
has contention at most 2, but achieving such low contention may
be harder for multi-reader multi-writer algorithms. Of course, the
contention is never worse that n, since we assume each process has at
most one pending operation at a time.

Space Just how big are those registers anyway? Much of the work in this
area assumes they are very big. But we can ask for the maximum
number of bits in any one register (width) or the total size (bit
complexity) or number (space complexity) of all registers, and will
try to minimize these quantities when possible. We can also look at
the size of the internal states of the processes for another measure of
space complexity.

15.7 Fancier registers

In addition to stock read-write registers, one can also imagine more tricked-out
registers that provide additional operations. These usually go by the name
of read-modify-write (RMW) registers, since the additional operations
consist of reading the state, applying some function to it, and writing the
state back, all as a single atomic action. Examples of RMW registers that
have appeared in real machines at various times in the past include:

Test-and-set bits A test-and-set operation sets the bit to 1 and returns

the old value.
CHAPTER 15. MODEL 119

Fetch-and-add registers A fetch-and-add operation adds some incre-

ment (typically -1 or 1) to the register and returns the old value.

Compare-and-swap registers A compare-and-swap operation writes

a new value only if the previous value is equal to a supplied test value.

These are all designed to solve various forms of mutual exclusion or

locking, where we want at most one process at a time to work on some shared
data structure.
Some more exotic read-modify-write registers that have appeared in the
literature are

Fetch-and-cons Here the contents of the register is a linked list; a fetch-

and-cons adds a new head and returns the old list.

Sticky bits (or sticky registers) With a sticky bit or sticky regis-
ter [Plo89], once the initial empty value is overwritten, all further
writes fail. The writer is not notified that the write fails, but may
be able to detect this fact by reading the register in a subsequent
operation.

Bank accounts Replace the write operation with deposit, which adds a
non-negative amount to the state, and withdraw, which subtracts a
non-negative amount from the state provided the result would not go
below 0; otherwise, it has no effect.

These solve problems that are hard for ordinary read/write registers under
bad conditions. Note that they all have to return something in response to
an invocation.
There are also blocking objects like locks or semaphores, but these don’t
fit into the RMW framework.
We can also consider generic read-modify-write registers that can compute
arbitrary functions (passed as an argument to the read-modify-write opera-
tion) in the modify step. Here we typically assume that the read-modify-write
operation returns the old value of the register. Generic read-modify-write
registers are not commonly found in hardware but can be easily simulated
(in the absence of failures) using mutual exclusion.2

2
See Chapter 17.
Chapter 16

Distributed shared memory

In distributed shared memory, our goal is to simulate a collection of

memory locations or registers, each of which supports a read operation
that returns the current state of the register and a write operation that
updates the state. Our implementation should be linearizable [HW90],
meaning that read and write operations appear to occur instantaneously
(atomically) at some point in between when the operation starts and the
operation finishes; equivalently, there should be some way to order all the
operations on the registers to obtain a sequential execution consistent
with the behavior of a real register (each read returns the value of the most
recent write) while preserving the observable partial order on operations
(where π1 precedes π2 if π1 finishes before π2 starts). Implicit in this definition
is the assumption that implemented operations take place over some interval,
between an invocation that starts the operation and a response that ends
the operation and returns its value.1
In the absence of process failures, we can just assign each register to
some process, and implement both read and write operations by remote
procedure calls to the process (in fact, this works for arbitrary shared-memory
objects). With process failures, we need to make enough copies of the register
that failures can’t destroy all of them. This creates an asymmetry between
simulations of message-passing from shared-memory and vice versa; in the
former case (discussed briefly in §16.1 below), a process that fails in the
underlying shared-memory system only means that the same process fails in
the simulated message-passing system. But in the other direction, not only
does the failure of a process in the underlying message-passing system mean
that the same process fails in the simulated shared-memory system, but the
1
More details on the shared-memory model are given in Chapter 15.

120
CHAPTER 16. DISTRIBUTED SHARED MEMORY 121

simulation collapses completely if a majority of processes fail.

16.1 Message passing from shared memory

We’ll start with the easy direction. We can build a reliable FIFO channel
from single-writer single-reader registers using polling. The naive approach
is that for each edge uv in the message-passing system, we create a (very big)
register ruv , and u writes the entire sequence of every message it has ever
sent to v to ruv every time it wants to do a new send. To receive messages,
v polls all of its incoming registers periodically and delivers any messages in
the histories that it hasn’t processed yet.2
The ludicrous register width can be reduced by adding in an acknowl-
edgment mechanism in a separate register ackvu ; the idea is that u will only
write one message at a time to ruv , and will queue subsequent messages until
v writes in ackvu that the message in ruv has been received. With some
tinkering, it is possible to knock ruv down to only three possible states (send-
ing 0, sending 1, and reset) and ackvu down to a single bit (value-received,
reset-received), but that’s probably overkill for most applications.
Process failures don’t affect any of these protocols, except that a dead
process stops sending and receiving.

16.2 Shared memory from message passing: the

Attiya-Bar-Noy-Dolev algorithm
Here we show how to implement shared memory from message passing. We’ll
assume that our system is asynchronous, that the network is complete, and
that we are only dealing with f < n/2 crash failures. We’ll also assume we
only want to build single-writer registers, just to keep things simple; we can
extend to multi-writer registers later.
Here’s the algorithm, which is due to Attiya, Bar-Noy, and Dolev [ABND95];
see also [Lyn96, §17.1.3]. (Section 9.3 of [AW04] gives an equivalent algorithm,
but the details are buried in an implementation of totally-ordered broadcast).
We’ll make n copies of the register, one on each process. Each process’s copy
will hold a pair (value, timestamp) where timestamps are (unbounded) integer
values. Initially, everybody starts with (⊥, 0). A process updates its copy
2
If we are really cheap about using registers, and are willing to accept even more
absurdity in the register size, we can just have u write every message it ever sends to ru ,
and have each v poll all the ru and filter out any messages intended for other processes.
CHAPTER 16. DISTRIBUTED SHARED MEMORY 122

with new values (v, t) upon receiving write(v, t) from any other process p,
provided t is greater than the process’s current timestamp. It then responds
to p with ack(v, t), whether or not it updated its local copy. A process will
also respond to a message read(u) with a response ack(value, timestamp, u);
here u is a nonce3 used to distinguish between different read operations so
that a process can’t be confused by out-of-date acknowledgments.
To write a value, the writer increments its timestamp, updates its value
and sends write(value, timestamp) to all other processes. The write operation
terminates when the writer has received acknowledgments containing the
new timestamp value from a majority of processes.
To read a value, a reader does two steps:

1. It sends read(u) to all processes (where u is any value it hasn’t used

before) and waits to receive acknowledgments from a majority of the
processes. It takes the value v associated with the maximum timestamp
t as its return value (no matter how many processes sent it).

2. It then sends write(v, t) to all processes, and waits for a response

ack(v, t) from a majority of the processes. Only then does it return.

(Any extra messages, messages with the wrong nonce, etc., are discarded.)
Both reads and writes cost Θ(n) messages (Θ(1) per process).
Intuition: Nobody can return from a write or a read until they are sure
that subsequent reads will return the same (or a later) value. A process
can only be sure of this if it knows that the values collected by a read will
include at least one copy of the value written or read. But since majorities
overlap, if a majority of the processes have a current copy of v, then the
majority read quorum will include it. Sending write(v, t) to all processes
and waiting for acknowledgments from a majority is just a way of ensuring
that a majority do in fact have timestamps that are at least t.
If we omit the write stage of a read operation, we may violate lineariz-
ability. An example would be a situation where two values (1 and 2, say),
have been written to exactly one process each, with the rest still holding the
initial value ⊥. A reader that observes 1 and (n − 1)/2 copies of ⊥ will return
1, while a reader that observes 2 and (n − 1)/2 copies of ⊥ will return 2. In
the absence of the write stage, we could have an arbitrarily long sequence
of readers return 1, 2, 1, 2, . . . , all with no concurrency. This would not be
3
A nonce is any value that is guaranteed to be used at most once (the term originally
comes from cryptography, which in turn got it from linguistics). In practice, a reader will
most likely generate a nonce by combining its process id with a local timestamp.
CHAPTER 16. DISTRIBUTED SHARED MEMORY 123

consistent with any sequential execution in which 1 and 2 are only written
once.

16.3 Proof of linearizability

Our intuition may be strong, but we still need a proof the algorithm works. In
particular, we want to show that for any trace T of the ABD protocol, there
is an trace of an atomic register object that gives the same sequence of invoke
and response events. The usual way to do this is to find a linearization
of the read and write operations: a total order that extends the observed
order in T where π1 < π2 in T if and only if π1 ends before π2 starts.
Sometimes it’s hard to construct such an order, but in this case it’s easy:
we can just use the timestamps associated with the values written or read
in each operation. Specifically, we define the timestamp of a write or read
operation as the timestamp used in the write(v, t) messages sent out during
the implementation of that operation, and we put π1 before π2 if:

1. π1 has a lower timestamp than π2 , or

2. π1 has the same timestamp as π2 , π1 is a write, and π2 is a read, or

3. π1 has the same timestamp as π2 and π1 <T π2 , or

4. none of the other cases applies, and we feel like putting π1 first.

The intent is that we pick some total ordering that is consistent with both
<T and the timestamp ordering (with writes before reads when timestamps
are equal). To make this work we have to show (a) that these two orderings
are in fact consistent, and (b) that the resulting ordering produces values
consistent with an atomic register: in particular, that each read returns the
value of the last preceding write.
Part (b) is easy: since timestamps only increase in response to writes,
each write is followed by precisely those reads with the same timestamp,
which are precisely those that returned the value written.
For part (a), suppose that π1 <T π2 . The first case is when π2 is a read.
Then before the end of π1 , a set S of more than n/2 processes send the π1
process an ack(v1, t1 ) message. Since local timestamps only increase, from
this point on any ack(v2 , t2 , u) message sent by a process in S has t2 ≥ t1 .
Let S 0 be the set of processes sending ack(v2 , t2 , u) messages processed by
π2 . Since |S| > n/2 and |S 0 | > n/2, we have S ∩ S 0 is nonempty and so S 0
includes a process that sent ack(v2 , t2 ) with t2 ≥ t1 . So π2 is serialized after
CHAPTER 16. DISTRIBUTED SHARED MEMORY 124

π1 . The second case is when π2 is a write; but then π1 returns a timestamp

that precedes the writer’s increment in π2 , and so again is serialized first.

16.4 Proof that f < n/2 is necessary

This is pretty much the standard partition argument that f < n/2 is necessary
to do anything useful in a message-passing system. Split the processes into
two sets S and S 0 of size n/2 each. Suppose the writer is in S. Consider an
execution where the writer does a write operation, but all messages between
S and S 0 are delayed. Since the writer can’t tell if the S 0 processes are slow
or dead, it eventually returns. Now let some reader in S 0 attempt to read
the simulated register, again delaying all messages between S and S 0 ; now
the reader is forced to return some value without knowing whether the S
processes are slow or dead. If the reader doesn’t return the value written,
we lose. If by some miracle it does, then we lose in the execution where the
write didn’t happen and all the processes in S really were dead.

16.5 Multiple writers

So far we have assumed a single writer. The main advantage of this approach
is that we don’t have to do much to manage timestamps: the single writer
can just keep track of its own. With multiple writers we can use essentially
the same algorithm, but each write needs to perform an initial round of
gathering timestamps so that it can pick a new timestamp bigger than those
that have come before. We also extend the timestamps to be of the form
hcount, idi, lexicographically ordered, so that two timestamps with the same
count field are ordered by process id. The modified write algorithm is:
1. Send read(u) to all processes and wait to receive acknowledgments
from a majority of the processes.

2. Set my timestamp to t = (maxq countq + 1, id) where the max is taken

over all processes q that sent me an acknowledgment. Note that this is
a two-field timestamp that is compared lexicographically, with the id
field used only to prevent duplicate timestamps.

3. Send write(v, t) to all processes, and wait for a response ack(v, t) from
a majority of processes.
This increases the cost of a write by a constant factor, but in the end we
still have only a linear number of messages. The proof of linearizability is
CHAPTER 16. DISTRIBUTED SHARED MEMORY 125

essentially the same as for the single-writer algorithm, except now we must
consider the case of two write operations by different processes. Here we have
that if π1 <T π2 , then π1 gets acknowledgments of its write with timestamp
t1 from a majority of processes before π2 starts its initial phase to compute
count. Since π2 waits for acknowledgments from a majority of processes as
well, these majorities overlap, so π2 ’s timestamp t2 must exceed t1 . So the
linearization ordering previously defined still works.

16.6 Other operations

The basic ABD framework can be extended to support other operations.
One such operation is a collect [SSW91], where we read n registers in
parallel with no guarantee that they are read at the same time. This can
trivially be implemented by running n copies of ABD in parallel, and can
be implemented with the same time and message complexity as ABD for a
single register by combining the messages from the parallel executions into
single (possibly very large) messages.
Chapter 17

Mutual exclusion

For full details see [AW04, Chapter 4] or [Lyn96, Chapter 10].

17.1 The problem

The goal is to share some critical resource between processes without more
than one using it at a time—this is the fundamental problem in time-sharing
systems.
The solution is to only allow access while in a specially-marked block of
code called a critical section, and only allow one process at a time to be
in a critical section.
A mutual exclusion protocol guarantees this, usually in an asyn-
chronous shared-memory model.
Formally: We want a process to cycle between states trying (trying to
get into critical section), critical (in critical section), exiting (cleaning up
so that other processes can enter their critical sections), and remainder
(everything else—essentially just going about its non-critical business). Only
in the trying and exiting states does the process run the mutual exclusion
protocol to decide when to switch to the next state; in the critical or remainder
states it switches to the next state on its own.
The ultimate payoff is that mutual exclusion solves for systems without
failures what consensus solves for systems with failures: if the only way to
update a data structure is to hold a lock on it, we are guaranteed to get
a nice clean sequence of atomic-looking updates. Of course, once we allow
failures back in, mutex becomes less useful, as our faulty processes start
crashing without releasing their locks, and with the data structure in some

126
CHAPTER 17. MUTUAL EXCLUSION 127

broken, half-updated state.1

17.2 Goals
(See also [AW04, §4.2], [Lyn96, §10.2].)
Core mutual exclusion requirements:

Mutual exclusion At most one process is in the critical state at a time.

No deadlock (progress) If there is at least one process in a trying state,

then eventually some process enters a critical state; similarly for exiting
and remainder states.

Note that the protocol is not required to guarantee that processes leave
the critical or remainder state, but we generally have to insist that the
processes at least leave the critical state on their own to make progress.
An additional useful property (not satisfied by all mutual exclusion
protocols; see [Lyn96, §10.4)]:

No lockout (lockout-freedom): If there is a particular process in a trying

or exiting state, that process eventually leaves that state. This means
that I don’t starve because somebody else keeps jumping past me and
seizing the critical resource before I can.

Stronger starvation guarantees include explicit time bounds (how many

rounds can go by before I get in) or bounded bypass (nobody gets in more
than k times before I do). Each of these imply lockout-freedom assuming no
deadlock.

17.3 Mutual exclusion using strong primitives

See [AW04, §4.3] or [Lyn96, 10.9]. The idea is that we will use some sort of
read-modify-write register, where the RMW operation computes a new
value based on the old value of the register and writes it back as a single
atomic operation, usually returning the old value to the caller as well.
1
In principle, if we can detect that a process has failed, we can work around this problem
by allowing some other process to bypass the lock and clean up. This may require that
the original process leaves behind notes about what it was trying to do, or perhaps copies
the data it is going to modify somewhere else before modifying it. But even this doesn’t
work if some zombie process can suddenly lurch to life and scribble its ancient out-of-date
values all over our shiny modern data structure.
CHAPTER 17. MUTUAL EXCLUSION 128

17.3.1 Test and set

A test-and-set operation does the following sequence of actions atomically:

1 oldValue ← read(bit)
2 write(bit, 1)
3 return oldValue

Typically there is also a second reset operation for setting the bit back
to zero. For some implementations, this reset operation may only be used
safely by the last process to get 0 from the test-and-set bit.
Because a test-and-set operation is atomic, if two processes both try to
perform test-and-set on the same bit, only one of them will see a return value
of 0. This is not true if each process simply executes the above code on a
stock atomic register: there is an execution in which both processes read
0, then both write 1, then both return 0 to whatever called the non-atomic
test-and-set subroutine.
Test-and-set provides a trivial implementation of mutual exclusion, shown
in Algorithm 17.1.

1 while true do
// trying
2 while TAS(lock) = 1 do nothing
// critical
3 (do critical section stuff)
// exiting
4 reset(lock)
// remainder
5 (do remainder stuff)
Algorithm 17.1: Mutual exclusion using test-and-set

It is easy to see that this code provides mutual exclusion, as once one
process gets a 0 out of lock, no other can escape the inner while loop until
that process calls the reset operation in its exiting state. It also provides
progress (assuming the lock is initially set to 0); the only part of the code
that is not straight-line code (which gets executed eventually by the fairness
condition) is the inner loop, and if lock is 0, some process escapes it, while if
lock is 1, some process is in the region between the TAS call and the reset
CHAPTER 17. MUTUAL EXCLUSION 129

call, and so it eventually gets to reset and lets the next process in (or itself,
if it is very fast).
The algorithm does not provide lockout-freedom: nothing prevents a
single fast process from scooping up the lock bit every time it goes through
the outer loop, while the other processes ineffectually grab at it just after it
is taken away. Lockout-freedom requires a more sophisticated turn-taking
strategy.

17.3.2 A lockout-free algorithm using an atomic queue

Basic idea: In the trying phase, each process enqueues itself on the end of a
shared queue (assumed to be an atomic operation). When a process comes
to the head of the queue, it enters the critical section, and when exiting it
dequeues itself. So the code would look something like Algorithm 17.2.
Note that this requires a queue that supports a head operation. Not all
implementations of queues have this property.

1 while true do
// trying
2 enq(Q, myId)
3 while head(Q) 6= myId do nothing
// critical
4 (do critical section stuff)
// exiting
5 deq(Q)
// remainder
6 (do remainder stuff)
Algorithm 17.2: Mutual exclusion using a queue

Here the proof of mutual exclusion is that only the process whose id is at
the head of the queue can enter its critical section. Formally, we maintain an
invariant that any process whose program counter is between the inner while
loop and the call to deq(Q) must be at the head of the queue; this invariant
is easy to show because a process can’t leave the while loop unless the test
fails (i.e., it is already at the head of the queue), no enq operation changes
the head value (if the queue is nonempty), and the deq operation (which
does change the head value) can only be executed by a process already at
the head (from the invariant).
Deadlock-freedom follows from proving a similar invariant that every
CHAPTER 17. MUTUAL EXCLUSION 130

element of the queue is the id of some process in the trying, critical, or

exiting states, so eventually the process at the head of the queue passes the
inner loop, executes its critical section, and dequeues its id.
Lockout-freedom follows from the fact that once a process is at position
k in the queue, every execution of a critical section reduces its position
by 1; when it reaches the front of the queue (after some finite number of
critical sections), it gets the critical section itself. Alternatively, we can argue
lockout-freedom by showing bounded bypass: once I am in the queue, no
process can execute two critical sections before I do, because once it leaves
its first critical section, it enqueues behind me.

17.3.2.1 Reducing space complexity

Following [AW04, §4.3.2], we can give an implementation of this algorithm
using a single read-modify-write (RMW) register instead of a queue; this
drastically reduces the (shared) space needed by the algorithm. The reason
this works is because we don’t really need to keep track of the position of
each process in the queue itself; instead, we can hand out numerical tickets
to each process and have the process take responsibility for remembering
where its place in line is.
The RMW register has two fields, first and last, both initially 0. In-
crementing last simulates an enqueue, while incrementing first simulates a
dequeue. The trick is that instead of testing if it is at the head of the queue,
a process simply remembers the value of the last field when it “enqueued”
itself, and waits for the first field to equal it.
Algorithm 17.3 shows the code from Algorithm 17.2 rewritten to use this
technique. The way to read the RMW operations is that the first argument
specifies the variable to update and the second specifies an expression for
computing the new value. Each RMW operation returns the old state of the
object, before the update.

17.4 Mutual exclusion using only atomic registers

While mutual exclusion is easier using powerful primitives, we can also solve
the problem using only registers.

17.4.1 Peterson’s algorithm

Algorithm 17.4 shows Peterson’s lockout-free mutual exclusion protocol for
two processes p0 and p1 [Pet81] (see also [AW04, §4.4.2] or [Lyn96, §10.5.1]).
CHAPTER 17. MUTUAL EXCLUSION 131

1 while true do
// trying
2 position ← RMW(V, hV.first, V.last + 1i)
// enqueue
3 while RMW(V, V ).first 6= position.last do
4 nothing
// critical
5 (do critical section stuff)
// exiting
6 RMW(V, hV.first + 1, V.lasti)
// dequeue
// remainder
7 (do remainder stuff)
Algorithm 17.3: Mutual exclusion using read-modify-write

It uses only atomic registers.

This uses three bits to communicate: present[0] and present[1] indicate
which of p0 and p1 are participating, and waiting enforces turn-taking. The
protocol requires that waiting be multi-writer, but it’s OK for present[0] and
present[1] to be single-writer.
In the description of the protocol, we write Lines 8 and 10 as two separate
lines because they include two separate read operations, and the order of
these reads is important.

17.4.1.1 Correctness of Peterson’s protocol

Intuitively, let’s consider all the different ways that the entry code of the two
processes could interact. There are basically two things that each process
does: it sets its own present variable in Line 5 and grabs the waiting variable
in Line 6. Here’s a typical case where one process gets in first:

1. p0 sets present[0] ← 1

2. p0 sets waiting ← 0

3. p0 reads present[1] = 0 and enters critical section

4. p1 sets present[1] ← 1

5. p1 sets waiting ← 1
CHAPTER 17. MUTUAL EXCLUSION 132

shared data:
1 waiting, initially arbitrary
2 present[i] for i ∈ {0, 1}, initially 0
3 Code for process i:
4 while true do
// trying
5 present[i] ← 1
6 waiting ← i
7 while true do
8 if present[¬i] = 0 then
9 break
10 if waiting 6= i then
11 break

// critical
12 (do critical section stuff)
// exiting
13 present[i] = 0
// remainder
14 (do remainder stuff)
Algorithm 17.4: Peterson’s mutual exclusion algorithm for two
processes
CHAPTER 17. MUTUAL EXCLUSION 133

6. p1 reads present[0] = 1 and waiting = 1 and loops

7. p0 sets present[0] ← 0

8. p1 reads present[0] = 0 and enters critical section

The idea is that if I see a 0 in your present variable, I know that you
aren’t playing, and can just go in.
Here’s a more interleaved execution where the waiting variable decides
the winner:

1. p0 sets present[0] ← 1

2. p0 sets waiting ← 0

3. p1 sets present[1] ← 1

4. p1 sets waiting ← 1

5. p0 reads present[1] = 1

6. p1 reads present[0] = 1

7. p0 reads waiting = 1 and enters critical section

8. p1 reads present[0] = 1 and waiting = 1 and loops

9. p0 sets present[0] ← 0

10. p1 reads present[0] = 0 and enters critical section

Note that it’s the process that set the waiting variable last (and thus sees
its own value) that stalls. This is necessary because the earlier process might
long since have entered the critical section.
Sadly, examples are not proofs, so to show that this works in general,
we need to formally verify each of mutual exclusion and lockout-freedom.
Mutual exclusion is a safety property, so we expect to prove it using invariants.
The proof in [Lyn96] is based on translating the pseudocode directly into
automata (including explicit program counter variables); we’ll do essentially
the same proof but without doing the full translation to automata. Below,
we write that pi is at line k if it the operation in line k is enabled but has
not occurred yet.

Lemma 17.4.1. If present[i] = 0, then pi is at Line 5 or 14.

CHAPTER 17. MUTUAL EXCLUSION 134

Proof. Immediate from the code.

Lemma 17.4.2. If pi is at Line 12, and p¬i is at Line 8, 10, or 12, then
waiting = ¬i.

Proof. We’ll do the case i = 0; the other case is symmetric. The proof is by
induction on the schedule. We need to check that any event that makes the
left-hand side of the invariant true or the right-hand side false also makes
the whole invariant true. The relevant events are:

• Transitions by p0 from Line 8 to Line 12. These occur only if present[1] =

0, implying p1 is at Line Line 5 or 14 by Lemma 17.4.1. In this case
the second part of the left-hand side is false.

• Transitions by p0 from Line 10 to Line 12. These occur only if waiting 6=

0, so the right-hand side is true.

• Transitions by p1 from Line 6 to Line 8. These set waiting to 1, making

the right-hand side true.

• Transitions that set waiting to 0. These are transitions by p0 from

Line 6 to Line 10, making the left-hand side false.

We can now read mutual exclusion directly off of Lemma 17.4.2: if

both p0 and p1 are at Line 12, then we get waiting = 1 and waiting = 0, a
contradiction.
To show progress, observe that the only place where both processes can
get stuck forever is in the loop at Lines 8 and 10. But then waiting isn’t
changing, and so some process i reads waiting = ¬i and leaves. To show
lockout-freedom, observe that if p0 is stuck in the loop while p1 enters the
critical section, then after p1 leaves it sets present[1] to 0 in Line 13 (which
lets p0 in if p0 reads present[1] in time), but even if it then sets present[1]
back to 1 in Line 5, it still sets waiting to 1 in Line 6, which lets p0 into
the critical section. With some more tinkering this argument shows that p1
enters the critical section at most twice while p0 is in the trying state, giving
2-bounded bypass; see [Lyn96, Lemma 10.12]. With even more tinkering we
get a constant time bound on the waiting time for process i to enter the
critical section, assuming the other process never spends more than O(1)
time inside the critical section.
CHAPTER 17. MUTUAL EXCLUSION 135

17.4.1.2 Generalization to n processes

(See also [AW04, §4.4.3].)
The easiest way to generalize Peterson’s two-process algorithm to n
processes is to organize a tournament in the form of log-depth binary tree;
this method was invented by Peterson and Fischer [PF77]. At each node
of the tree, the roles of the two processes are taken by the winners of the
subtrees, i.e., the processes who have entered their critical sections in the
two-process algorithms corresponding to the child nodes. The winner of
the tournament as a whole enters the real critical section, and afterwards
walks back down the tree unlocking all the nodes it won in reverse order.
It’s easy to see that this satisfies mutual exclusion, and not much harder
to show that it satisfies lockout-freedom—in the latter case, the essential
idea is that if a winner at some node reaches the root infinitely often, then
lockout-freedom at that node means that a winner of each child node reaches
the root infinitely often.
The most natural way to implement the nodes is to have present[0] and
present[1] at each node be multi-writer variables that can be written to by
any process in the appropriate subtree. Because the present variables don’t
do much, we can also implement them as the OR of many single-writer
variables (this is what is done in [Lyn96, §10.5.3]), but there is no immediate
payoff to doing this since the waiting variables are still multi-writer.
Nice properties of this algorithm are that it uses only bits and that it’s
very fast: O(log n) time in the absence of contention.

17.4.2 Fast mutual exclusion

With a bit of extra work, we can reduce the no-contention cost of mutual
exclusion to O(1), while keeping whatever performance we previously had
in the high-contention case. The trick (due to Lamport [Lam87]) is to put
an object at the entrance to the protocol that diverts a solo process onto a
“fast path” that lets it bypass the n-process mutex that everybody else ends
up on.
Our presentation mostly follows [AW04][§4.4.5], which uses the splitter
abstraction of Moir and Anderson [MA95] to separate out the mechanism
for diverting a lone process.2 Code for a splitter is given in Algorithm 17.5.
A splitter assigns to each processes that arrives at it the value right, down,
or stop. The useful properties of splitters are that if at least one process
2
Moir and Anderson call these things one-time building blocks, but the name
splitter has become standard in subsequent work.
CHAPTER 17. MUTUAL EXCLUSION 136

shared data:
1 atomic register race, big enough to hold an id, initially ⊥
2 atomic register door, big enough to hold a bit, initially open
3 procedure splitter(id)
4 race ← id
5 if door = closed then
6 return right
7 door ← closed
8 if race = id then
9 return stop
10 else
11 return down

Algorithm 17.5: Implementation of a splitter

arrives at a splitter, then (a) at least one process returns right or stop; and
(b) at least one process returns down or stop; (c) at most one process returns
stop; and (d) any process that runs by itself returns stop. The first two
properties will be useful when we consider the problem of renaming in
Chapter 24; we will prove them there. The last two properties are what we
want for mutual exclusion.
The names of the variables race and door follow the presentation in
[AW04, §4.4.5]; Moir and Anderson [MA95], following Lamport [Lam87],
call these X and Y . As in [MA95], we separate out the right and down
outcomes—even though they are equivalent for mutex—because we will need
them later for other applications.
The intuition behind Algorithm 17.5 is that setting door to closed closes
the door to new entrants, and the last entrant to write its id to race wins
(it’s a slow race), assuming nobody else writes race and messes things up.
The added cost of the splitter is always O(1), since there are no loops.
To reset the splitter, write open to door. This allows new processes to
enter the splitter and possibly return stop.
Lemma 17.4.3. After each time that door is set to open, at most one process
running Algorithm 17.5 returns stop.
Proof. To simplify the argument, we assume that each process calls splitter
at most once.
Let t be some time at which door is set to open (−∞ in the case of the
initial value). Let St be the set of processes that read open from door after
CHAPTER 17. MUTUAL EXCLUSION 137

time t and before the next time at which some process writes closed to door,
and that later return stop by reaching Line 9.
Then every process in St reads door before any process in St writes door.
It follows that every process in St writes race before any process in St reads
race. If some process p is not the last process in St to write race, it will not
see its own id, and will not return stop. But only one process can be the last
process in St to write race.3

Lemma 17.4.4. If a process runs Algorithm 17.5 by itself starting from a

configuration in which door = open, it returns stop.

Proof. Follows from examining a solo execution: the process sets race to id,
reads open from door, then reads id from race. This causes it to return stop
as claimed.

To turn this into an n-process mutex algorithm, we use the splitter to

separate out at most one process (the one that gets stop) onto a fast path
that bypasses the slow path taken by the rest of the processes. The slow-
path process first fight among themselves to get through an n-process mutex;
the winner then fights in a 2-process mutex with the process (if any) on the
fast path.
Releasing the mutex is the reverse of acquiring it. If I followed the fast
path, I release the 2-process mutex first then reset the splitter. If I followed
the slow path, I release the 2-process mutex first then the n-process mutex.
This gives mutual exclusion with O(1) cost for any process that arrives before
there is any contention (O(1) for the splitter plus O(1) for the 2-process
mutex).
A complication is that if nobody wins the splitter, there is no fast-path
process to reset it. If we don’t want to accept that the fast path just breaks
forever in this case, we have to include a mechanism for a slow-path process
to reset the splitter if it can be assured that there is no fast-path process
left in the system. The simplest way to do this is to have each process mark
a bit in an array to show it is present, and have each slow-path process,
while still holding all the mutexes, check on its way out if the door bit is set
and no processes claim to be present. If it sees all zeros (except for itself)
after seeing door = closed, it can safely conclude that there is no fast-path
process and reset the splitter itself. The argument then is that the last
slow-path process to leave will do this, re-enabling the fast path once there is
3
It’s worth noting that this last process still might not return stop, because some later
process—not in St —might overwrite race. This can happen even if nobody ever resets the
splitter.
CHAPTER 17. MUTUAL EXCLUSION 138

no contention again. This approach is taken implicitly in Lamport’s original

algorithm, which combines the splitter and the mutex algorithms into a
single miraculous blob.

17.4.3 Lamport’s Bakery algorithm

See [AW04, §4.4.1] or [Lyn96, §10.7].
This is a lockout-free mutual exclusion algorithm that uses only single-
writer registers (although some of the registers may end up holding arbitrarily
large values). Code for the Bakery algorithm is given as Algorithm 17.6.

shared data:
1 choosing[i], an atomic bit for each i, initially 0
2 number[i], an unbounded atomic register, initially 0
3 Code for process i:
4 while true do
// trying
5 choosing[i] ← 1
6 number[i] ← 1 + maxj6=i number[j]
7 choosing[i] ← 0
8 for j 6= i do
9 loop until choosing[j] = 0
10 loop until number[j] = 0 or hnumber[i], ii < hnumber[j], ji
// critical
11 (do critical section stuff)
// exiting
12 number[i] ← 0
// remainder
13 (do remainder stuff)
Algorithm 17.6: Lamport’s Bakery algorithm

Note that several of these lines are actually loops; this is obvious for
Lines 9 and 10, but is also true for Line 6, which includes an implicit loop to
read all n − 1 values of number[j].
Intuition for mutual exclusion is that if you have a lower number than
I do, then I block waiting for you; for lockout-freedom, eventually I have
the smallest number. (There are some additional complications involving
the choosing bits that we are sweeping under the rug here.) For a real proof
see [AW04, §4.4.1] or [Lyn96, §10.7].
CHAPTER 17. MUTUAL EXCLUSION 139

Selling point is a strong near-FIFO guarantee and the use of only single-
writer registers (which need not even be atomic—it’s enough that they return
correct values when no write is in progress). Weak point is unbounded
registers.

17.4.4 Lower bound on the number of registers

There is a famous result due to Burns and Lynch [BL93] that any mutual
exclusion protocol using only read/write registers requires at least n of
them. Details are in [Lyn96, §10.8]. A slightly different version of the
argument is given in [AW04, §4.4.4]. The proof is another nice example of an
indistinguishability proof, where we use the fact that if a group of processes
can’t tell the difference between two executions, they behave the same in
both.
Assumptions: We have a protocol that guarantees mutual exclusion and
progress. Our base objects are all atomic registers.
Key idea: In order for some process p to enter the critical section, it has
to do at least one write to let the other processes know it is doing so. If
not, they can’t tell if p ever showed up at all, so eventually either some p0
will enter the critical section and violate mutual exclusion or (in the no-p
execution) nobody enters the critical section and we violate progress. Now
suppose we can park a process pi on each register ri with a pending write to
i; in this case we say that pi covers ri . If every register is so covered, we
can let p go ahead and do whatever writes it likes and then deliver all the
covering writes at once, wiping out anything p did. Now the other processes
again don’t know if p exists or not. So we can say something stronger: before
some process p can enter a critical section, it has to write to an uncovered
register.
The hard part is showing that we can cover all the registers without
letting p know that there are other processes waiting—if p can see that other
processes are waiting, it can just sit back and wait for them to go through
the critical section and make progress that way. So our goal is to produce
states in which (a) processes p1 . . . , pk (for some k) between them cover k
registers, and (b) the resulting configuration is indistinguishable from an idle
configuration to pk+1 . . . pn , where an idle configuration is one in which
every process is in its remainder section.
Lemma 17.4.5. Starting from any idle configuration C, there exists an exe-
cution in which only processes p1 . . . pk take steps that leads to a configuration
C 0 such that (a) C 0 is indistinguishable by any of pk+1 . . . pn from some idle
configuration C 00 and (b) k distinct registers are covered by p1 . . . pk in C 0 .
CHAPTER 17. MUTUAL EXCLUSION 140

Proof. The proof is by induction on k. For k = 0, let C 00 = C 0 = C.

For larger k, the essential idea is that starting from C, we first run
to a configuration C1 where p1 . . . pk−1 cover k − 1 registers and C1 is
indistinguishable from an idle configuration by the remaining processes, and
then run pk until it covers one more register. If we let p1 . . . pk−1 go, they
overwrite anything pk wrote. Unfortunately, they may not come back to
covering the same registers as before if we rerun the induction hypothesis
(and in particular might cover the same register that pk does). So we have
to look for a particular configuration C1 that not only covers k − 1 registers
but also has an extension that covers the same k − 1 registers.
Here’s how we find it: Start in C. Run the induction hypothesis to get
C1 ; here there is a set W1 of k − 1 registers covered in C1 . Now let processes
p1 through pk−1 do their pending writes, then each enter the critical section,
leave it, and finish, and rerun the induction hypothesis to get to a state C2 ,
indistinguishable from an idle configuration by pk and up, in which k − 1
registers in W2 are covered. Repeat to get sets W3 , W4 , etc. Since this
r

sequence is unbounded, and there are only k−1 distinct sets of registers to
cover (where r is the number of registers), eventually we have Wi = Wj for
some i =6 j. The configurations Ci and Cj are now our desired configurations
covering the same k − 1 registers.
Now that we have Ci and Cj , we run until we get to Ci . We now run pk
until it is about to write some register not covered by Ci (it must do so, or
otherwise we can wipe out all of its writes while it’s in the critical section and
then go on to violate mutual exclusion). Then we let the rest of p1 through
pk−1 do all their writes (which immediately destroys any evidence that pk
ran at all) and run the execution that gets them to Cj . We now have k − 1
registers covered by p1 through pk−1 and a k-th register covered by pk , in a
configuration that is indistinguishable from idle: this proves the induction
step.

The final result follows by the fact that when k = n we cover n registers;
this implies that there are n registers to cover.
It’s worth noting that the execution constructed in this proof might be
very, very long. It’s not clear what happens if we consider executions in which,
say, the critical section is only entered a polynomial number of times. If we
are willing to accept a small probability of failure over polynomially-many
entries, there is a randomized mutex that runs in O(log n) space [AHTW18].
CHAPTER 17. MUTUAL EXCLUSION 141

17.5 RMR complexity

It’s not hard to see that we can’t build a shared-memory mutex without
busy-waiting: any process that is waiting can’t detect that the critical section
is safe to enter without reading a register, but if that register tells it that it
should keep waiting, it is back where it started and has to read it again. This
makes our standard step-counting complexity measures useless for describe
the worst-case complexity of a mutual exclusion algorithm.
However, the same argument that suggests we can ignore local computa-
tion in a message-passing model suggests that we can ignore local operations
on registers in a shared-memory model. Real multiprocessors have memory
hierarchies where memory that is close to the CPU (or one of the CPUs)
is generally much faster than memory that is more distant. This suggests
charging only for remote memory references, or RMRs, where each
register is local to one of the processes and only operations on non-local
registers are expensive. This has the advantage of more accurately modeling
real costs [MCS91, And90], and allowing us to build busy-waiting mutual
exclusion algorithms with costs we can actually analyze.
As usual, there is a bit of a divergence here between theory and practice.
Practically, we are interested in algorithms with good real-time performance,
and RMR complexity becomes a heuristic for choosing how to assign memory
locations. This gives rise to very efficient mutual exclusion algorithms for
real machines, of which the most widely used is the beautiful MCS algorithm
of Mellor-Crummey and Scott [MCS91]. Theoretically, we are interested in
the question of how efficiently we can solve mutual exclusion in our formal
model, and RMR complexity becomes just another complexity measure, one
that happens to allow busy-waiting on local variables.

17.5.1 Cache-coherence vs. distributed shared memory

The basic idea of RMR complexity is that a process doesn’t pay for operations
on local registers. But what determines which operations are local?
In the cache-coherent model (CC for short), once a process reads a
register it retains a local copy as long as nobody updates it. So if I do a
sequence of read operations with no intervening operations by other processes,
I may pay an RMR for the first one (if my cache is out of date), but the rest
are free. The assumption is that each process can cache registers, and there
is some cache-coherence protocol that guarantees that all the caches stay up
to date. We may or may not pay RMRs for write operations or other read
operations, depending on the details of the cache-coherence protocol, but for
CHAPTER 17. MUTUAL EXCLUSION 142

upper bounds it is safest to assume that we do.

In the distributed shared memory model (DSM), each register is
assigned permanently to a single process. Other processes can read or write
the register, but only the owner gets to do so without paying an RMR. Here
memory locations are nailed down to specific processes.
In general, we expect the cache-coherent model to be cheaper than the
distributed shared-memory model, if we ignore constant factors. The reason
is that if we run a DSM algorithm in a CC model, then the process p to
which a register r is assigned incurs an RMR only if some other process
q accesses p since p’s last access. But then we can amortize p’s RMR by
charging q double. Since q incurs an RMR in the CC model, this tells us that
we pay at most twice as many RMRs in DSM as in CC for any algorithm.
The converse is not true: there are (mildly exotic) problems for which it
is known that CC algorithms are asymptotically more efficient than DSM
algorithms [Gol11, DH04].

17.5.2 RMR complexity of Peterson’s algorithm

As a warm-up, let’s look at the RMR complexity of Peterson’s two-process
mutual exclusion algorithm (Algorithm 17.4). Acquiring the mutex requires
going through mostly straight-line code, except for the loop that tests
present[¬i] and waiting.
In the DSM model, spinning on present[¬i] is not a problem (we can
make it a local variable of process i). But waiting is trouble. Whichever
process we don’t assign it to will pay an RMR every time it looks at it. So
Peterson’s algorithm behaves badly by the RMR measure in this model.
Things are better in the CC model. Now process i may pay RMRs for its
first reads of present[¬i] and waiting, but any subsequent reads are free unless
process ¬i changes one of them. But any change to either of the variables
causes process i to leave the loop. It follows that process i pays at most 3
RMRs to get through the busy-waiting loop, giving an RMR complexity of
O(1).
RMR complexities for parts of a protocol that access different registers
add just like step complexities, so the Peterson-Fischer tree construction
described in §17.4.1.2 works here too. The result is O(log n) RMRs per
critical section access, but only in the CC model.
CHAPTER 17. MUTUAL EXCLUSION 143

17.5.3 Mutual exclusion in the DSM model

Yang and Anderson [YA95] give a mutual exclusion algorithm for the DSM
model that requires Θ(log n) RMRs to reach the critical section. This is now
known to be optimal for deterministic algorithms [AHW08]. The core of the
algorithm is a 2-process mutex similar to Peterson’s, with some tweaks so
that each process spins only on its own registers. Pseudocode is given in
Algorithm 17.7; this is adapted from [YA95, Figure 1].

1 C[side(i)] ← i
2 T ←i
3 P [i] ← 0
4 rival ← C[¬side(i)]
5 if rival 6= ⊥ and T = i then
6 if P [rival] = 0 then
7 P [rival] = 1
8 while P [i] = 0 do spin
9 if T = i then
10 while P [i] ≤ 1 do spin

// critical section goes here

11 C[side(i)] ← ⊥
12 rival ← T
13 if rival 6= i then
14 P [rival] ← 2
Algorithm 17.7: Yang-Anderson mutex for two processes

The algorithm is designed to be used in a tree construction where a

process with id in the range {1 . . . n/2} first fights with all other processes
in this range, and similarly for processes in the range {n/2 + 1 . . . n}. The
function side(i) is 0 for the first group of processes and 1 for the second.
The variables C[0] and C[1] are used to record which process is the winner
for each side, and also take the place of the present variables in Peterson’s
algorithm. Each process has its own variable P [i] that it spins on when
blocked; this variable is initially 0 and ranges over {0, 1, 2}; this is used to
signal a process that it is safe to proceed, and tests on P substitute for tests
on the non-local variables in Peterson’s algorithm. Finally, the variable T is
used (like waiting in Peterson’s algorithm) to break ties: when T = i, it’s i’s
turn to wait.
Initially, C[0] = C[1] = ⊥ and P [i] = 0 for all i.
CHAPTER 17. MUTUAL EXCLUSION 144

When I want to enter my critical section, I first set C[side(i)] so you can
find me; this also has the same effect as setting present[side(i)] in Peterson’s
algorithm. I then point T to myself and look for you. I’ll block if I see
C[¬side(i)] = 1 and T = i. This can occur in two ways: one is that I really
write T after you did, but the other is that you only wrote C[¬side(i)] but
haven’t written T yet. In the latter case, you will signal to me that T may
have changed by setting P [i] to 1. I have to check T again (because maybe I
really did write T later), and if it is still i, then I know that you are ahead of
me and will succeed in entering your critical section. In this case I can safely
spin on P [i] waiting for it to become 2, which signals that you have left.
There is a proof that this actually works in [YA95], but it’s 27 pages
of very meticulously-demonstrated invariants (in fairness, this includes the
entire algorithm, including the tree parts that we omitted here). For intuition,
this is not much more helpful than having a program mechanically check all
the transitions, since the algorithm for two processes is effectively finite-state
if we ignore the issue with different processes i jumping into the role of
side(i).
A slightly less rigorous but more human-accessible proof would be analo-
gous to the proof of Peterson’s algorithm. We need to show two things: first,
that no two processes ever both enter the critical section, and second, that
no process gets stuck.
For the first part, consider two processes i and j, where side(i) = 0 and
side(j) = 1. We can’t have both i and j skip the loops, because whichever
one writes T last sees itself in T . Suppose that this is process i and that
j skips the loops. Then T = i and P [i] = 0 as long as j is in the critical
section, so i blocks. Alternatively, suppose i writes T last but does so after
j first reads T . Now i and j both enter the loops. But again i sees T = i on
its second test and blocks on the second loop until j sets P [i] to 2, which
doesn’t happen until after j finishes its critical section.
Now let us show that i doesn’t get stuck. Again we’ll assume that i wrote
T second.
If j skips the loops, then j sets P [i] = 2 on its way out as long as T = i;
this falsifies both loop tests. If this happens after i first sets P [i] to 0, only
i can set P [i] back to 0, so i escapes its first loop, and any j 0 that enters
from the 1 side will see P [i] = 2 before attempting to set P [i] to 1, so P [i]
remains at 2 until i comes back around again. If j sets P [i] to 2 before i sets
P [i] to 0 (or doesn’t set it at all because T = j, then C[side(j)] is set to ⊥
before i reads it, so i skips the loops.
If j doesn’t skip the loops, then P [i] and P [j] are both set to 1 after i
and j enter the loopy part. Because j waits for P [j] 6= 0, when it looks at
CHAPTER 17. MUTUAL EXCLUSION 145

T the second time it will see T = i 6= j and will skip the second loop. This
causes it to eventually set P [i] to 2 or set C[side(j)] to ⊥ before i reads it
as in the previous case, so again i eventually reaches its critical section.
Since the only operations inside a loop are on local variables, the algorithm
has O(1) RMR complexity. For the full tree this becomes O(log n).

17.5.4 Lower bounds

For deterministic algorithms, there is a lower bound due to Attiya, Hendler,
and Woelfel [AHW08] that shows that any one-shot mutual exclusion algo-
rithm for n processes incurs Ω(n log n) total RMRs in either the CC or DSM
models (which implies that some single process incurs Ω(log n) RMRs). This
is based on an earlier breakthrough lower bound of Fan and Lynch [FL06]
that proved the same lower bound for the number of times a register changes
state. Both bounds are information-theoretic: a family of n! executions
is constructed containing all possible orders in which the processes enter
the critical section, and it is shown that each RMR or state change only
contributes O(1) bits to choosing between them.
For randomized algorithms, Hendler and Woelfel [HW11] have an al-
gorithm that uses O(log n/ log log n) expected RMRs against an adaptive
adversary, beating the deterministic lower bound. This is the best possible
for an adaptive adversary, due to a matching lower bound of Giakkoupis and
Woelfel [GW12b] that holds even for systems that provide compare-and-swap
objects.
For an oblivious adversary, an algorithm of Giakkoupis and Woelfel [GW14]
achieves O(1) expected RMRs in the DSM model. With compare-and-swap,
a more recent algorithm of Giakkoupis and Woelfel [GW17] gives O(1) ex-
pected RMRs in the CC model. Curiously, there also exist linearizable
O(1)-RMR implementations of CAS from registers in this model [GHHW12];
however, it is not clear that these implementations can be combined with the
Giakkoupis-Woelfel algorithm to give O(1) expected RMRs using registers,
because variations in scheduling of randomized implementations may produce
subtle conditioning that gives different behavior from actual atomic objects
in the context of a randomized algorithm [GHW11].
Chapter 18

The wait-free hierarchy

In a shared memory model, it may be possible to solve some problems

using wait-free protocols, in which any process can finish the protocol in a
bounded number of steps, no matter what the other processes are doing (see
Chapter 26 for more on this and some variants).
The wait-free hierarchy hrm classifies asynchronous shared-memory
object types T by consensus number, where a type T has consensus
number n if with objects of type T and atomic registers (all initialized to
appropriate values1 ) it is possible to solve wait-free consensus (i.e., agreement,
validity, wait-free termination) for n processes but not for n + 1 processes.
The consensus number of any type is at least 1, since 1-process consensus
requires no interaction, and may range up to ∞ for particularly powerful
objects.
The general idea is that a type T with consensus number c can’t simulate
at type T 0 with a higher consensus number c0 , because then we could use
the simulation to convert a c0 -process consensus protocol using T 0 into a
c0 -process consensus protocol using T . The converse claim, that objects
with the same or higher consensus numbers can simulate those with lower
ones, is not necessarily true: even though n-process consensus can implement
any object for n processes (see §18.2), it may be that for more than n
1
The justification for assuming that the objects can be initialized to an arbitrary state
is a little tricky. The idea is that if we are trying to implement consensus from objects of
type T that are themselves implemented in terms of objects of type S, then it’s natural to
assume that we initialize our simulated type-T objects to whatever states are convenient.
Conversely, if we are using the ability of type-T objects to solve n-process consensus to
show that they can’t be implemented from type-S objects (which can’t solve n-process
consensus), then for both the type-T and type-S objects we want these claims to hold no
matter how they are initialized.

146
CHAPTER 18. THE WAIT-FREE HIERARCHY 147

processes there is some object that has consensus number n but that cannot
be implemented from an arbitrary n-consensus object.2
The wait-free hierarchy was suggested by work by Maurice Herlihy [Her91b]
that classified many common (and some uncommon) shared-memory objects
by consensus number, and showed that an unbounded collection of objects
with consensus number n together with atomic registers gives a wait-free
implementation of any object in an n-process system.
Various subsequent authors noticed that this did not give a robust
hierarchy in the sense that combining two types of objects with consensus
number n could solve wait-free consensus for larger n, and the hierarchy
hrm was proposed by Prasad Jayanti [Jay97] as a way of classifying objects
that might be robust: an object is at level n of the hrm hierarchy if having
unboundedly many objects plus unboundedly many registers solves n-process
wait-free consensus but not (n + 1)-process wait-free consensus.3
Whether or not the resulting hierarchy is in fact robust for arbitrary
deterministic objects is still open, but Ruppert [Rup00] subsequently showed
that it is robust for RMW registers and objects with a read operation that
returns the current state, and there is a paper by Borowsky, Gafni, and
Afek [BGA94] that sketches a proof based on a topological characterization
of computability4 that hrm is robust for deterministic objects that don’t
discriminate between processes (unlike, say, single-writer registers). So for
well-behaved shared-memory objects (deterministic, symmetrically accessible,
with read operations, etc.), consensus number appears to give a real classi-
fication that allows us to say for example that any collection of read-write
registers (consensus number 1), fetch-and-increments (2), test-and-set bits
(2), and queues (2) is not enough to build a compare-and-swap (∞).5
Another useful result from the Borowsky-Gafni-Afek paper is that the
consensus number is not generally dependent on what assumptions we make
about the initial state of the objects. Specifically, [BGA94, Lemma 3.2]
states that as long as there is some sequence of operations that takes an
object from a fixed initial state to a desirable initial state for consensus, then
2
The existence of such objects was eventually demonstrated by Afek, Ellen, and
Gafni [AEG16].
3
The r in hrm stands for the registers, the m for having many objects of the given type.
Jayanti [Jay97] also defines a hierarchy hr1 where you only get finitely many objects. The
h stands for “hierarchy,” or, more specifically, h(T ) stands for the level of the hierarchy at
which T appears [Jay11].
4
See Chapter 28.
5
Ruppert’s paper is particularly handy because it gives an algorithm for computing
the consensus number of the objects it considers. However, for infinite-state objects, this
requires solving the halting problem (as previously shown by Jayanti and Toueg [JT92].
CHAPTER 18. THE WAIT-FREE HIERARCHY 148

we can safely assume that the object is in the desirable state. The core idea
of the proof is that each process can initialize its own copy of the object
and then announce that it is ready; each process will then participate in a
sequence of consensus protocols using the objects that they observe are ready,
with the output of each protocol used as the input to the next. Because the
first object Si to be announced as initialized will be visible to all processes,
they will all do consensus using Si . Any subsequent protocols that may be
used by only a subset of the processes will not change the common agreed
output from the Si protocol.6
When considering multiple objects, the usual assumption is that objects
are combined by putting them next to each other. If we can combine two
objects by constructing a single object with operations of both—which is
essentially what happens when we apply different machine language instruc-
tions to the same memory location—then the object with both operations
may have a higher consensus number than the object with either opera-
tion individually. This was observed by Ellen et al. [EGSZ16]. A simple
example would be a register than supports increment (+1) and doubling
(×2) operations. A register with only one of these operations is equivalent
to a counter and has consensus number 1. But a register with both oper-
ations has consensus number at least 2, since if it is initialized to 2, we
can tell which of the two operations went first by looking at the final value:
3 = 2 + 1, 4 = 2 × 2, 5 = (2 × 2) + 1, 6 = (2 + 1) × 2.
We won’t attempt to do the robustness proofs of Borowsky et al. [BGA94]
or Ruppert [Rup00]. Instead, we’ll concentrate on Herlihy’s original results
and show that specific objects have specific consensus numbers when used in
isolation. The procedure in each case will be to show an upper bound on the
consensus number using a variant of Fischer-Lynch-Paterson (made easier
because we are wait-free and don’t have to worry about fairness) and then
show a matching lower bound (for non-trivial upper bounds) by exhibiting
an n-process consensus protocol for some n. Essentially everything below is
taken from Herlihy’s paper [Her91b], so reading that may make more sense
than reading these notes.

18.1 Classification by consensus number

Here we show the position of various types in the wait-free hierarchy. The
quick description is shown in Table 18.1; more details (mostly adapted
6
The result in the paper is stated for a consensus protocol that uses a single copy of the
object, but it generalizes in the obvious way to those that use multiple copies of the object.
CHAPTER 18. THE WAIT-FREE HIERARCHY 149

Consensus Defining Examples

number characteristic
1 Read with Registers, counters,
interfering generalized counters, max registers,
no-return atomic snapshots.
RMW.
2 Interfering Test-and-set, fetch-and-add, queues,
RMW; queue- process-to-memory swap.
like structures.
m m-process consensus objects.
2m − 2 Atomic m-register write.
∞ First write-like Queue with peek, fetch-and-cons,
operation wins. sticky bits, compare-and-swap,
memory-to-memory swap, memory-to-
memory copy.

Table 18.1: Position of various types in the wait-free hierarchy

from [Her91b]) are given below.

18.1.1 Level 1: atomic registers, counters, other interfering

RMW registers that don’t return the old value
First observe that any type has consensus number at least 1, since 1-process
consensus is trivial.
We’ll argue that a large class of particularly weak objects has consensus
number exactly 1, by running FLP with 2 processes. Recall from Chap-
ter 11 that in the Fischer-Lynch-Paterson [FLP85] proof we classify states
as bivalent or univalent depending on whether both decision values are still
possible, and that with at least one failure we can always start in a bivalent
state (this doesn’t depend on what objects we are using, since it depends
only on having invisible inputs). Since the system is wait-free there is no
constraint on adversary scheduling, and so if any bivalent state has a bivalent
successor we can just do it. So to solve consensus we have to reach a bivalent
configuration C that has only univalent successors, and in particular has a
0-valent and a 1-valent successor produced by applying operations x and y
of processes px and py .
Assuming objects don’t interact with each other behind the scenes, x
and y must be operations of the same object. Otherwise Cxy = Cyx and we
CHAPTER 18. THE WAIT-FREE HIERARCHY 150

get a contradiction.
Now let’s suppose we are looking at atomic registers, and consider cases:

• x and y are both reads, Then x and y commute: Cxy = Cyx, and we
get a contradiction.

• x is a read and y is a write. Then py can’t tell the difference between

Cyx and Cxy, so running py to completion gives the same decision
value from both Cyx and Cxy, another contradiction.

• x and y are both writes. Now py can’t tell the difference between Cxy
and Cy, so we get the same decision value for both, again contradicting
that Cx is 0-valent and Cy is 1-valent.

There’s a pattern to these cases that generalizes to other objects. Suppose

that an object has a read operation that returns its state and one or more
read-modify-write operations that don’t return anything (perhaps we could
call them “modify-write” operations). We’ll say that the MW operations are
interfering if, for any two operations x and y, either:

• x and y commute: Cxy = Cyx.

• One of x and y overwrites the other: Cxy = Cy or Cyx = Cx.

Then no pair of read or modify-write operations can get us out of a

bivalent state, because (a) reads commute; (b) for a read and MW, the
non-reader can’t tell which operation happened first; (c) and for any two
MW operations, either they commute or the overwriter can’t detect that the
first operation happened. So any MW object with uninformative, interfering
MW operations has consensus number 1.
For example, consider a counter that supports operations read, increment,
decrement, and write: a write overwrites any other operation, and increments
and decrements commute with each other, so the counter has consensus
number 1. The same applies to a generalized counter that supports an
atomic x ← x + a operation; as long as this operation doesn’t return the old
value, it still commutes with other atomic increments.
Max registers [AACH12], which have read operations that return the
largest value previously written, also have commutative updates, so they also
have consensus number 1. This gives an example of an object not invented
at the time of Herlihy’s paper that is still covered by Herlihy’s argument.
CHAPTER 18. THE WAIT-FREE HIERARCHY 151

18.1.2 Level 2: interfering RMW objects that return the old

value, queues (without peek)
Suppose now that we have a RMW object that returns the old value, and
suppose that it is non-trivial in the sense that it has at least one RMW
operation where the embedded function f that determines the new value is
not the identity (otherwise RMW is just read). Then there is some value v
such that f (v) 6= v. To solve two-process consensus, have each process pi first
write its preferred value to a register ri , then execute the non-trivial RMW
operation on the RMW object initialized to v. The first process to execute
its operation sees v and decides its own value. The second process sees f (v)
and decides the first process’s value (which it reads from the register). It
follows that non-trivial RMW object has consensus number at least 2.
In many cases, this is all we get. Suppose that the operations of some
RMW type T are non-interfering in a way analogous to the previous definition,
where now we say that x and y commute if they leave the object in the same
state (regardless of what values are returned) and that y overwrites x if the
object is always in the same state after both x and xy (again regardless
of what is returned). The two processes px and py that carry out x and y
know what happened, but a third process pz doesn’t. So if we run pz to
completion we get the same decision value after both Cx and Cy, which
means that Cx and Cy can’t be 0-valent and 1-valent. It follows that no
collection of RMW registers with interfering operations can solve 3-process
consensus, and thus all such objects have consensus number 2. Examples
of these objects include test-and-set bits, fetch-and-add registers, and
swap registers that support an operation swap that writes a new value and
returns the previous value.
There are some other objects with consensus number 2 that don’t fit this
pattern. Define a wait-free queue as an object with enqueue and dequeue
operations (like normal queues), where dequeue returns ⊥ if the queue is
empty (instead of blocking). To solve 2-process consensus with a wait-free
queue, initialize the queue with a single value (it doesn’t matter what the
value is). We can then treat the queue as a non-trivial RMW register where
a process wins if it successfully dequeues the initial value and loses if it gets
empty.7
7
But wait! What if the queue starts empty?
This turns out to be a surprisingly annoying problem, and was one of the motivating
examples for hrm as opposed to Herlihy’s vaguer initial definition.
With one empty queue and nothing else, Jayanti and Toueg [JT92, Theorem 7] show that
there is no solution to consensus for two processes. This is also true for stacks (Theorem 8
CHAPTER 18. THE WAIT-FREE HIERARCHY 152

However, enqueue operations are non-interfering: if px enqueues vx and

py enqueues vy , then any third process can detect which happened first;
similarly we can distinguish enq(x)deq() from deq()enq(x). So to show we
can’t do three process consensus we do something sneakier: given a bivalent
state C with allegedly 0- and 1-valent successors Cenq(x) and Cenq(y),
consider both Cenq(x)enq(y) and Cenq(y)enq(x) and run px until it does
a deq() (which it must, because otherwise it can’t tell what to decide) and
then stop it. Now run py until it also does a deq() and then stop it. We’ve
now destroyed the evidence of the split and poor hapless pz is stuck. In the
case of Cdeq()enq(x) and Cenq(x)deq() on a non-empty queue we can kill
the initial dequeuer immediately and then kill whoever dequeues x or the
value it replaced, and if the queue is empty only the dequeuer knows. In
either case we reach indistinguishable states after killing only 2 witnesses,
and the queue has consensus number at most 2.
Similar arguments work on stacks, deques, and so forth—these all have
consensus number exactly 2.

18.1.3 Level ∞: objects where the first write wins

These are objects that can solve consensus for any number of processes. Here
are a bunch of level-∞ objects:
Queue with peek Has operations enq(x) and peek(), which returns the
first value enqueued. (Maybe also deq(), but we don’t need it for
consensus). Protocol is to enqueue my input and then peek and return
the first value in the queue.

Fetch-and-cons Returns old cdr and adds new car on to the head of a list.
Use preceding protocol where peek() = tail(car :: cdr).

Sticky bit Has a write operation that has no effect unless register is in the
initial ⊥ state. Whether the write succeeds or fails, it returns nothing.
The consensus protocol is to write my input and then return result of
a read.

Compare-and-swap Has CAS(old, new) operation that writes new only if

previous value is old. Use it to build a sticky bit.

Load-linked/store-conditional Like compare-and-swap split into two op-

erations. The operation reads a memory location and marks it. The
from the same paper). But adding a register (Theorem 9) lets you do it. A second empty
queue also works.
CHAPTER 18. THE WAIT-FREE HIERARCHY 153

operation succeeds only if the location has not been changed since the
preceding load-linked by the same process. Can be used to build a
sticky bit.

Memory-to-memory swap Has swap(ri , rj ) operation that atomically

swaps contents of ri with rj , as well as the usual read and write
operations for all registers. Use to implement fetch-and-cons. Alterna-
tively, use two registers input[i] and victory[i] for each process i, where
victory[i] is initialized to 0, and a single central register prize, initialized
to 1. To execute consensus, write your input to input[i], then swap
victory[i] with prize. The winning value is obtained by scanning all
the victory registers for the one that contains a 1, then returning the
corresponding input value.)

Memory-to-memory copy Has a copy(ri , rj ) operation that copies ri to

rj atomically. Use the same trick as for memory-to-memory swap,
where a process copies prize to victory[i]. But now we have a process
follow up by writing 0 to prize. As soon as this happens, the victory
values are now fixed; take the leftmost 1 as the winner.8
Herlihy [Her91b] gives a slightly more complicated version of this
procedure, where there is a separate prize[i] register for each i, and
after doing its copy a process writes 0 to all of the prize registers. This
shows that memory-to-memory copy solves consensus for arbitrarily
many processes even if we insist that copy operations can never overlap.
The same trick also works for memory-to-memory swap, since we can
treat a memory-to-memory swap as a memory-to-memory copy given
that we don’t care what value it puts in the prize[i] register.

Bank accounts A bank account object stores a non-negative integer,

and supports a read operation that returns the current value and a
withdraw(k) operation that reduces the value by k, unless this would
reduce the value below 0, in which case it has no effect.
To solve (binary) consensus with a bank account, start it with 3, and
have each process with input b attempt to withdraw 3 − b from the
account. After the first withdrawal, the object will hold either 0 or 1,
and no further withdrawals will have any effect. So the bank account
acts exactly like a sticky bit where 3 represents ⊥.9
8
Or use any other rule that all processes apply consistently.
9
If you have more money, you can extend this construction to any fixed set of values.
For example, to choose among values v in 0 . . . m − 1, start with 2m and have a process
CHAPTER 18. THE WAIT-FREE HIERARCHY 154

For many years, I assumed that this example demonstrated why cryp-
tocurrencies all seem to use embedded consensus protocols of some
sort. However, it turns out that there is a critical assumption needed
for this proof, which is that more than one process can spend from the
same account. Without this assumption, it has been shown by Guer-
raoui et al. [GKM+ 19] that the consensus number of a single-spender
bank account is 1, and more generally that the consensus number of a
k-spender bank account is exactly k.

18.1.4 Level 2m − 2: simultaneous m-register write

Here we have a (large) collection of atomic registers augmented by an m-
register write operation that performs all the writes simultaneously. The
intuition for why this is helpful is that if p1 writes r1 and rshared while p2
writes r2 and rshared then any process can look at the state of r1 , r2 and
rshared and tell which write happened first. Code for this procedure is given
in Algorithm 18.1; note that up to 4 reads may be necessary to determine
the winner because of timing issues.10
The workings of Algorithm 18.1 are straightforward:

• If the process reads r1 = r2 = ⊥, then we don’t care which went first,

because the reader (or somebody else) already won.

• If the process reads r1 = 1 and then r2 = ⊥, then p1 went first.

• If the process reads r2 = 2 and then r1 = ⊥, then p2 went first. (This

requires at least one more read after checking the first case.)

• Otherwise the process saw r1 = 1 and r2 = 2. Now read rshared : if it’s

1, p2 went first; if it’s 2, p1 went first.

Algorithm 18.1 requires 2-register writes, and will give us a protocol for 2
processes (since the reader above has to participate somewhere to make the
first case work). For m processes, we can do the same thing with m-register
writes. We have a register rpq = rqp for each pair of distinct processes p
m

and q, plus a register rpp for each p; this gives a total of 2 + m = O(m2 )
with input v subtract 2m − v.
10
The main issue is that processes can only read the registers one at a time. An
alternative to running Algorithm 18.1 is to use a double-collect snapshot (see §19.1) to
simulate reading all three registers at once. However, this might require as many as twelve
read operations, since a process doing a snapshot has to re-read all three registers if any of
them change.
CHAPTER 18. THE WAIT-FREE HIERARCHY 155

1 v1 ← r1
2 v2 ← r2
3 if v1 = v2 = ⊥ then
4 return no winner
5 if v1 = 1 and v2 = ⊥ then
// p1 went first
6 return 1
// read r1 again
7 v10 ← r1
8 if v2 = 2 and v10 = ⊥ then
// p2 went first
9 return 2
// both p1 and p2 wrote
10 if rshared = 1 then
11 return 2
12 else
13 return 1
Algorithm 18.1: Determining the winner of a race between 2-register
writes. The assumption is that p1 and p2 each wrote their own ids
to ri and rshared simultaneously. This code can be executed by any
process (including but not limited to p1 or p2 ) to determine which of
these 2-register writes happened first.
CHAPTER 18. THE WAIT-FREE HIERARCHY 156

registers. All registers are initialized to ⊥. Process p then writes its initial
preference to some single-writer register pref p and then simultaneously writes
p to rpq for all q (including rpp ). It then attempts to figure out the first
writer by applying the above test for each q to rpq (standing in for rshared ),
rpp (r1 ) and rqq (r2 ). If it won against all the other processes, it decides its
own value. If not, it repeats the test recursively for some p0 that beat it until
it finds a process that beat everybody, and returns its value. So m-register
writes solve m-process wait-free consensus.
A further tweak gets 2m − 2: run two copies of an (m − 1)-process
protocol using separate arrays of registers to decide a winner for each group.
Then add a second phase where processes contend across the groups. This
involves each process p from group 1 writing the winning id for its group
simultaneously into sp and spq for each q in the other group. To figure out
who won in the end, build a graph of all victories, where there is an edge
from p to q if and only if p beat q in Phase 1 or p’s id was written before q’s
id in Phase 2. The winner is the (unique) process with at least one outgoing
edge and no incoming edges, which will be the process that won its own
group (by writing first) and whose value was written first in Phase 2.
One thing to note about the second phase is that, unlike mutex, we can’t
just have the winners of the two groups fight each other, since this would
not give the wait-free property for non-winners. Instead, we have to allow a
non-winner p to pick up the slack for a slow winner and fight on behalf of
the entire group. This requires an m-process write operation to write sp and
all spq at once.

18.1.4.1 Matching impossibility result

It might seem that the technique used to boost from m-process consensus to
(2m − 2)-process consensus could be repeated to get up to at least Θ(m2 ), but
this turns out not to be the case. The essential idea is to show that in order
to escape bivalence, we have to get to a configuration C where every process
is about to do an m-register write leading to a univalent configuration (since
reads don’t help for the usual reasons, and normal writes can be simulated
by m-register writes with an extra m − 1 dummy registers), and then argue
that these writes can’t overlap too much. So suppose we are in such a
configuration, and suppose that Cx is 0-valent and Cy is 1-valent, and
we also have many other operations z1 . . . zk that lead to univalent states.
Following Herlihy [Her91b], we argue in two steps:

1. There is some register that is written to by x alone out of all the

CHAPTER 18. THE WAIT-FREE HIERARCHY 157

pending operations. Proof: Suppose not. Then the 0-valent configura-

tion Cxyz1 . . . zk is indistinguishable from the 1-valent configuration
Cyz1 . . . zk by any process except px , and we’re in trouble.

2. There is some register that is written to by x and y but not by any of

the zi . Proof:: Suppose not. The each register is written by at most
one of x and y, making it useless for telling which went first; or it is
overwritten by some zi , hiding the value that tells which went first.
So Cxyz1 . . . zk is indistinguishable from Cyxz1 . . . zk for any process
other than px and py , and we’re still in trouble.

Now suppose we have 2m − 1 processes. The first part says that each of
the pending operations (x, y, all of the zi ) writes to 1 single-writer register
and at least k two-writer registers where k is the number of processes leading
to a different univalent value. This gives k + 1 total registers simultaneously
written by this operation. Now observe that with 2m − 1 process, there is
some set of m processes whose operations all lead to a b-valent state; so
for any process to get to a (¬b)-valent state, it must write m + 1 registers
simultaneously. It follows that with only m simultaneous writes we can only
do (2m − 2)-consensus.

18.1.5 Level m: m-process consensus objects

An m-process consensus object has a single consensus operation that,
the first m times it is called, returns the input value in the first operation,
and thereafter returns only ⊥. Clearly this solves m-process consensus. To
show that it doesn’t solve (m + 1)-process consensus even when augmented
with registers, run a bivalent initial configuration to a configuration C where
any further operation yields a univalent state. By an argument similar to
the m-register write case, we can show that the pending operations in C
must all be consensus operations on the same consensus object (anything
else commutes or overwrites). Now run Cxyz1 . . . zk and Cyxz1 . . . zk , where
x and y lead to 0-valent and 1-valent states, and observe that pk can’t
distinguish the resulting configurations because all it got was ⊥. (Note: this
works even if the consensus object isn’t in its initial state, since we know
that before x or y the configuration is still bivalent.)
So the m-process consensus object has consensus number m. This shows
that hrm is nonempty at each level.
A natural question at this point is whether the inability of m-process
consensus objects to solve (m + 1)-process consensus implies robustness of the
hierarchy. One might consider the following argument: given any object at
CHAPTER 18. THE WAIT-FREE HIERARCHY 158

level m, we can simulate it with an m-process consensus object, and since we

can’t combine m-process consensus objects to boost the consensus number,
we can’t combine any objects they can simulate either. The problem here is
that while m-process consensus objects can simulate any object in a system
with m processes (see below), it may be that some objects can do more in a
system with m + 1 objects while still not solving (m + 1)-process consensus.
A simple way to see this would be to imagine a variant of the m-process
consensus object that doesn’t fail completely after m operations; for example,
it might return one of the first two inputs given to it instead of ⊥. This
doesn’t help with solving consensus, but it might (or might not) make it too
powerful to implement using standard m-process consensus objects.

18.2 Universality of consensus

Universality of consensus says that any type that can implement n-
process consensus can, together with atomic registers, give a wait-free im-
plementation of any object in a system with n processes. That consensus
is universal was shown by Herlihy [Her91b] and Plotkin [Plo89]. Both of
these papers spend a lot of effort on making sure that both the cost of each
operation and the amount of space used is bounded. But if we ignore these
constraints, the same result can be shown using a mechanism similar to the
replicated state machines of §12.3.
Here the processes repeatedly use consensus to decide between candidate
histories of the simulated object, and a process successfully completes an
operation when its operation (tagged to distinguish it from other similar
operations) appears in a winning history. A round structure avoids too much
confusion.
Details are given in Algorithm 18.2.
There are some subtleties to this algorithm. The first time that a process
calls consensus (on c[r]), it may supply a dummy input; the idea is that it is
only using the consensus object to obtain the agreed-upon history from a
round it missed. It’s safe to do this, because no process writes r to its round
register until c[r] is complete, so the dummy input can’t be accidentally
chosen as the correct value.
It’s not hard to see that whatever hr+1 is chosen in c[r + 1] is an extension
of hr (it is constructed by appending operations to hr ), and that all processes
agree on it (by the agreement property of the consensus object c[r + 1]. So
this gives us an increasing sequence of consistent histories. We also need to
show that these histories are linearizable. The obvious linearization is just
CHAPTER 18. THE WAIT-FREE HIERARCHY 159

1 procedure apply(π)
// announce my intended operation
2 op[i] ← π
3 while true do
// find a recent round
4 r ← maxj round[j]
// obtain the history as of that round
5 if hr = ⊥ then
6 hr ← consensus(c[r], ⊥)
7 if π ∈ hr then
8 return value π returns in hr
// else attempt to advance
9 h0 ← hr
10 for each j do
11 if op[j] 6∈ h0 then
12 append op[j] to h0

13 hr+1 ← consensus(c[r + 1], h0 )

14 round[i] ← r + 1

Algorithm 18.2: A universal construction based on consensus

CHAPTER 18. THE WAIT-FREE HIERARCHY 160

the most recent version of hr . Suppose some call to apply(π1 ) finishes before
a call to apply(π2 ) starts. Then π1 is contained in some hr when apply(π1 )
finishes, and since π2 can only enter h by being appended at the end, we get
π1 linearized before π2 .
Finally, we need to show termination. The algorithm is written with a
loop, so in principle it could run forever. But we can argue that no process
after executes the loop more than twice. The reason is that a process p puts
its operation in op[p] before it calculates r; so any process that writes r0 > r
to round sees p’s operation before the next round. It follows that p’s value
gets included in the history no later than round r + 2. (We’ll see this sort of
thing again when we do atomic snapshots in Chapter 19.)
Building a consistent shared history is easier with some particular objects
that solve consensus. For example, a fetch-and-cons object that supplies
an operation that pushes a new head onto a linked list and returns the old
head trivially implements the common history above without the need for
helping. One way to implement fetch-and-cons is with memory-to-memory
swap; to add a new element to the list, create a cell with its next pointer
pointing to itself, then swap the next field with the head pointer for the entire
list.
The solutions we’ve described here have a number of deficiencies that
make them impractical in a real system (even more so than many of the
algorithms we’ve described). If we store entire histories in a register, the
register will need to be very, very wide. If we store entire histories as a linked
list, it will take an unbounded amount of time to read the list. For solutions
to these problems, see [AW04, 15.3] or the papers of Herlihy [Her91b] and
Plotkin [Plo89].
Chapter 19

Atomic snapshots

We’ve seen in the previous chapter that there are a lot of things we can’t
make wait-free with just registers. But there are a lot of things we can.
Atomic snapshots are a tool that let us do a lot of these things easily.
An atomic snapshot object acts like a collection of n single-writer
multi-reader atomic registers with a special snapshot operation that returns
(what appears to be) the state of all n registers at the same time. This
is easy without failures: we simply lock the whole register file, read them
all, and unlock them to let all the starving writers in. But it gets harder if
we want a protocol that is wait-free, where any process can finish its own
snapshot or write even if all the others lock up.
We’ll give the usual sketchy description of a couple of snapshot algo-
rithms. More details on early snapshot results can be found in [AW04, §10.3]
or [Lyn96, §13.3]. There is also a reasonably recent survey by Fich on upper
and lower bounds for the problem [Fic05].

19.1 The basic trick: two identical collects equals

a snapshot
Let’s tag any value written with a sequence number, so that each value
written has a seqno field attached to it that increases over time. We can
now detect if a new write has occurred between two reads of the same
variable. Suppose now that we repeatedly perform collects—reads of all
n registers—until two successive collects return exactly the same vector of
values and sequence numbers. We can then conclude that precisely these
values were present in the registers at some time in between the two collects.
This gives us a very simple algorithm for snapshot. Unfortunately, it doesn’t

161
CHAPTER 19. ATOMIC SNAPSHOTS 162

terminate if there are a lot of writers around.1 So we need some way to slow
the writers down, or at least get them to do snapshots for us.

19.2 Snapshots using double collects with helping

This is the approach taken by Afek and his five illustrious co-authors [AAD+ 93]
(see also [AW04, §10.3] or [Lyn96, §13.3.2]): before a process can write to its
register, it first has to complete a snapshot and leave the results behind with
its write.2 This means that if some slow process (including a slow writer,
since now writers need to do snapshots too) is prevented from doing the
two-collect snapshot because too much writing is going on, eventually it can
just grab and return some pre-packaged snapshot gathered by one of the
many successful writers.
Specifically, if a process executing a single snapshot operation σ sees
values written by a single process i with three different sequence numbers
s1 , s2 and s3 , then it can be assured that the snapshot σ3 gathered with
sequence number s3 started no earlier than s2 was written (and thus no
earlier than σ started, since σ read s1 after it started) and ended no later
than σ ended (because σ saw it). It follows that the snapshot can safely
return σ3 , since that represents the value of the registers at some time inside
σ3 ’s interval, which is contained completely within σ’s interval.
So a snapshot repeatedly does collects until either (a) it gets two identical
collects, in which case it can return the results (a direct scan, or (b) it sees
three different values from the same process, in which case it can take the
snapshot collected by the second write (an indirect scan). See pseudocode
in Algorithm 19.1.
Amazingly, despite the fact that updates keep coming and everybody is
trying to do snapshots all the time, a snapshot operation of a single process
is guaranteed to terminate after at most n + 1 collects. The reason is that
1
This isn’t always a problem, since there may be external factors that keep the writers
from showing up too much. Maurice Herlihy and I got away with using exactly this
snapshot algorithm in an ancient, pre-snapshot paper on randomized consensus [AH90a].
The reread-until-no-change idea was used as early as 1977 by Lamport [Lam77].
2
The algorithm is usually called the AADGMS algorithm by people who can remember
all the names—or at least the initials—of the team of superheroes who came up with
it (Afek, Attiya, Dolev, Gafni, Merritt, and Shavit). Historically, this was one of three
independent solutions to the problem that appeared at about the same time. A similar
algorithm for composite registers was given by James Anderson [And94] and a somewhat
different algorithm for consistent scan was given by Aspnes and Herlihy [AH90b]. The
Afek et al. algorithm had the advantage of using bounded registers (in its full version),
and so it and its name for atomic snapshot prevailed over its competitors.
CHAPTER 19. ATOMIC SNAPSHOTS 163

in order to prevent case (a) from holding, the adversary has to supply at
least one new value in each collect after the first. But it can only supply one
new value for each of the n − 1 processes that aren’t doing collects before
case (b) is triggered (it’s triggered by the first process that shows up with a
second new value). Adding up all the collects gives 1 + (n − 1) + 1 = n + 1
collects before one of the cases holds. Since each collect takes n − 1 read
operations (assuming the process is smart enough not to read its own register),
a snapshot operation terminates after at most n2 − 1 reads.

1 procedure updatei (A, v)

2 s ← scan(A)
3 A[i] ← hA[i].count + 1, v, si
4 procedure scan(A)
5 initial ← collect(A)
6 previous ← initial while true do
7 s ← scan(A)
8 if s = previous then
// Two identical collects
9 return s
10 else if ∃j : s[j].count ≥ initial[j].count + 2 do
// Three different counts from j
11 return s[j].snapshot
12 else
// Nothing useful, try again
13 previous ← s

Algorithm 19.1: Snapshot of [AAD+ 93] using unbounded registers

For a write operation, a process first performs a snapshot, then writes

the new value, the new sequence number, and the result of the snapshot
to its register (these are very wide registers). The total cost is n2 − 1 read
operations for the snapshot plus 1 write operation.

19.2.1 Linearizability
We now need to argue that the snapshot vectors returned by the Afek et al.
algorithm really work, that is, that between each matching invoke-snapshot
and respond-snapshot there was some actual time where the registers in the
array contained precisely the values returned in the respond-snapshot action.
CHAPTER 19. ATOMIC SNAPSHOTS 164

We do so by assigning a linearization point to each snapshot vector, a time

at which it appears in the registers (which for correctness of the protocol had
better lie within the interval between the snapshot invocation and response).
For snapshots obtained through case (a), take any time between the two
collects. For snapshots obtained through case (b), take the linearization point
already assigned to the snapshot vector provided by the third write. In the
latter case we argue by induction on termination times that the linearization
point lies inside the snapshot’s interval.
Note that this means that all snapshots were ultimately collected by two
successive collects returning identical values, since any case-(b) snapshot
sits on top of a finite regression of case-(b) snapshots that must end with a
case-(a) snapshot. This means that any snapshot corresponds to an actual
global state of the registers at some point in the execution, which is not true
of all snapshot algorithms. It also means that we can replace the registers in
the snapshot array with other objects that allow us to detect updates (say,
counters or max registers) and still get snapshots.
In an actual execution, the fact that we are waiting for double collects
with no intervening updates means that if there are many writers, eventually
all of them will stall waiting for a case-(a) snapshot to complete. So that
snapshot will complete because all the writers are stuck. In a sense, requiring
writers to do snapshots first almost gives us a form of locking, but without
the vulnerability to failures of a real lock.

19.2.2 Using bounded registers

The simple version of the Afek et al. algorithm requires unbounded registers
(since sequence numbers may grow forever). One of the reasons why this
algorithm required so many smart people was to get rid of this assumption:
the paper describes a (rather elaborate) mechanism for recycling sequence
numbers that prevents unbounded growth (see also [Lyn96, 13.3.3]). In
practice, unbounded registers are probably not really an issue once one
has accepted very large registers, but getting rid of them is an interesting
theoretical problem.
It turns out that with a little cleverness we can drop the sequence numbers
entirely. The idea is that we just need a mechanism to detect when somebody
has done a lot of writes while a snapshot is in progress. A naive approach
would be to have sequence numbers wrap around mod m for some small
constant modulus m; this fails because if enough snapshots happen between
two of my collects, I may notice none of them because all the sequence
numbers wrapped around all the way. But we can augment mod-m sequence
CHAPTER 19. ATOMIC SNAPSHOTS 165

numbers with a second handshaking mechanism that detects when a large

enough number of snapshots have occurred; this acts like the guard bit on
an automobile odometer, than signals when the odometer has overflowed
to prevent odometer fraud by just running the odometer forward an extra
million miles or so.
The result is the full version of Afek et al. [AAD+ 93]. (Our presentation
here follows [AW04, 10.3].) The key mechanism for detecting odometer fraud
is a handshake, a pair of single-writer bits used by two processes to signal
each other that they have done something. Call the processes S (for same)
and D (for different), and supposed we have handshake bits hS and hD . We
then provide operations tryHandshake (signal that something is happening)
and checkHandshake (check if something happened) for each process; these
operations are asymmetric. The code is:

tryHandshake(S): hS ← hD (make the two bits the same)

tryHandshake(D): hD ← ¬hS (make the two bits different)

checkHandshake(S): return hS 6= hD (return true if D changed its bit)

checkHandshake(D): return hS = hD (return true if S changed its bit)

The intent is that checkHandshake returns true if the other process

called tryHandshake after I did. The situation is a bit messy, however, since
tryHandshake involves two register operations (reading the other bit and
then writing my own). So in fact we have to look at the ordering of these
read and write events. Let’s assume that checkHandshake is called by S (so
it returns true if and only if it sees different values). Then we have two cases:
1. checkHandshake(S) returns true. Then S reads a different value in
hD from the value it read during its previous call to tryHandshake(S).
It follows that D executed a write as part of a tryHandshake(D)
operation in between S’s previous read and its current read.

2. checkHandshake(S) returns false. Then S reads the same value in hD

as it read previously. This does not necessarily mean that D didn’t
write hD during this interval—it is possible that D is just very out
of date, and did a write that didn’t change the register value—but it
does mean that D didn’t perform both a read and a write since S’s
previous read.
How do we use this in a snapshot algorithm? The idea is that before
performing my two collects, I will execute tryHandshake on my end of a
CHAPTER 19. ATOMIC SNAPSHOTS 166

pair of handshake bits for every other process. After performing my two
collects, I’ll execute checkHandshake. I will also assume each update (after
performing a snapshot) toggles a mod-2 sequence number bit on the value
stored in its segment of the snapshot array. The hope is that between the
toggle and the handshake, I detect any changes. (See [AW04, Algorithm 30]
for the actual code.)
Does this work? Let’s look at cases:

1. The toggle bit for some process q is unchanged between the two snap-
shots taken by p. Since the bit is toggled with each update, this means
that an even number of updates to q 0 s segment occurred during the
interval between p’s writes. If this even number is 0, we are happy: no
updates means no call to tryHandshake by q, which means we don’t
see any change in q’s segment, which is good, because there wasn’t any.
If this even number is 2 or more, then we observe that each of these
events precedes the following one:

• p’s call to tryHandshake.

• p’s first read.
• q’s first write.
• q’s call to tryHandshake at the start of its second scan.
• q’s second write.
• p’s second read.
• p’s call to checkHandshake.

It follows that q both reads and writes the handshake bits in between
p’s calls to tryHandshake and checkHandshake, so p correctly sees
that q has updated its segment.

2. The toggle bit for q has changed. Then q did an odd number of updates
(i.e., at least one), and p correctly detects this fact.

What does p do with this information? Each time it sees that q has done
a scan, it updates a count for q. If the count reaches 3, then p can determine
that q’s last scanned value is from a scan that is contained completely within
the time interval of p’s scan. Either this is a direct scan, where q actually
performs two collects with no changes between them, or it’s an indirect
scan, where q got its value from some other scan completely contained within
q’s scan. In the first case p is immediately happy; in the second, we observe
that this other scan is also contained within the interval of p’s scan, and so
CHAPTER 19. ATOMIC SNAPSHOTS 167

(after chasing down a chain of at most n − 1 indirect scans) we eventually

reach a direct scan contained within it that provided the actual value. In
either case p returns the value of pair of adjacent collects with no changes
between them that occurred during the execution of its scan operation, which
gives us linearizability.

19.3 Faster snapshots using lattice agreement

The Afek et al. algorithm and its contemporaries all require O(n2 ) operations
for each snapshot. It is possible to get this bound down to O(n) using a
more clever algorithm, [IMCT94] which is the best we can reasonably hope
for in the worst case given that (a) even a collect (which doesn’t guarantee
anything about linearizability) requires Θ(n) operations when implemented
in the obvious way, and (b) there is a linear lower bound, due to Jayanti,
Tan, and Toueg [JTT00], on a large class of wait-free objects that includes
snapshot.3
The first step, due to Attiya, Herlihy, and Rachman [AHR95], is a
reduction to a related problem called lattice agreement.

19.3.1 Lattice agreement

A lattice is a partial order in which every pair of elements x, y has a least
upper bound x ∨ y called the join of x and y and a greatest lower bound
x ∧ y called the meet of x and y. For example, we can make a lattice out
of sets by letting join be union and meet be intersection; or we can make a
lattice out of integers by making join be max and meet be min.
In the lattice agreement problem, each process starts with an input xi
and produces an output yi , where both are elements of some lattice. The
requirements of the problem are:

Comparability For all i, j, yi ≤ yj or yj ≤ yi .

Downward validity For all i, xi ≤ yi .

Upward validity For all i, yi ≤ x1 ∨ x2 ∨ x3 ∨ . . . ∨ xn .

These requirements are analogous to the requirements for consensus. Com-

parability acts like agreement: the views returned by the lattice-agreement
protocol are totally ordered. Downward validity says that each process will
3
But see §21.6 for a faster alternative if we allow either randomization or limits on the
number of times the array is updated.
CHAPTER 19. ATOMIC SNAPSHOTS 168

include its own input in its output. Upward validity acts like validity: an
output can’t include anything that didn’t show up in some input.
For the snapshot algorithm, we also demand wait-freedom: each process
terminates after a bounded number of its own steps, even if other processes
fail.
Note that if we are really picky, we can observe that we don’t actually
need meets; a semi-lattice that provides only joins is enough. In practice
we almost always end up with a full-blown lattice, because (a) we are working
with finite sets, and (b) we generally want to include a bottom element ⊥
that is less than all the other elements, to represent the “empty” state of
our data structure. But any finite join-semi-lattice with a bottom element
turns out to be a lattice, since we can define x ∧ y as the join of all elements
z such that z ≤ x and z ≤ y. We don’t use the fact that we are in a lattice
anywhere, but it does save us two syllables not to have to say “semi-lattice
agreement.”

19.3.2 Connection to vector clocks

The first step in reducing snapshot to lattice agreement is to have each writer
generate a sequence of increasing timestamps r1 , r2 , . . . , and a snapshot
corresponds to some vector of timestamps ht1 , t2 . . . tn i, where ti indicates
the most recent write by pi that is included in the snapshot (in other words,
we are using vector clocks again; see §6.2.3). Now define v ≤ v 0 if vi ≤ vi0 for
all i; the resulting partial order is a lattice, and in particular we can compute
x ∨ y by the rule (x ∨ y)i = xi ∨ yi .
Suppose now that we have a bunch of snapshots that satisfy the com-
parability requirement. This means they are totally ordered. Then we can
construct a sequential execution by ordering the snapshots in increasing order
with each update operation placed before the first snapshot that includes
it. This sequential execution is not necessarily a linearization of the original
execution, and a single lattice agreement object won’t support more than one
operation for each process, but the idea is that we can nonetheless use lattice
agreement objects to enforce comparability between concurrent executions
of snapshot, while doing some other tricks (exploiting, among other things,
the validity properties of the lattice agreement objects) to get linearizability
over the full execution.
CHAPTER 19. ATOMIC SNAPSHOTS 169

19.3.3 The full reduction

The Attiya-Herlihy-Rachman algorithm is given as Algorithm 19.2. It uses
an array of registers Ri to hold round numbers (timestamps); an array Si
to hold values to scan; an unboundedly humongous array Vir to hold views
obtained by each process in some round; and a collection of lattice-agreement
objects LAr , one for each round.

1 procedure scan()
2 for attempt ← 1 to 2 do
3 Ri ← r ← max(R1 . . . Rn ; Ri + 1)
4 collect ← read(S1 . . . Sn )
5 view ← LAr (collect)
// max computation requires a collect
6 if max(R1 . . . Rn ) ≤ Ri then
7 Vir ← view
8 return Vir

// finding nonempty Vjr also requires a collect

9 Vir ← some nonempty Vjr
10 return Vir
Algorithm 19.2: Lattice agreement snapshot

The algorithm makes two attempts to obtain a snapshot. In both cases,

the algorithm advances to the most recent round it sees (or its previous
round plus one, if nobody else has reached this round yet), attempts a collect,
and then runs lattice-agreement to try to get a consistent view. If after
getting its first view it finds that some other process has already advanced
to a later round, it makes a second attempt at a new, higher round r0 and
uses some view that it obtains in this second round, either directly from
lattice agreement, or (if it discovers that it has again fallen behind), it uses
an indirect view from some speedier process.
The reason why I throw away my view if I find out you have advanced to
a later round is not because the view is bad for me but because it’s bad for
you: I might have included some late values in my view that you didn’t see,
breaking consistency between rounds. But I don’t have to do this more than
once; if the same thing happens on my second attempt, I can use an indirect
view as in [AAD+ 93], knowing that it is safe to do so because any collect
that went into this indirect view started after I did.
The update operation is the usual update-and-scan procedure; for com-
CHAPTER 19. ATOMIC SNAPSHOTS 170

pleteness this is given as Algorithm 19.3. To make it easier to reason about

the algorithm, we assume that an update returns the result of the embedded
scan.

1 procedure updatei (v)

2 Si ← (Si .seqno + 1, v)
3 return scan()
Algorithm 19.3: Update for lattice agreement snapshot

19.3.4 Why this works

We need to show three facts:

1. All views returned by the scan operation are comparable; that is, there
exists a total order on the set of views (which can be extended to a
total order on scan operations by breaking ties using the execution
order).

2. The view returned by an update operation includes the update (this

implies that future views will also include the update, giving the correct
behavior for snapshot).

3. The total order on views respects the execution order: if π1 and π2 are
scan operations that return v1 and v2 , then π1 <S π2 implies v1 ≤ v2 .
(This gives us linearization.)

Let’s start with comparability. First observe that any view returned
is either a direct view (obtained from LAr ) or an indirect view (obtained
from Vjr for some other process j). In the latter case, following the chain of
indirect views eventually reaches some direct view. So all views returned for
a given round are ultimately outputs of LAr and thus satisfy comparability.
But what happens with views from different rounds? The lattice-
agreement objects only operate within each round, so we need to ensure that
any view returned in round r is included in any subsequent rounds. This is
where checking round numbers after calling LAr comes in.
Suppose some process i returns a direct view; that is, it sees no higher
round number in either its first attempt or its second attempt. Then at
the time it starts checking the round number in Line 6, no process has yet
written a round number higher than the round number of i’s view (otherwise
CHAPTER 19. ATOMIC SNAPSHOTS 171

i would have seen it). So no process with a higher round number has yet
executed the corresponding collect operation. When such a process does
so, it obtains values that are at least as large as those fed into LAr , and i’s
round-r view is less than or equal to the vector of these values by upward
validity of LAr , and thus less than or equal to the vector of values returned
by LAr0 for r0 > r, by downward validity of LAr0 . So we have comparability
of all direct views, which implies comparability of all indirect views as well.
To show that each view returned by a scan includes any preceding update,
we observe that either a process returns its first-try scan (which includes
the update by downward validity) or it returns the results of a scan in the
second-try round (which includes the update by downward validity in the
later round, since any collect in the second-try round starts after the update
occurs). So no updates are missed.
Now let’s consider two scan operations π1 and π2 where π1 precedes π2
in the execution. We want to show that, for the views v1 and v2 that these
scans return, v1 ≤ v2 . Pick some time between when π1 finishes and π2
starts, and let s be the contents of the registers at this time. Then v1 ≤ s by
upward validity, since any input fed to a lattice agreement object before π1
finishes was collected from a register whose value was no greater than it is in
s. Similarly, s ≤ v2 by downward validity, because v2 is at least as large as
the collect value read by π2 , and this is at least as large as s. So v1 ≤ s ≤ v2 .

19.3.5 Implementing lattice agreement

There are several known algorithms for implementing lattice agreement,
including the original algorithm of Attiya, Herlihy, and Rachman [AHR95]
and an adaptive algorithm of Attiya and Fouren [AF01]. The best of them
(assuming multi-writer registers) is Inoue et al.’s linear-time lattice agreement
protocol [IMCT94].
The intuition behind this protocol is to implement lattice agreement
using divide-and-conquer. The processes are organized into a tree, with each
leaf in the tree corresponding to some process’s input. Internal nodes of
the tree hold data structures that will report increasingly large subsets of
the inputs under them as they become available. At each internal node, a
double-collect snapshot is used to ensure that the value stored at that node
is always the union of two values that appear in its children at the same time.
This is used to guarantee that, so long as each child stores an increasing
sequence of sets of inputs, the parent does so also.
Each process ascends the tree updating nodes as it goes to ensure that
its value is included in the final result. A clever data structure is used to
CHAPTER 19. ATOMIC SNAPSHOTS 172

ensure that out-of-date smaller sets don’t overwrite larger ones at any node,
and the cost of using this data structure and carrying out the double-collect
snapshot at a node with m leaves below it is shown to be O(m). So the total
cost of a snapshot is O(n + n/2 + n/4 + . . . 1) = O(n), giving the linear time
bound.
Let’s now look at the details of this protocol. There are two main
components: the Union algorithm used to compute a new value for each
node of the tree, and the ReadSet and WriteSet operations used to store the
data in the node. These are both rather specialized algorithms and depend
on the details of the other, so it is not trivial to describe them in isolation
from each other; but with a little effort we can describe exactly what each
component demands from the other, and show that it gets it.
The Union algorithm does the usual two-collects-without change trick to
get the values of the children and then stores the result. In slightly more
detail:

1. Perform ReadSet on both children. This returns a set of leaf values.

2. Perform ReadSet on both children again.

3. If the values obtained are the same in both collects, call WriteSet on
the current node to store the union of the two sets and proceed to the
parent node. Otherwise repeat the preceding step.

The requirement of the Union algorithm is that calling ReadSet on a

given node returns a non-decreasing sequence of sets of values; that is, if
ReadSet returns some set S at a particular time and later returns S 0 , then
S ⊆ S 0 . We also require that the set returned by ReadSet is a superset
of any set written by a WriteSet that precedes it, and that it is equal to
some such set. This last property only works if we guarantee that the values
stored by WriteSet are all comparable (which is shown by induction on the
behavior of Union at lower levels of the tree).
Suppose that all these conditions hold; we want to show that the values
written by successive calls to Union are all comparable, that is, for any values
S, S 0 written by union we have S ⊆ S 0 or S 0 ⊆ S. Observe that S = L ∪ R
and S 0 = L0 ∪ R0 where L, R and L0 , R0 are sets read from the children.
Suppose that the Union operation producing S completes its snapshot before
the operation producing S 0 . Then L ⊆ L0 (by the induction hypothesis) and
R ⊆ R0 , giving S ⊆ S 0 .
We now show how to implement the ReadSet and WriteSet operations.
The main thing we want to avoid is the possibility that some large set gets
CHAPTER 19. ATOMIC SNAPSHOTS 173

overwritten by a smaller, older one. The solution is to have m registers

a[1 . . . m], and write a set of size s to every register in a[1 . . . s] (each register
gets a copy of the entire set). Because register a[s] gets only sets of size s or
larger, there is no possibility that our set is overwritten by a smaller one. If
we are clever about how we organize this, we can guarantee that the total
cost of all calls to ReadSet by a particular process is O(m), as is the cost of
the single call to WriteSet in Union.
Pseudocode for both is given as Algorithm 19.4. This is a simplified
version of the original algorithm from [IMCT94], which does the writes in
increasing order and thus forces readers to finish incomplete writes that they
observe, as in Attiya-Bar-Noy-Dolev [ABND95] (see also Chapter 16).

shared data: array a[1 . . . m] of sets, initially ∅

local data: index p, initially 0

1 procedure WriteSet(S)
2 for i ← |S| down to 1 do
3 a[i] ← S

4 procedure ReadSet()
// update p to last nonempty position
5 while true do
6 s ← a[p]
7 if p = m or a[p + 1] = ∅ then
8 break
9 else
10 p←p+1

11 return s
Algorithm 19.4: Increasing set data structure

Naively, one might think that we could just write directly to a[|S|] and
skip the previous ones, but this makes it harder for a reader to detect that
a[|S|] is occupied. By writing all the previous registers, we make it easy to
tell if there is a set of size |S| or bigger in the sequence, and so a reader can
start at the beginning and scan forward until it reaches an empty register,
secure in the knowledge that no larger value has been written.4 Since we
4
This trick of reading in one direction and writing in another dates back to a paper by
Lamport from 1977 [Lam77].
CHAPTER 19. ATOMIC SNAPSHOTS 174

want to guarantee that no reader every spends more that O(m) operations
on an array of m registers (even if it does multiple calls to ReadSet), we also
have it remember the last location read in each call to ReadSet and start
there again on its next call. For WriteSet, because we only call it once, we
don’t have to be so clever, and can just have it write all |S| ≤ m registers.
We need to show linearizability. We’ll do so by assigning a specific
linearization point to each high-level operation. Linearize each call to ReadSet
at the last time that it reads a[p]. Linearize each call to WriteSet(S) at the
first time at which a[|S|] = S and a[i] 6= ∅ for every i < |S| (in other words,
at the first time that some reader might be able to find and return S); if
there is no such time, linearize the call at the time at which it returns. Since
every linearization point is inside its call’s interval, this gives a linearization
that is consistent with the actual execution. But we have to argue that it
is also consistent with a sequential execution, which means that we need
to show that every ReadSet operation returns the largest set among those
whose corresponding WriteSet operations are linearized earlier.
Let R be a call to ReadSet and W a call to WriteSet(S). If R returns S,
then at the time that R reads S from a[|S|], we have that (a) every register
a[i] with i < |S| is non-empty (otherwise R would have stopped earlier), and
(b) |S| = m or a[|S| + 1] = ∅ (as otherwise R would have kept going after
later reading a[|S| + 1]. From the rule for when WriteSet calls are linearized,
we see that the linearization point of W precedes this time and that the
linearization point of any call to WriteSet with a larger set follows it. So
the return value of R is consistent.
The payoff: unless we do more updates than snapshots, don’t want to
assume multi-writer registers, are worried about unbounded space, have a
beef with huge registers, or care about constant factors, it costs no more
time to do a snapshot than a collect. So in theory we can get away with
assuming snapshots pretty much wherever we need them.

19.4 Practical snapshots using LL/SC

Though atomic registers are enough for snapshots, it is possible to get
a much more efficient snapshot algorithm using stronger synchronization
primitives. An algorithm of Riany, Shavit, and Touitou [RST01] uses load-
linked/store-conditional objects to build an atomic snapshot protocol
with linear-time snapshots and constant-time updates using small registers.
We’ll give a sketch of this algorithm here.
The RST algorithm involves two basic ideas: the first is a snapshot
CHAPTER 19. ATOMIC SNAPSHOTS 175

algorithm for a single scanner (i.e., only one process can do snapshots) in
which each updater maintains two copies of its segment, a high copy (that
may be more recent than the current scan) and a low copy (that is guaranteed
to be no more recent than the current scan). The idea is that when a scan is
in progress, updaters ensure that the values in memory at the start of the
scan are not overwritten before the scan is completed, by copying them to
the low registers, while the high registers allow new values to be written
without waiting for the scan to complete. Unbounded sequence numbers,
generated by the scanner, are used to tell which values are recent or not.
As long as there is only one scanner, nothing needs to be done to ensure
that all scans are consistent, and indeed the single-scanner algorithm can be
implemented using only atomic registers. But extending the algorithm to
multiple scanners is tricky. A simple approach would be to keep a separate
low register for each concurrent scan—however, this would require up to n
low registers and greatly increase the cost of an update. Instead, the authors
devise a mechanism, called a coordinated collect, that allows the scanners
collectively to implement a sequence of virtual scans that do not overlap.
Each virtual scan is implemented using the single-scanner algorithm, with its
output written to a common view array that is protected from inconsistent
updates using LL/SC operations (CAS also works). A scanner participates
in virtual scans until it obtains a virtual scan that is useful to it (this means
that the virtual scan has to take place entirely within the interval of the
process’s actual scan operation); the simplest way to arrange this is to have
each scanner perform two virtual scans and return the value obtained by the
second one.
The paper puts a fair bit of work into ensuring that only O(n) view
arrays are needed, which requires handling some extra special cases where
particularly slow processes don’t manage to grab a view before it is reallocated
for a later virtual scan. We avoid this complication by simply assuming an
unbounded collection of view arrays; see the paper for how to do this right.
A more recent paper by Fatourou and Kallimanis [FK07] gives improved
time and space complexity using the same basic technique.

19.4.1 Details of the single-scanner snapshot

The single-scanner snapshot is implemented using a shared currSeq variable
(incremented by the scanner but used by all processes) and an array memory of
n snapshot segments, each of which is divided into a high and low component
consisting of a value and a timestamp. Initially, currSeq is 0, and all memory
locations are initialized to (⊥, 0). This part of the algorithm does not require
CHAPTER 19. ATOMIC SNAPSHOTS 176

LL/SC.
A call to scan copies the first of memory[j].high or memory[j].low that
has a sequence number less than the current sequence number. Pseudocode
is given as Algorithm 19.5.

1 procedure scan()
2 currSeq ← currSeq + 1
3 for j ← 0 to n − 1 do
4 h ← memory[j].high
5 if h.seq < currSeq then
6 view[j] ← h.value
7 else
8 view[j] ← memory[j].low.value

Algorithm 19.5: Single-scanner snapshot: scan

The update operation for process i cooperates by copying memory[i].high

to memory[i].low if it’s old.
The update operation always writes its value to memory[i].high, but
preserves the previous value in memory[i].low if its sequence number indicates
that it may have been present at the start of the most recent call to scan.
This means that scan can get the old value if the new value is too recent.
Pseudocode is given in Algorithm 19.6.

1 procedure update()
2 seq ← currSeq
3 h ← memory[i].high
4 if h.seq 6= seq then
5 memory[i].low ← h
6 memory[i].high ← (value, seq)
Algorithm 19.6: Single-scanner snapshot: update

To show this actually works, we need to show that there is a linearization

of the scans and updates that has each scan return precisely those values
whose corresponding updates are linearized before it. The ordering is based
on when each scan operation S increments currSeq and when each update
operation U reads it; specifically:
• If U reads currSeq after S increments it, then S < U .
CHAPTER 19. ATOMIC SNAPSHOTS 177

• If U reads currSeq before S increments it and S reads memory[i].high

(where i is the process carrying out U ) before U writes it, then S < U .

• If U reads currSeq before S increments it, but S reads memory[i].high

after U writes it, then U < S.

Updates are ordered based on intervening scans (i.e., U1 < U2 if U1 < S

and S < U2 by the above rules), or by the order in which they read currSeq
if there is no intervening scan.
To show this is a linearization, we need first to show that it extends the
ordering between operations in the original schedule. Each of the above rules
has π1 < π2 only if some low-level operation of π1 precedes some low-level
operation of π2 , with the exception of the transitive ordering of two update
events with an intervening scan. But in this last case we observe that if
U1 < S, then U1 writes memory[i].high before S reads it, so if U1 precedes
U2 in the actual execution, U2 must write memory[i].high after S reads it,
implying S < U2 .
Now we show that the values returned by scan are consistent with the
linearization ordering; that, is, for each i, scan copies to view[i] the value in
the last update by process i in the linearization. Examining the code for scan,
we see that a scan operation S takes memory[i].high if its sequence number
is less than currSeq, i.e., if the update operation U that wrote it read currSeq
before S incremented it and wrote memory[i].high before S read it; this gives
U < S. Alternatively, if scan takes memory[i].low, then memory[i].low was
copied by some update operation U 0 from the value written to memory[i].high
by some update U that read currSeq before S incremented it. Here U 0 must
have written memory[i].high before S read it (otherwise S would have taken
the old value left by U ) and since U precedes U 0 (being an operation of the
same process) it must therefor also have written memory[i].high before S read
it. So again we get the first case of the linearization ordering and U < S.
So far we have shown only that S obtains values that were linearized
before it, but not that it ignores values that were linearized after it. So now
let’s consider some U with S < U . Then one of two cases holds:

• U reads currSeq after S increments it. Then U writes a sequence

number in memory[i].high that is greater than or equal to the currSeq
value used by S; so S returns memory[i].low instead, which can’t have
a sequence number equal to currSeq and thus can’t be U ’s value either.

• U reads currSeq before S increments it but writes memory[i].high after

S reads it. Now S won’t return U ’s value from memory[i].high (it didn’t
CHAPTER 19. ATOMIC SNAPSHOTS 178

read it), and won’t get it from memory[i].low either (because the value
that is in memory[i].high will have seq < currSeq, and so S will take
that instead).

So in either case, if S < U , then S doesn’t return U ’s value. This

concludes the proof of correctness.

19.4.2 Extension to multiple scanners

See the paper for details.
The essential idea: view now represents a virtual scan viewr generated
cooperatively by all the scanners working together in some asynchronous
round r. To avoid conflicts, we update viewr using LL/SC or compare-and-
swap (so that only the first scanner to write wins), and pretend that reads
of memory[i] by losers didn’t happen. When viewr is full, start a new virtual
scan and advance to the next round (and thus the next viewr+1 ).

19.5 Applications
Here we describe a few things we can do with snapshots.

19.5.1 Multi-writer registers from single-writer registers

One application of atomic snapshot is building multi-writer registers from
single-writer registers. The idea is straightforward: to perform a write, a
process does a snapshot to obtain the maximum sequence number, tags its
own value with this sequence number plus one, and then writes it. A read
consists of a snapshot followed by returning the value associated with the
largest sequence number (breaking ties by process id). (See [Lyn96, §13.5]
for a proof that this actually works.) This requires using a snapshot that
doesn’t use multi-writer registers, and turns out to be overkill in practice;
there are simpler algorithms that give O(n) cost for reads and writes based
on timestamps (see [AW04, 10.2.3]).
With additional work, it is even possible to eliminate the requirement
of multi-reader registers, and get a simulation of multi-writer multi-reader
registers that goes all the way down to single-writer single-read registers, or
even single-writer single-reader bits. See [AW04, §§10.2.1–10.2.2] or [Lyn96,
§13.4] for details.
CHAPTER 19. ATOMIC SNAPSHOTS 179

19.5.2 Counters
Given atomic snapshots, it’s easy to build a counter (supporting increment,
decrement, and read operations); or, in more generality, a generalized counter
(supporting increments by arbitrary amounts); or, in even more generality,
an object supporting any collection of commutative and associative update
operations (as long as these operations don’t return anything). The idea
is that each process stores in its segment the total of all operations it has
performed so far, and a read operation is implemented using a snapshot
followed by summing the results. This is a case where it is reasonable
to consider multi-writer registers in building the snapshot implementation,
because there is not necessarily any circularity in doing so.

19.5.3 Resilient snapshot objects

The previous examples can be generalized to objects with operations that
either read the current state of the object but don’t update it or update the
state but return nothing, provided the update operations either overwrite each
other (so that Cxy = Cy or Cyx = Cx) or commute (so that Cxy = Cyx).
This was shown by Aspnes and Herlihy [AH90b] and improved on by
Anderson and Moir [AM93] by eliminating unbounded space usage. An-
derson and Moir also defined the terms snapshot objects for those with
separate read and update operations and resilience for the property that
all operations commute or overwrite. The basic idea underneath both of
these papers is to use the multi-writer register construction given above, but
break ties among operations with the same sequence numbers by first placing
overwritten operations before overwriting operations and only then using
process ids.
This almost shows that snapshots can implement any object with con-
sensus number 1 where update operations return nothing, because an object
that is not resilient violates the commute-or-overwrite condition in some
configuration has consensus number at least 2 (see §18.1.2)—in Herlihy’s
terminology, non-resilient objects have interfering operations. It doesn’t quite
work (as observed in the Anderson-Moir paper), because the tie-breaking
procedure assumes a static ordering on which operations overwrite each
other, so that given operations x and y where y overwrites x, y overwrites
x in any configuration. But there may be objects with a dynamic ordering
to how operations interfere, where y overwrites x in some configuration, x
overwrites y in another, and perhaps even the two operations commute in
yet another. This prevents us from achieving consensus, but also breaks the
CHAPTER 19. ATOMIC SNAPSHOTS 180

tie-breaking technique. So it may be possible that there are objects with

consensus number 1 and no-return updates that we still can’t implement
using only registers.
Chapter 20

Lower bounds on
perturbable objects

Being able to do snapshots in linear time means that we can build linearizable
counters, generalized counters, max registers, and so on, in linear time, by
having each reader take a snapshot and combine the contributions of each
updater using the appropriate commutative and associative operation. A
natural question is whether we can do better by exploiting the particular
features of these objects.
Unfortunately, the Jayanti-Tan-Toueg [JTT00] lower bound for per-
turbable objects says each of these objects requires n − 1 space and n − 1
steps for a read operation in the worst case, for any solo-terminating deter-
ministic implementation from historyless objects. Like Burns-Lynch, this
is a worst-case bound based on a covering argument, so it may be possible
to evade it in some cases using either randomization or a restriction on the
length of an execution (see Chapter 21).
Perturbable means that the object has a particular property that makes
the proof work, essentially that the outcome of certain special executions
can be changed by stuffing lots of extra update operations in the middle (see
below for details).
Solo-terminating means that a process finishes its current operation
in a finite number of steps if no other process takes steps in between; it is a
much weaker condition, for example, than wait-freedom.
Historyless objects are those for which any operation either never
changes the state (like a read, but it could be weaker) or always sets the
state to a value that depends only on the operation and not the previous
value (like a write, but it may also return some information about the old

181
CHAPTER 20. LOWER BOUNDS ON PERTURBABLE OBJECTS 182

state). The point of historyless objects is that covering arguments work for
them: if there is a process with a pending update operations on some object,
the adversary can use it at any time to wipe out the state of the object and
hide any previous operations from any process except the updater (who, in
a typical covering argument, is quietly killed to keep it from telling anybody
what it saw).
Atomic registers are a common example of a historyless object: the read
never changes the state, and the write always replaces it. Swap objects
(with a swap operation that writes a new state while returning the old state)
are the canonical example, since they can implement any other historyless
object (and even have consensus number 2, showing that even extra consensus
power doesn’t necessarily help here). Test-and-sets (which are basically one-
bit swap objects where you can only swap in 1) are also historyless. In
contrast, anything that looks like a counter or similar object where the new
state is a combination of the old state and the operation is not historyless.
This is important because many of these objects turn out to be perturbable,
and if they were also historyless, we’d get a contradiction.
Below is a sketch of the proof. See the original paper [JTT00] for more
details.
The basic idea is to build a sequence of executions of the form Λk Σk π,
where Λk is a preamble consisting of various complete update operations and
k incomplete update operations by processes p1 through pn−1 , Σk delivers
k delayed writes from the incomplete operations in Λk , and π is a read
operation by pn . To make our life easier, we’ll assume that π performs only
read steps.1
We’ll expand Λk Σk to Λk+1 Σk+1 by inserting new operations in between
Λk and Σk , and argue that because those operations can change the value
returned by π, one of them must write an object not covered in Σk , which
will (after some more work) allow us to cover yet another object.
In order for these covered objects to keep accumulating, the reader has
to keep looking at them. To a first approximation, this means that we want
the first k reads done by π to be from objects written in Σk : since the
values seen by the reader for these objects never change, the (deterministic)
1
The idea is that if π does anything else, then the return values of other steps can
be simulated by doing a read in place of the first step and using the property of being
historyless to compute the return values of subsequent steps. There is still a possible
objection that we might have some historyless objects that don’t even provide read steps.
The easiest way to work around this is to assume that our objects do in fact provide a read
step, because taking the read step away isn’t going to make implementing the candidate
perturbable object any easier.
CHAPTER 20. LOWER BOUNDS ON PERTURBABLE OBJECTS 183

reader will continue to read them even as we add more operations before Σk .
Unfortunately, this does not quite match all possible cases, because it may
be that π performs useless reads of objects that aren’t covered in Σk but
that aren’t written to by anybody anyway. So we have the more technical
condition that π has an initial prefix that only includes covered reads and
useless reads: formally, there is a prefix π 0 of π that includes at least one read
operation of every object covered by Σk , such that any other read operation
in π 0 reads an object whose state cannot be changed by any step that can be
performed by any sequence of operations by processes p1 through pn−1 that
can be inserted between Λk and Σk π.
The induction hypothesis is that an execution Λk Σk with these properties
exists for each k ≤ n − 1.
For the base case, Λ0 Σ0 = hi. This covers 0 reads by π.
For the induction step, we start with Λk Σk , and look for a sequence
of operations γ that we can insert in between Λk and Σk that changes
what π returns in Λk γΣk π from what it returned in Λk γΣk . This is where
perturbability comes in: an object is defined to be perturbable if such a
sequence γ always exists.
Some examples of γ:
• For a snapshot object, let γ write to a component that is not written
to by any of the operations in Σk .
• For a max register, let γ include a bigger write than all the others.
• For a counter, let γ include at least n increments. The same works for
a mod-m counter if m is at least 2n. (Why n increments? With fewer
increments, we can make π return the same value by being sneaky
about when the partial increments represented in Σk are linearized.)
In contrast, historyless objects (including atomic registers) are not per-
turbable: if Σk includes a write that sets the value of the object, no set of
operations inserted before it will change this value. This is good, because we
know that it only takes one atomic register to implement an atomic register.
Assuming that our object is perturbable, now we want to use the existence
of γ to generate our bigger execution Λk+1 Σk+1 . As in the Burns-Lynch
mutex bound [BL93], we will be arguing that γ must include a write to an
object that is not covered by the k delayed writes. Also as in Burns-Lynch,
it turns out that it is not enough just to delay this particular write, because
it might not cover the specific object we want.
Instead, we look for an alternative γ 0 that changes the value of the earliest
object read by π that can be changed. We know that some such γ 0 exists,
CHAPTER 20. LOWER BOUNDS ON PERTURBABLE OBJECTS 184

because γ writes to some such object, so there must be a first place in the
execution of π where the output of an object can change, and there must be
some γ 0 that makes that change. Note however that γ 0 that hits that earliest
object need not be the same as the γ used to demonstrate perturbability,
and indeed it may be that γ 0 is very different from γ—in particular, it may
be much longer.
So now we expand γ 0 = αβδ, where β is the magic write to the uncovered
object, and let Λk+1 = Λk αδ 0 and Σk+1 = βΣk , where δ 0 consists of running
all incomplete operations in α except the one that include β to completion.
We’ve now covered k + 1 distinct objects in Σk . It remains only to show that
the technical condition that any uncovered object that π reads before reading
all the covered objects can’t have its value changed by inserting additional
operations.
Suppose that there is a sequence of operations κ such that Λk+1 κ changes
one of these forbidden uncovered objects. But Λk+1 κ = Λk ακ, and so
γ 00 = ακ changes an object that either (a) can’t be changed because of the
technical condition in the induction hypothesis for k, or (b) changes an object
that π reads before the object covered by β. In the second case, this γ 00
changes an earlier object that γ 0 , contradicting the choice of γ 0 .
It follows that we do in fact manage to cover k + 1 objects while satisfying
the technical condition, and the induction hypothesis holds for k + 1.
We can repeat this step until we’ve covered n − 1 objects. This implies
that there are at least n − 1 objects (the space lower bound), and in the
worst case some reader reads all of them (the step complexity lower bound).
Chapter 21

Restricted-use objects

The Jayanti-Tan-Toueg bound puts a hard floor under the worst-case com-
plexity of almost anything interesting we’d like to do wait-free in a system
that provides only historyless objects as primitives. As with the consensus hi-
erarchy lower bounds, we could interpret this as a reason to demand stronger
primivitives. Or we could look for ways to bypass the JTT bound.
One approach is to modify our target objects so that they are no longer
perturbable. This can be done by limiting their use: a counter or max register
that can only change its value a limited number of times is not perturbable,
because once we hit the limit, there is no perturbing sequence of operations
that we can insert between Λk and Σk in the JTT execution that changes
the value returned by the eventual reader. This observation motivated a
line of work on restricted-use max registers [AACH12] and restricted-use
snapshots [AACHE15] that have polylogarithmic worst-case individual step
complexity assuming a polynomial limit on updates. While restricted-use
objects might not be all that exciting on their own, they in turn have served
as building blocks for implementations of snapshots with polylogarithmic
expected individual step complexity [ACH13] or polylogarithmic amortized
individual step complexity [BHMT20].
In this chapter, we will concentrate on the original restricted-use max
register construction of Aspnes, Attiya, and Censor-Hillel [AACH12], and
its extension to give restricted-use snapshots by Aspnes et al. [AACHE15].

21.1 Max registers

We will start by implementing a restricted-use max register [AACH12], for
which read operation returns the largest value previously written, as opposed

185
CHAPTER 21. RESTRICTED-USE OBJECTS 186

to the last value previously written. So after writes of 0, 3, 5, 2, 6, 11, 7, 1,

9, a read operation will return 11.
In general, max registers are perturbable objects in the sense of the
Jayanti-Tan-Toueg bound, so in the worst case a max-register read will have
to read at least n − 1 distinct atomic registers, giving an n − 1 lower bound on
both step complexity and space. But we can get around this by considering
bounded max registers, which only hold values in some range 0 . . . m − 1.
These are not perturbable because once we hit the upper bound we can no
longer insert new operations to change the value returned by a read. This
allows for a much more efficient implementation (at least in terms of step
complexity) when m is not too big.

21.2 Implementing bounded max registers

This implementation is from a paper by Aspnes, Attiya, and Censor-Hillel [AACH12].
The same paper shows that it is in a certain sense the only possible imple-
mentation of a wait-free restricted max register (see §21.5).
For m = 1, the implementation is trivial: write does nothing and read
always returns 0.
For larger m, we’ll show how to paste together two max registers left and
right with m0 and m1 values together to get a max register r with m0 + m1
values. We’ll think of each value stored in the max register as a bit-vector,
with bit-vectors ordered lexicographically. In addition to left and right, we
will need a 1-bit atomic register switch used to choose between them. The
read procedure is straightforward and is shown in Algorithm 21.1; essentially
we just look at switch, read the appropriate register, and prepend the value
of switch to what we get.

1 procedure read(r)
2 if switch = 0 then
3 return 0 : read(left)
4 else
5 return 1 : read(right)

Algorithm 21.1: Max register read operation

For write operations, we have two somewhat asymmetrical cases depend-

ing on whether the value we are writing starts with a 0 bit or a 1 bit. These
are shown in Algorithm 21.2.
CHAPTER 21. RESTRICTED-USE OBJECTS 187

1 procedure write(r, 0x)

2 if switch = 0 then
3 write(left, x)

4 procedure write(r, 1x)

5 write(right, x)
6 switch ← 1
Algorithm 21.2: Max register write operations

The intuition is that the max register is really a big tree of switch
variables, and we store a particular bit-vector in the max register by setting
to 1 the switches needed to make read follow the path corresponding to
that bit-vector. The procedure for writing 0x tests switch first, because once
switch gets set to 1, any 0x values are smaller than the largest value, and we
don’t want them getting written to left where they might confuse particularly
slow readers into returning a value we can’t linearize. The procedure for
writing 1x sets switch second, because (a) it doesn’t need to test switch, since
1x always beats 0x, and (b) it’s not safe to send a reader down into right
until some value has actually been written there.
It’s easy to see that read and write operations both require exactly
one operation per bit of the value read or written. To show that we get
linearizability, we give an explicit linearization ordering (see the paper for a
full proof that this works):

1. All operations that read 0 from switch go in the first pile.

(a) Within this pile, we sort operations using the linearization ordering
for left.

2. All operations that read 1 from switch or write 1 to switch go in the

second pile, which is ordered after the first pile.

(a) Within this pile, operations that touch right are ordered using
the linearization ordering for right. Operations that don’t (which
are the “do nothing” writes for 0x values) are placed consistently
with the actual execution order.

To show that this gives a valid linearization, we have to argue first that
any read operation returns the largest earlier write argument and that we
don’t put any non-concurrent operations out of order.
CHAPTER 21. RESTRICTED-USE OBJECTS 188

For the first part, any read in the 0 pile returns 0 : read(left), and
read(left) returns (assuming left is a linearizable max register) the largest
value previously written to left, which will be the largest value linearized
before the read, or the all-0 vector if there is no such value. In either case
we are happy. Any read in the 1 pile returns 1 : read(right). Here we have
to guard against the possibility of getting an all-0 vector from read(right)
if no write operations linearize before the read. But any write operation
that writes 1x doesn’t set switch to 1 until after it writes to right, so no read
operation ever starts read(right) until after at least one write to right has
completed, implying that that write to right linearizes before the read from
right. So in all the second-pile operations linearize as well.

21.3 Encoding the set of values

If we structure our max register as a balanced tree of depth k, we are
essentially encoding the values 0 . . . 2k −1 in binary, and the cost of performing
a read or write operation on an m-valued register is exactly k = dlg me. But
if we are willing to build an unbalanced tree, any prefix code will work.
The paper describes a method of building a max register where the cost
of each operation that writes or reads a value v is O(log v). The essential
idea is to build a tree consisting of a rightward path with increasingly large
left subtrees hanging off of it, where each of these left subtrees is twice as big
as the previous. This means that after following a path encoded as 1k 0, we
hit a 2k -valued max register. The value returned after reading some v 0 from
this max register is v 0 + (2k − 1), where the 2k − 1 term takes into account
all the values represented by earlier max registers in the chain. Formally,
this is equivalent to encoding values using an Elias gamma code, tweaked
slightly by changing the prefixes from 0k 1 to 1k 0 to get the ordering right.

21.4 Unbounded max registers

While the unbalanced-tree construction could be used to get an unbounded
max register, it is possible that read operations might not terminate (if
enough writes keep setting 1 bits on the right path before the read gets to
them) and for very large values the cost even of terminating reads becomes
higher than what we can get out of a snapshot.
Here is the snapshot-based method: if each process writes its own contri-
bution to the max register to a single-writer register, then we can read the
max register by taking a snapshot and returning the maximum value. (It is
CHAPTER 21. RESTRICTED-USE OBJECTS 189

not hard to show that this is linearizable.) This gives an unbounded max
register with read and write cost O(n). So by choosing this in preference
to the balanced tree when m is large, the cost of either operation on a max
register is min (dlg me , O(n)).
We can combine this with the unbalanced tree by terminating the right
path with a snapshot-based max register. This gives a cost for reads and
writes of values v of O(min(log v, n)).

21.5 Lower bound

The min(dlg me , n − 1) cost of a max register read turns out to be exactly
optimal. Intuitively, we can show by a covering argument that once some
process attempts to write to a particular atomic register, then any subsequent
writes convey no additional information (because they can be overwritten by
the first delayed write)—so in effect, no algorithm can use get more than
one bit of information out of each atomic register. But as always, turning
this intuition into an actual proof requires a bit more work.
We will consider solo-terminating executions in which n − 1 writers do
any number of max-register writes in some initial prefix Λ, followed by a
single max-register read π by process pn . Let T (m, n) be the optimal reader
cost for executions with this structure with m values, and let r be the first
register read by process pn , assuming it is running an algorithm optimized
for this class of executions (we do not even require it to be correct for other
executions).
We are now going split up our set of values based on which will cause a
write operation to write to r. Let Sk be the set of all sequences of writes that
only write values ≤ k. Let t be the smallest value such that some execution
in St writes to r (there must be some such t, or our reader can omit reading
r, which contradicts the assumption that it is optimal).

Case 1 Since t is smallest, no execution in St−1 writes to r. If we restrict

writes to values ≤ t − 1, we can omit reading r, giving T (t, n) ≤
T (m, n) − 1 or T (m, n) ≥ T (t, n) + 1.

Case 2 Let α be some execution in St that writes to r.

• Split α as α0 δβ where δ is the first write to r by some process pi .

• Construct a new execution α0 η by letting all the max-register
writes except the one performing δ finish.
CHAPTER 21. RESTRICTED-USE OBJECTS 190

• Now consider any execution α0 ηγδ, where γ is any sequence of

max-register writes with values ≥ t that excludes pi and pn . Then
pn always sees the same value in r following these executions,
but otherwise (starting after α0 η) we have an (n − 1)-process
max-register with values t through m − 1.
• Omit the read of r again to get T (m, n) ≥ T (m − t, n − 1) + 1.

We’ve shown the recurrence T (m, n) ≥ mint (max(T (t, n), T (m−t, n)))+1,
with base cases T (1, n) = 0 and T (m, 1) = 0. The solution to this recurrence
is exactly min(dlg me , n − 1), which is the same, except for a constant factor
on n, as the upper bound we got by choosing between a balanced tree for
small m and a snapshot for m ≥ 2n−1 . For small m, the recursive split we
get is also the same as in the tree-based algorithm: call the r register switch
and you can extract a tree from whatever algorithm somebody gives you. So
this says that the tree-based algorithm is (up to choice of the tree) essentially
the unique optimal bounded max register implementation for m ≤ 2n−1 .
It is also possible to show lower bounds on randomized implementations
of max registers and other restricted-use objects. See [AACH12, ACAH16,
HK14] for examples.

21.6 Max-register snapshots

With some tinkering, it’s possible to extend the max-register construction
to get an array of max registers that supports snapshots. The description
in this section follows [AACHE15], with some updates to fix a bug noted in
the original paper in an erratum published by the authors [AACHE18].
Formally, a max array is an object a that supports an operation
write(a, i, v) that sets a[i] ← max(v, a[i]), and an operation read(a) that
returns a snapshot of all components of the array. The first step in building
this beast is to do it for two components. The resulting 2-component max
array can then be used as a building block for larger max arrays and for
fast restricted-used snapshots in general.
A k × ` max array a is one that permits values in the range 0 . . . k − 1
in a[0] and 0 . . . ` − 1 in a[1]. We think of a[0] as the head of the max array
and a[1] as the tail. We’ll show how to construct such an object recursively
from smaller objects of the same type, analogous to the construction of an
m-valued max register (which we can think of as a m × 1 max array). The
idea is to split head into two pieces left and right as before, while representing
tail as a master copy stored in a max register at the top of the tree plus
CHAPTER 21. RESTRICTED-USE OBJECTS 191

cached copies at every internal node. These cached copies are updated by
readers at times carefully chosen to ensure linearizability.
The base of the construction is an `-valued max register r, used directly
as a 1 × ` max array; this is the case where the head component is trivial and
we only need to store a.tail = r. Here calling write(a, 0, v) does nothing,
while write(a, 1, v) maps to write(r, v), and read(a) returns h0, read(r)i.
For larger values of k, paste a kleft × ` max array left and a kright × ` max
array right together to get a (kleft + kright ) × ` max array. This construction
uses a switch variable as in the basic construction, along with an `-valued
max register tail that is used to store the value of a[1].
Calls to write(a, 0, v) and read(a) follow the structure of the correspond-
ing operations for a simple max register, with some extra work in read to
make sure that the value in tail propagates into left and right as needed to
ensure the correct value is returned.
A call to write(a, 1, v) operation writes tail directly, and then calls
read(a) to propagate the new value as well.1
Pseudocode is given in Algorithm 21.3.
The individual step complexity of each operation is easily computed.
Assuming a balanced tree, write(a, 0, v) takes exactly dlg ke steps, while
write(a, 1, v) costs exactly dlg `e steps plus the cost of read(a). Read
operations are more complicated. In the worst case, we have two reads of
a.tail and a write to a.right[1] at each level, plus up to two operations on
a.switch, for a total cost of at most (3dlg ke − 1)(dlg `e + 2) = O(log k log `)
steps. This dominates other costs in write(a, 1, v), so the asymptotic cost
of both write and read operations is O(log k log `).
In the special case where k = `, both writes and reads have their step
complexities squared compared to a single-component k-valued max register.

21.6.1 Linearizability
In broad outline, the proof of linearizability follows the proof for a simple
max register. But as with snapshots, we have to show that the ordering of
the head and tail components are consistent.
The key observation is the following lemma.

Lemma 21.6.1. Fix some execution of a max array a implemented as in

Algorithm 21.3. Suppose this execution contains a read(a) operation πleft
1
This call to read(a) was omitted in the original published version of the algo-
rithm [AACHE15], but was added in an erratum by the authors [AACHE18]. Without it,
the implementation can violate linearizability in some executions.
CHAPTER 21. RESTRICTED-USE OBJECTS 192

1 procedure write(a, i, v)
2 if i = 0 then
3 if v < kleft then
4 if a.switch = 0 then
5 write(a.left, 0, v)
6 else
7 write(a.right, 0, v − kleft )
8 a.switch ← 1
9 else
10 write(a.tail, v)
11 read(a)

12 procedure read(a)
13 x ← read(a.tail)
14 if a.switch = 0 then
15 write(a.left, 1, x)
16 return read(a.left)
17 else
18 x ← read(a.tail)
19 write(a.right, 1, x)
20 return hkleft , 0i + read(a.right)

Algorithm 21.3: Recursive construction of a 2-component max array

CHAPTER 21. RESTRICTED-USE OBJECTS 193

that returns vleft from a.left and a read(a) operation πright that returns vright
from a.right. Then vleft [1] ≤ vright [1].

Proof. Both vleft [1] and vright [1] are values that were previously written to
their respective max arrays by read(a) operations (such writes necessarily
exist because any process that reads a.left or a.right writes a.left[1] or a.right[1]
first). From examining the code, we have that any value written to a.left[1]
was read from a.tail before a.switch was set to 1, while any value written to
a.right[1] was read from a.tail after a.switch was set to 1. Since max-register
reads are non-decreasing, we have than any value written to a.left[1] is less
than or equal to any value written to a.right[1], proving the claim.

The rest of the proof is tedious but straightforward: we linearize the

read(a) and write(a[0]) operations as in the max-register proof, then fit
the write(a[1]) operations in based on the tail values of the reads. The full
result is:

Theorem 21.6.2. If a.left and a.right are linearizable max arrays, and a.tail
is a linearizable max register, then Algorithm 21.3 implements a linearizable
max array.

It’s worth noting that the same unbalanced-tree construction used in

§§21.3 and 21.4 can be used here as well. This makes the step complexity for
read(a) scale as O(log v[0] log v[1]), where v is the value returned. For writes
the step complexity may depend in a complicated way on what values are
being written and to which side, but in the worst case, it is O(log v[0] log v[1]),
where v is the value in the register when the write finishes. (This is a
consequence of the embedded read(a) in write(a, 1, v).)

21.7 Restricted-use snapshots

To build an ordinary snapshot object from 2-component max arrays, we
construct a balanced binary tree in which each leaves holds a pointer to
an individual snapshot element and each internal node holds a pointer to a
partial snapshot containing all of the elements in the subtree of which it is
the root. The pointers themselves are non-decreasing indices into arrays of
values that consist of ordinary (although possibly very wide) atomic registers.
When a process writes a new value to its component of the snapshot
object, it increases the pointer value in its leaf and then propagates the new
value up the tree by combining together partial snapshots at each step, using
2-component max arrays to ensure linearizability. The resulting algorithm
CHAPTER 21. RESTRICTED-USE OBJECTS 194

is similar in many ways to the lattice agreement procedure of Inoue et

al. [IMCT94] (see §19.3.5), except that it uses a more contention-tolerant
snapshot algorithm than double collects and we allow processes to update
their values more than once. It is also similar to the f -array construction
of Jayanti [Jay02] for efficient computation of array aggregates (sum, min,
max, etc.) using LL/SC, the main difference being that because the index
values are non-decreasing, max arrays can substitute for LL/SC.
Each node in the tree except the root is represented by one component
of a 2-component max array that we can think of as being owned by its
parent, with the other component being the node’s sibling in the tree. To
propagate a value up the tree, at each level the process takes a snapshot
of the two children of the node and writes the sum of the indices to the
node’s component in its parent’s max array (or to an ordinary max register
if we are at the root). Before doing this last write, a process will combine
the partial snapshots from the two child nodes and write the result into
a separate array indexed by the sum. In this way any process that reads
the node’s component can obtain the corresponding partial snapshot in a
single register operation. At the root this means that the cost of obtaining
a complete snapshot is dominated by the cost of the max-register read, at
O(log v), where v is the number of updates ever performed.
A picture of this structure, adapted from the proceedings version of [AACHE15],
appears in Figure 21.1. The figure depicts an update in progress, with red
values being the new values written as part of the update. Only some of the
tables associated with the nodes are shown.
The cost of an update is dominated by the O(log n) max-array operations
needed to propagate the new value to the root. This takes O(log2 v log n)
steps. Here v can be taken to be the number of update operations, which
controls the maximum value on either side of the 2-component max arrays.
The linearizability proof is trivial: linearize each update by the time at
which a snapshot containing its value is written to the root (which necessarily
occurs within the interval of the update, since we don’t let an update finish
until it has propagated its value to the top), and linearize reads by when
they read the root. This immediately gives us an O(log3 n) implementa-
tion—as long as we only want to use it polynomially many times—of anything
we can build from snapshot, including counters, generalized counters, and
(by [AH90b, AM93]) any other object whose operations all commute with or
overwrite each other in a static pattern.
CHAPTER 21. RESTRICTED-USE OBJECTS 195

cms
5 cmr
bmr
br
ar
0 5 a

ms
0 0 3 3 mr
m

c
b 1 2
a

s
m r

Figure 21.1: Snapshot from max arrays; see also [AACHE15, Fig. 2]

21.7.1 Randomized and amortized snapshots

Randomization can eliminate the need to limit the number of times the
snapshot is used. The JTT bound still applies, so there will be occasional
expensive operations, but we can spread these out randomly so that any
particular operation has low expected cost. This gives a cost of O(log3 n) ex-
pected steps for an unrestricted snapshot. This result was claimed by Aspnes
and Censor-Hillel [ACH13], although the construction given in the conference
version of this paper (a) is based on the original, uncorrected version of the
max array from [AACHE15], (b) incorrectly computes the running time of
the algorithm, and (c) has a rather rococo proof of linearizability that is
suspect in various additional ways. A submitted journal version of the same
paper [ACH18] may be more convincing.
Convincing or not, this result is largely dominated by a more recent result
by Baig et al. [BHMT20] that gives a deterministic snapshot implementation
with O(log n) amortized step complexity.
Note that none of these algorithms contradict the JTT lower bound: in
the worst case, each will have operations that take Ω(n) steps. But the hope
is that these operations are few, and in the amortized case, paid for by many
cheap operations. Also, even though we may beat JTT most of the time,
other lower bounds may still apply; see for example [ACAH16, HK14].
CHAPTER 21. RESTRICTED-USE OBJECTS 196

For some specific applications, further improvements are possible. The

original max-registers paper [AACH12] gives an implementation of counters
using a similar tree construction with only max registers that costs only
O(log2 n) for increments; here the trick is to observe that counters that can
only be incremented by one are much easier to make linearizable, because
there is no possibility of seeing an intermediate value that couldn’t be present
in a sequential execution. Whether it is possible to improve the cost of
snapshot is an open problem.
Chapter 22

Common2

Last updated 2019. Some material may be out of date.

The common2 class, defined by Afek, Weisberger, and Weisman [AWW93]

consists of all read-modify-write objects where the modify functions either (a)
all commute with each other or (b) all overwrite each other. We can think of it
as the union of two simpler classes, the set of read-modify-write objects where
all update operations commute, called commuting objects [AW99]; and
the set of read-modify-write objects where all updates produce a value that
doesn’t depend on the previous state, called historyless objects [FHS98]).
From §18.1.2, we know that both commuting objects and historyless
objects have consensus number at most 2, and that these objects have con-
sensus number exactly 2 provided they supply at least one non-trivial update
operation. The main result of Afek et al. [AWW93] is that commuting and
historyless objects can all be implemented from any object with consensus
number 2, even in systems with more than 2 processes. This gives a com-
pleteness result analogous to completeness results in complexity theory: any
non-trivial common2 object can be used to implement any other common2
object.
The common2 conjecture was that common2 objects could also im-
plement any object with consensus number 2, This is now known to be
false [AEG16].
The main result in the paper has two parts, reflecting the two parts of
the common2 class: a proof that 2-process consensus plus registers is enough
to implement all commuting objects (which essentially comes down to build
a generalized fetch-and-add that returns an unordered list of all preceding
operations); and a proof that 2-process consensus plus registers is enough to
implement all overwriting objects (which is equivalent to showing that we can

197
CHAPTER 22. COMMON2 198

implement swap objects). The construction of the generalized fetch-and-add

is pretty nasty, so we’ll concentrate on the implementation of swap objects.
We will also skip the swap implementation in [AWW93], and instead describe,
in §§22.3 and 22.4, a simpler (though possibly less efficient) algorithm from
a later paper by Afek, Morrison, and Wertheim [AMW11]. Before we do
this, we’ll start with some easier results from the older paper, including an
implementation of n-process test-and-set from 2-process consensus. This
will show that anything we can do with test-and-set we can do with any
common2 object.

22.1 Test-and-set and swap for two processes

The first step is to get test-and-set.
Algorithm 22.1 shows how to turn 2-process consensus into 2-process
test-and-set. The idea is that whoever wins the consensus protocol wins the
test-and-set. This is linearizable, because if I run TAS2 before you do, I win
the consensus protocol by validity.

1 procedure TAS2()
2 if Consensus2(myId) = myId then
3 return 0
4 else
5 return 1

Algorithm 22.1: Building 2-process TAS from 2-process consensus

Once we have test-and-set for two processes, we can easily get one-shot
swap for two processes. The trick is that a one-shot swap object always
returns ⊥ to the first process to access it and returns the other process’s value
to the second process. We can distinguish these two roles using test-and-set
and add a register to send the value across. Pseudocode is in Algorithm 22.2.

22.2 Building n-process TAS from 2-process TAS

To turn the TAS2 into full-blown n-process TAS, start by staging a tournament
along the lines of [PF77] (§17.4.1.2). Each process walks up a tree of nodes,
and at each node it attempts to beat every process from the other subtree
using a TAS2 object (we can’t just have it fight one process, because we
don’t know which one process will have won the other subtree, and our TAS2
CHAPTER 22. COMMON2 199

1 procedure swap(v)
2 a[myId] = v
3 if TAS2() = 0 then
4 return ⊥
5 else
6 return a[¬myId]

Algorithm 22.2: Two-process one-shot swap from TAS

objects may only work for two specific processes). A process drops out if
it ever sees a 1. We can easily show that at most one process leaves each
subtree with all zeros, including the whole tree itself.
Unfortunately, this process does not give a linearizable test-and-set object.
It is possible that p1 loses early to p2 , but then p3 starts (elsewhere in the
tree) after p1 finishes, and races to the top, beating out p2 . To avoid this,
we can follow [AWW93] and add a gate bit that locks out latecomers.1
The resulting construction looks something like Algorithm 22.3. This
gives a slightly different interface from straight TAS; instead of returning 0
for winning and 1 for losing, the algorithm returns ⊥ if you win and the id
of some process that beats you if you lose.2 It’s not hard to see that this
gives a linearizable test-and-set after translating the values back to 0 and 1
(the trick for linearizability is that any process that wins saw an empty gate,
and so started before any other process finished). It also sorts the processes
into a rooted tree, with each process linearizing after its parent (this latter
claim is a little trickier, but basically comes down to a loser linearizing after
the process that defeated it either on gate or on one of the TAS2 objects).
This algorithm is kind of expensive: the losers that drop out early are
relatively lucky, but the winning process has to win a TAS2 against everybody,
for a total of Θ(n) TAS operations. We can reduce the cost to O(log n) if
our TAS2 objects allow arbitrary processes to execute them. This is done,
for example, in the RatRace test-and-set implementation of Alistarh et
al. [AAG+ 10], using a randomized implementation of TAS2 due to Tromp
and Vitányi [TV02] (see §24.5.2).
1
The original version of this trick is from an earlier paper [AGTV92], where the gate
bit is implemented as an array of single-writer registers.
2
Note that this process may also be a loser, just one that made it further up the tree
than you did. We can’t expect to learn the id of the ultimate winner, because that would
solve n-process consensus.
CHAPTER 22. COMMON2 200

1 procedure compete(i)
// check the gate
2 if gate 6= ⊥ then
3 return gate
4 gate ← i
// Do tournament, returning id of whoever I lose to
5 node ← leaf for i
6 while node 6= root do
7 for each j whose leaf is below sibling of node do
8 if TAS2(t[i, j]) = 1 then
9 return j

10 node ← node.parent
// I win!
11 return ⊥
Algorithm 22.3: Tournament algorithm with gate

22.3 Obstruction-free swap from test-and-set

We’ll start by describing the “strawman algorithm” from the AMW paper.
This is presented by the authors as a stepping-stone to their real algorithm,
which we will describe below in §22.4.
The code is given in Algorithm 22.4. This implements a swap object that
is linearizable but not wait-free.
This algorithm uses two infinite arrays s and t of test-and-set objects
and an infinite array r of atomic registers. The si objects are essentially
being used to implement a fetch-and-increment, and if we have a fetch-and-
increment lying around we can replace the loop at Line 4 with an operation
on that object instead. The ri registers record values to return. The ti
registers implement a block/pass mechanism where a later process can force
an earlier process to try again if it didn’t record its value in time. This
solves the problem of a process going to sleep after acquiring a particular
slot i from the fetch-and-increment but before writing down a value that
somebody else can use.
The algorithm is obstruction-free, because in any reachable configuration,
only finitely many test-and-sets have been accessed, so there is some value i
with sj = tj = 0 for all j ≥ i. A process running in isolation will eventually
hit one of these slots, win both test-and-sets, and return.
CHAPTER 22. COMMON2 201

1 procedure swap(v)
2 i←0
3 while true do
// Look for a starting point
4 while TAS(si ) = 1 do
5 i←i+1
6 vi ← v
// Check if we’ve been blocked
7 if TAS(ti ) = 0 then
// We win, find our predecessor
8 for j ← i − 1 down to 0 do
9 if TAS(tj ) = 1 then
// Use this value
10 return vj

// Didn’t find anybody, we are first

11 return ⊥
12 else
// Pick a new start and try again

Algorithm 22.4: Obstruction-free swap from test-and-set

CHAPTER 22. COMMON2 202

For linearizability, the value of i when each operation returns gives an

obvious linearization ordering. This ordering is consistent with the observed
history, because if I finish with value i1 before you start, then at the time
that I finish all sj for j ≤ i1 have sj = 1. So you can’t win any of them, and
get a slot i2 > i1 . But we still have to show that the return values make
sense.
Consider some swap operation π.
Suppose that π starts at position i and wins every tj down to position
k, where it loses. Then no other operation wins any tj with k < j < i, so
there is no process that leaves with any slot between k and i. In addition,
the operation π 0 that did win tk must have taken slot k in Line 7, because
any other process would have needed to win tk+1 before attempting to win
tk . So π 0 linearizes immediately before π, which is good, because π returns
the value vk that π 0 wrote before it won tk .
Alternatively, suppose that π never loses tj for any j ≤ i. Then no other
operation takes a slot less than i, and π linearizes first. In this case, it must
return ⊥, which it does.

22.4 Wait-free swap from test-and-set

Now we want to make the strawman algorithm wait-free. The basic idea
is similar: we will have an ordered collection of test-and-set objects, and a
process will move right until it can capture one that determines its place
in the linearization ordering, and then it will move left to block any other
processes from taking an earlier place unless they have already written out
their values. To avoid starvation, we assign a disjoint collection of test-and-
set objects to each operation, so that every operation eventually wins one of
its own test-and-sets. Unfortunately this only works if we make the ordering
dense, so that between each pair of test-and-sets there are infinitely many
other test-and-sets.
AMW do this in terms of a binary tree, but I find it easier to think of the
test-and-sets as being indexed by dyadic rationals strictly between 0 and 1.3
The idea is that the i-th operation to start executing the swap object will use
test-and-sets tq where q = k/2i for all odd k in the range 1 . . . 2i − 1. In order
to avoid having to check the infinitely many possible values smaller than q,
we will use two auxiliary objects: a readable fetch-and-increment maxDepth
that hands out denominators and tracks the largest denominator used so far,
3
The two representations are isomorphic: make each value k/2q be the parent of
k/2 ± 1/2q+1 .
q
CHAPTER 22. COMMON2 203

and a max register accessed that keeps track of the largest position accessed
so far.
AMW implement accessed using a snapshot, which we will do as well to
avoid complications from trying to build a max register out of an infinitely
deep tree.4 Note that AMW don’t call this data structure a max register,
but we will, because we like max registers.
Code for the swap procedure is given in Algorithm 22.5.
To show Algorithm 22.5 works, we need the following technical lemma,
which, among other things, implies that node 1 − 2depth is always available
to be captured by the process at depth depth. This is essentially just a
restatement of Lemma 1 from [AMW11].
Lemma 22.4.1. For any x = k/2q , where k is odd, no process attempts to
capture any y ∈ [x, x + 1/2q ) before some process writes x to accessed.
Proof. Suppose that the lemma fails, let y = `/2r be the first node captured
in violation of the lemma, and let x = k/2q be such that y ∈ [x, x + 1/2q )
but x has not been written to accessed when y is captured. Let p be the
process that captures y.
Now consider y 0 = x − 1/2r , the last node to the left of x at the same
depth as y. Why didn’t p capture y 0 ?
One possibility is that some other process p0 blocked y 0 during its return
phase. This p0 must have captured a node z > y 0 . If z > y, then p0 would
have blocked y first, preventing p from capturing it. So y 0 < z < y.
The other possibility is that p never tried to capture y 0 , because some
other process p0 wrote some value z > y 0 to accessed first. This value z must
also be less than y (or else p would not have tried to capture y).
In both cases, there is a process p0 that captures a value z with y 0 < z < y,
before p captures y and thus before anybody writes x to accessed.
Since y 0 < x and y 0 < z, either y 0 < z < x or y 0 < x < z. In the first case,
z ∈ [y 0 , y 0 + 1/2r ) is captured before y 0 is written to accessed. In the second
case z ∈ [x, x + 1/2q ) is captured before x is written to accessed. Either
way, y is not the first capture to violate the lemma, contradicting our initial
assumption.
Using Lemma 22.4.1, it is straightforward to show that Algorithm 22.5 is
wait-free. If I get q for my value of depth, then no process will attempt to
4
The issue is not so much that we can’t store arbitrary dyadics, since we can encode them
using an order-preserving prefix-free code, but that, without some sort of helping mechanism,
a read running concurrently with endlessly increasing writes (e.g. 1/2, 3/4, 7/8, . . . ) might
not be wait-free. Plus as soon as the denominator exceeds 2n , which happens after only n
calls to swap, O(n)-step snapshots are cheaper anyway.
CHAPTER 22. COMMON2 204

1 procedure swap(v)
// Pick a new row just for me
2 depth ← fetchAndIncrement(maxDepth)
// Capture phase
3 repeat
// Pick leftmost node in my row greater than
accessed
n o
4 cap ← min x x = k/2depth for odd k, x > accessed
// Post my value
5 reg[cap] ← v
// Try to capture the test-and-set
6 win ← TAS(tst[cap]) = 0
7 writeMax(accessed, cap)
8 until win
// Return phase
// Max depth reached by anybody left of cap
9 maxPreviousDepth ← read(maxDepth)
10 ret ← cap
// Block previous nodes until we find one we can take
11 repeat
12 ret ← max {x = k/2q | q ≤ maxPreviousDepth, k odd, x < ret}
13 if ret < 0 then
14 return ⊥
15 until TAS(tst[ret]) = 1
16 return reg[ret]
Algorithm 22.5: Wait-free swap from test-and-set [AMW11]
CHAPTER 22. COMMON2 205

capture any y in [1 − 2q , 1) before I write 1 − 2q to accessed. But this means

that nobody can block me from capturing 1 − 2q , because processes can only
block values smaller than the one they already captured. I also can’t get
stuck in the return phase, because there are only finitely many values with
denominator less than 2maxPreviousDepth .
It remains to show that the implementation is linearizable. The obvious
linearization ordering is given by sorting each operation i by its captured
node cap. Linearizability requires then that if we imagine a directed graph
containing an edge ij for each pair of operations i and j such that i captures
capi and returns reg[capj ], then this graph forms a path that corresponds to
this linearization ordering.
Since each process only returns one value, it trivially holds that each
node in the graph has out-degree at most 1. For the in-degree, suppose that
we have operations i, j, and k with capi < capj < capk such that j and k
both return reg[capi ]. Before k reaches tst[capi ], it must first capture all
the test-and-sets between capi and capk that have depth less than or equal
to maxPreviousDepthk . This will include tst[capj ], because j must write to
maxDepth before doing anything, and this must occur before k starts the
return phase if j sees a value of accessed that is less that capk .
A similar argument show that there is at most one process that returns
⊥; this implies that there is at most one process with out-degree 0.
So now we have a directed graph where every process has in-degree and
out-degree at most one, which implies that each weakly-connected component
will be a path. But each component will also have exactly one terminal
node with out-degree 0. Since there is only one such node, there is only one
component, and the entire graph is a single path. This concludes the proof
of linearizability.

22.5 Implementations using stronger base objects

The terrible step complexity of known wait-free implementations of Com-
mon2 objects like swap or fetchAndIncrement from 2-process consensus
objects and registers has led to work on finding better implementations
assuming stronger base objects. Using load-linked/store-conditional, Ellen
and Woelfel [EW13] provide implementations of several Common2 objects, in-
cluding fetchAndIncrement, fetchAndAdd, and swap that all have O(log n)
individual step complexity.5 This is know to be optimal due to a lower bound
5
What they actually implement is the ability to do fetch-and-f , where f is any binary
associative function, using an object they call an aggregator. Each of these objects is
CHAPTER 22. COMMON2 206

of Jayanti [Jay98].
The lower bound applies a fortiori to the case where we don’t have
LL/SC or CAS and have to rely on 2-process consensus objects. But it’s not
out of the question that there is a matching upper bound in this case.

obtained by choosing an appropriate f .

Chapter 23

Randomized consensus and

test-and-set

We’ve seen that we can’t solve consensus in an asynchronous system message-

passing or shared-memory system with one crash failure [FLP85, LAA87],
but that the problem becomes solvable using failure detectors [CT96]. An
alternative that also allows us to solve consensus is to allow the processes
to use randomization, by providing each process with a local coin that can
generate random values that are immediately visible only to that process.
The resulting randomized consensus problem replaces the termination
requirement with probabilistic termination: all processes terminate with
probability 1. The agreement and validity requirements remain the same.
In this chapter, we will describe how randomization interacts with the
adversary, give a bit of history of randomized consensus, and then concen-
trate on recent algorithms for randomized consensus and the closely-related
problem of randomized test-and-set. Much of the material in this chapter is
adapted from notes for a previous course on randomized algorithms [Asp11]
and a few of my own papers [Asp12b, AE11, Asp12a].

23.1 Role of the adversary in randomized algo-

rithms
Because randomized processes are unpredictable, we need to become a
little more sophisticated in our handling of the adversary. As in previous
asynchronous protocols, we assume that the adversary has control over
timing, which we model by allowing the adversary to choose at each step
which process performs the next operation. But now the adversary may do

207
CHAPTER 23. RANDOMIZED CONSENSUS AND TEST-AND-SET208

so based on knowledge of the state of the protocol and its past evolution.
How much knowledge we give the adversary affects its power. Several classes
of adversaries have been considered in the literature; ranging from strongest
to weakest, we have:
1. An adaptive adversary. This adversary is a function from the state
of the system to the set of processes; it can see everything that has
happened so far (including coin-flips internal to processes that have not
yet been revealed to anybody else), but can’t predict the future. It’s
known that an adaptive adversary can force any randomized consensus
protocol to take Θ(n2 ) total steps [AC08]. The adaptive adversary
is also called a strong adversary following a foundational paper of
Abrahamson [Abr88].
2. An intermediate adversary or weak adversary [Abr88] is one
that limits the adversary’s ability to observe or control the system
in some way, without completely eliminating it. For example, a
content-oblivious adversary [Cha96] or value-oblivious adver-
sary [Aum97] is restricted from seeing the values contained in registers
or pending write operations and from observing the internal states
of processes directly. A location-oblivious adversary [Asp12b] can
distinguish between values and the types of pending operations, but
can’t discriminate between pending operations based one which register
they are operating on. These classes of adversaries are modeled by
imposing an equivalence relation on partial executions and insisting
that the adversary make the same choice of processes to go next in
equivalent situations. Typically they arise because somebody invented
a consensus protocol for the oblivious adversary (described below) and
then looked for the next most powerful adversary that still let the
protocol work.
Weak adversaries often allow much faster consensus protocols than
adaptive adversaries. Each of the above adversaries permits consensus
to be achieved in O(log n) expected individual work using an appropri-
ate algorithm. But from a mathematical standpoint, weak adversaries
are a bit messy, and once you start combining algorithms designed for
different weak adversaries, it’s natural to move all the way down to the
weakest reasonable adversary, the oblivious adversary.
3. A oblivious adversary has no ability to observe the system at all;
instead, it fixes a sequence of process ids in advance, and at each step
the next process in the sequence runs.
CHAPTER 23. RANDOMIZED CONSENSUS AND TEST-AND-SET209

We will describe below a protocol that guarantees O(log log n) expected

individual work for an oblivious adversary. It is not known whether
this is optimal; in fact, is is consistent with the best known lower bound
(due to Attiya and Censor [AC08]) that consensus can be solved in
O(1) expected individual steps against an oblivious adversary.
Each of these adversaries is defined based on choosing steps of particular
objects, with particular constraints on knowledge based on the states of those
objects. This interacts badly with abstractions like linearizability: an adver-
sary might be able to play games with the internals of an implementation
of an object that allows it more power than it would have with an actual
sequential version of the object. So even though linearizable implementations
are indistinguishable from sequential objects for deterministic protocols, for
randomized protocols they can give very different results for both adaptive
and oblivious adversaries [GHW11]; and in the specific case of consensus,
it can be shown that there are randomized consensus protocols that termi-
nate with probability 1 against an adaptive adversary when implemented
with atomic registers, but fail to terminate with nonzero probability when
implemented using an arbitrary linearizable implementation [HHT20].
These results don’t necessarily imply the failure of any specific consensus
protocol implemented using a specific atomic register simulation, but they
do justify suspicion. The easiest way to deal with this suspicion is to assume
that our atomic registers are, in fact, atomic, so that’s what we will do here.

23.2 History
The use of randomization to solve consensus in an asynchronous system
with crash failures was proposed by Ben-Or [BO83] for a message-passing
model. Chor, Israeli, and Li [CIL94] gave the first wait-free consensus
protocol for a shared-memory system, which assumed a particular kind of
weak adversary. Abrahamson [Abr88] defined strong and weak adversaries
and gave the first wait-free consensus
 2  protocol for a strong adversary; its
expected step complexity was Θ 2n . After failing to show that exponential
time was necessary, Aspnes and Herlihy [AH90a] showed how to do consensus
in O(n4 ) total step complexity, a value that was soon reduced to O(n2 log n)
by Bracha and Rachman [BR91]. This remained the best known bound for
the strong-adversary model until Attiya and Censor [AC08] showed matching
Θ(n2 ) upper and lower bounds on total step complexity. A later paper by
Aspnes and Censor [AC09] showed that it was also possible to get an O(n)
bound on individual step complexity.
CHAPTER 23. RANDOMIZED CONSENSUS AND TEST-AND-SET210

For weak adversaries, the best known upper bound on individual step
complexity was O(log n) for a long time [Cha96, Aum97, Asp12b], with
an O(n) bound on total step complexity for some models [Asp12b]. More
recent work has lowered the bound to O(log log n), under the assumption of
an oblivious adversary [Asp12a]. No non-trivial lower bound on expected
individual step complexity is known, although there is a known lower bound
on the distribution of of the individual step complexity [ACH10].
In the following sections, we will concentrate on the more recent weak-
adversary algorithms. These have the advantage of being fast enough that
one might reasonably consider using them in practice, assuming that the
weak-adversary assumption does not create trouble, and they are also require
less probabilistic machinery to analyze than the strong-adversary algorithms.

23.3 Reduction to simpler primitives

To show how to solve consensus using randomization, it helps to split the
problem in two: we will first see how to detect when we’ve achieved agreement,
and then look at how to achieve agreement.

23.3.1 Adopt-commit objects

Most known randomized consensus protocols have a round-based structure
that alternates between generating and detecting agreement. Gafni [Gaf98]
proposed adopt-commit protocols as a tool for detecting agreement, and
these protocols were later abstracted as adopt-commit objects [MRRT08,
AGGT09]. The version described here is largely taken from [AE11], which
shows bounds on the complexity of adopt-commit objects.
An adopt-commit object supports a single operation, AdoptCommit (u),
where u is an input from a set of m values. The result of this operation is an
output of the form (commit, v) or (adopt, v), where the second component is
a value from this set and the first component is a decision bit that indicates
whether the process should decide value v immediately or adopt it as its
preferred value in later rounds of the protocol.
The requirements for an adopt-commit object are the usual requirements
of validity and termination, plus:

1. Coherence. If the output of some operation is (commit, v), then every

output is either (adopt, v) or (commit, v).

2. Convergence. If all inputs are v, all outputs are (commit, v).

CHAPTER 23. RANDOMIZED CONSENSUS AND TEST-AND-SET211

These last two requirement replace the agreement property of consensus.

They are also strictly weaker than consensus, which means that a consensus
object (with all its output labeled commit) is also an adopt-commit object.
The reason we like adopt-commit objects is that they allow the simple
consensus protocol shown in Algorithm 23.1.

1 preference ← input
2 for r ← 1 . . . ∞ do
3 (b, preference) ← AdoptCommit(AC[r], preference)
4 if b = commit then
5 return preference
6 else
7 do something to generate a new preference

Algorithm 23.1: Consensus using adopt-commit

The idea is that the adopt-commit takes care of ensuring that once
somebody returns a value (after receiving commit), everybody else who
doesn’t return adopts the same value (follows from coherence). Conversely,
if everybody already has the same value, everybody returns it (follows from
convergence). The only missing piece is the part where we try to shake all
the processes into agreement. For this we need a separate object called a
conciliator.

23.3.2 Conciliators
Conciliators are a weakened version of randomized consensus that replace
agreement with probabilistic agreement: the processes can disagree some-
times, but must agree with constant probability despite interference by the
adversary. An algorithm that satisfies termination, validity, and probabilistic
agreement is called a conciliator.1
The important feature of conciliators is that if we plug a conciliator that
guarantees agreement with probability at least δ into Algorithm 23.1, then
on average we only have to execute the loop 1/δ times before every process
agrees. This gives an expected cost equal to 1/δ times the total cost of
AdoptCommit and the conciliator. Typically we will aim for constant δ.
1
Warning: This name has not really caught on in the general theory-of-distributed-
computing community, and so far only appears in papers that have a particular researcher
as a co-author [Asp12a, AE11, Asp12b, AACV17]. Unfortunately, there doesn’t seem to
be a better name for the same object that has caught on. So we are stuck with it for now.
CHAPTER 23. RANDOMIZED CONSENSUS AND TEST-AND-SET212

23.4 Implementing an adopt-commit object

What’s nice about adopt-commit objects is that they can be implemented
deterministically. Here we’ll give a simple adopt-commit object for two values,
0 and 1. Optimal (under certain assumptions) constructions of m-valued
adopt-commits can be found in [AE11].
Pseudocode is given in Algorithm 23.2.

shared data: a[0], a[1], initially false; proposal, initially ⊥

1 procedure AdoptCommit(v)
2 a[v] ← 1
3 if proposal = ⊥ then
4 proposal ← v
5 else
6 v ← proposal
7 if a[¬v] = false then
8 return (commit, v)
9 else
10 return (adopt, v)

Algorithm 23.2: A 2-valued adopt-commit object

Structurally, this is pretty similar to a splitter (see §17.4.2), except that

we use values instead of process ids.
We now show correctness. Termination and validity are trivial. For
coherence, observe that if I return (commit, v) I must have read a[¬v] = false
before any process with ¬v writes a[¬v]; it follows that all such processes
will see proposal 6= ⊥ and return (adopt, v). For convergence, observe that
if all processes have the same input v, they all write it to proposal and all
observe a[¬v] = false, causing them all to return (commit, v).

23.5 Conciliators and shared coins

For an adaptive adversary, the usual way to implement a conciliator is
from a weak shared coin [AH90a], which is basically a non-cryptographic
version of the common coin [Rab83] found in many cryptographic Byzantine
agreement protocols. Formally, a weak shared coin is an object that has
no inputs and returns either 0 or 1 to all processes with some minimum
probability δ. By itself this does not give validity, so converting a weak
CHAPTER 23. RANDOMIZED CONSENSUS AND TEST-AND-SET213

shared coin into a conciliator requires extra machinery to bypass the coin if
the processes that have accessed the conciliator so far are all in agreement;
see Algorithm 23.3. The intuition is that having some processes (who all
agree with each other) skip the shared coin is not a problem, because with
probability δ the remaining processes will agree with them as well.

shared data:
binary registers r0 and r1 , initially 0;
weak shared coin sharedCoin
1 procedure coinCoinciliator()
2 rv ← 1
3 if r¬v = 1 then
4 return sharedCoin()
5 else
6 return v

Algorithm 23.3: Shared coin conciliator from [Asp12b]

This still leaves the problem of how to build a shared coin. In the
message-passing literature, the usual approach is to use cryptography,2 but
because we are assuming an arbitrarily powerful adversary, we can’t use
cryptography.
If we don’t care how small δ gets, we could just have each process flip its
own local coin and hope that they all come up the same. (This is more or
less what was done by Abrahamson [Abr88].) But that might take a while. If
we aren’t willing to wait exponentially long, a better approach is to combine
many individual local coins using some sort of voting.
A version of this approach, based on a random walk, was used by Aspnes
and Herlihy [AH90a] to get consensus in (bad) polynomial expected time
against an adaptive adversary. A better version was developed by Bracha
and Rachman [BR91]. In their version, each process repeatedly generates a
random ±1 vote and adds it to a common pool (which just means writing the
sum and count of all its votes so far out to a single-writer register). Every
Θ(n/ log n) votes, the process does a collect (giving an overhead of Θ(log n)
operations per vote) and checks to see if the total number of votes is greater
than a Θ(n2 ) threshold. If it is, the process returns the sign of the total vote.
Bracha and Rachman showed that despite processes seeing different
combinations of votes (due to the collects running at possibly very different
2
For example, Canetti and Rabin [CR93] solved Byzantine agreement in O(1) time
building a shared coin on top of secret sharing.
CHAPTER 23. RANDOMIZED CONSENSUS AND TEST-AND-SET214

speeds), the difference between what each process sees and the actual sum of
all votes ever generated is at most O(n) with high probability. This means
that if the total vote is more than cn from 0 for some c, which occurs with
constant probability, then every processes is likely to return the same value.
This gives a weak shared coin with constant bias, and thus also a consensus
protocol, that runs in O(n2 log n) expected total steps.
This remained the best known protocol for many years, leaving an
annoying gap between the upper bound and the best known lower bound
of Ω(n2 / log2 n) [Asp98]. Eventually, Attiya and Censor [AC08] produced
an entirely new argument to bring the lower bound up to Ω(n2 ) and at the
same time gave a simple tweak to the Bracha-Rachman protocol to bring
the upper bound down to O(n2 ), completely settling (up to constant factors)
the asymptotic expected total step complexity of strong-adversary consensus.
But the question of how quickly one could solve weak-adversary adversary
consensus remained (and still remains) open.

23.6 A one-register conciliator for an oblivious ad-

versary

shared data: register r, initially ⊥

1 k←0
2 while r = ⊥ do
2k
3 with probability 2n do
4 write v to r
5 else
6 do a dummy operation
7 k ←k+1
8 return r
Algorithm 23.4: Impatient first-mover conciliator from [Asp12b]

Algorithm 23.4 implements a conciliator for an oblivious adversary3 using

a single register. This particular construction is taken from [Asp12b], and
is based on an earlier algorithm of Chor, Israeli, and Li [CIL94]. The cost
of this algorithm is expected O(n) total work and O(log n) individual work.
Later (§23.7.2), we will see a different algorithm [Asp12a] that reduces the
3
Or any adversary weak enough not to be able to block the write based on how the
coin-flip turned out.
CHAPTER 23. RANDOMIZED CONSENSUS AND TEST-AND-SET215

individual work to O(log log n), although the total work for that algorithm
may be O(n log log n).
The basic idea is that processes alternate between reading a register r
and (maybe) writing to the register; if a process reads a non-null value from
the register, it returns it. Any other process that reads the same non-null
value will agree with the first process; the only way that this can’t happen is
if some process writes a different value to the register before it notices the
first write.
The random choice of whether to write the register or not avoids this
problem. The idea is that even though the adversary can schedule a write at
a particular time, because it’s oblivious, it won’t be able to tell if the process
wrote (or was about to write) or did a no-op instead.
The basic version of this algorithm, due to Chor, Israeli, and Li [CIL94],
1
uses a fixed 2n probability of writing to the register. So once some process
writes to the register, the chance that any of the remaining n − 1 processes
write to it before noticing that it’s non-null is at most n−12n < 1/2. It’s also
not hard to see that this algorithm uses O(n) total operations, although it
may be that one single process running by itself has to go through the loop
2n times before it finally writes the register and escapes.
Using increasing probabilities avoids this problem, because any process
that executes the main loop dlg ne + 1 times will write the register. This
establishes the O(log n) per-process bound on operations. At the same time,
an O(n) bound on total operations still holds, since each write has at least
1
a 2n chance of succeeding. The price we pay for the improvement is that
we increase the chance that an initial value written to the register gets
overwritten by some high-probability write. But the intuition is that the
probabilities can’t grow too much, because the probability that I write on
my next write is close to the sum of the probabilities that I wrote on my
previous writes—suggesting that if I have a high probability of writing next
time, I should have done a write already.
Formalizing this intuition requires a little bit of work. Fix the schedule,
and let pi be the probability that the i-th write operation in this schedule
succeeds. Let t be the least value for which ti=1 pi ≥ 1/4. We’re going to
P

argue that with constant probability one of the first t writes succeeds, and
that the next n − 1 writes by different processes all fail.
CHAPTER 23. RANDOMIZED CONSENSUS AND TEST-AND-SET216

The probability that none of the first t writes succeed is

t t
e−pi
Y Y
(1 − pi ) ≤
i=1 i=1
t
!
X
= exp pi
i=1
≤ e−1/4 .

Now observe that if some process p writes at or before the t-th write,
then any process q with a pending write either did no writes previously, or
its last write was among the first t − 1 writes, whose probabilities sum to
P 1
less than 1/4. In either case, q has a i∈Sq pi + 2n chance of writing on
its pending attempt, where Sq is the set of indices in 1 . . . t − 1 where q
previously attempted to write.
Summing up these probabilities over all processes gives a total of n−1 2n +
P P
p ≤ 1/2 + 1/4 = 3/4. So with probability at least e −1/4 (1 − 3/4) =
q i∈Sq i
e−1/4 /4, we get agreement.

23.7 Sifters
A faster conciliator can be obtained using a sifter, which is a mechanism for
rapidly discarding processes using randomization [AA11] while keeping at
least one process around. The simplest sifter has each process either write a
register (with low probability) or read it (with high probability); all writers
and all readers that see ⊥ continue to the next stage of the protocol, while
all readers who see a non-null value drop out. If the probability of writing
√
is tuned carefully, this will reduce n processes to at most 2 n processes on
average; by iterating this mechanism, the expected number of remaining
processes can be reduced to 1 +  after O(log log n + log(1/)) phases.
As with previous implementations of test-and-set (see Algorithm 22.3),
it’s often helpful to have a sifter return not only that a process lost but which
process it lost to. This gives the implementation shown in Algorithm 23.5.
To use a sifter effectively, p should be tuned to match the number of
processes that are likely to use it. This is because of the following lemma:
Lemma 23.7.1. Fix p, and let X processes executed a sifter with parameter
p. Let Y be the number of processes for which the sifter returns ⊥. Then
1
E [X | Y ] ≤ pX + . (23.7.1)
p
CHAPTER 23. RANDOMIZED CONSENSUS AND TEST-AND-SET217

1 procedure sifter(p, r)
2 with probability p do
3 r ← id
4 return ⊥
5 else
6 return r

Algorithm 23.5: A sifter

Proof. In order to return ⊥, a process must either (a) write to r, which

occurs with probability p, or (b) read r before any other process writes to
it. The expected number of writers, conditioned on X, is exactly pX. The
expected number of readers before the first write has a geometric distribution
truncated by X. Removing the truncation gives exactly p1 expected readers,
which is an upper bound on the correct value.

For n initial processes, the choice of p that minimizes the bound in

√
(23.7.1) is √1n , giving at most 2 n expected survivors. Iterating this process
√ q √
with optimal p at each step gives a sequence of at most n, 2 n, 2 2 n,
etc., expected survivors after each sifter. The twos are a little annoying, but
a straightforward induction bounds the expected survivors after i rounds
−i
by 4 · n2 . In particular, we get at most 8 expected survivors after dlg lg ne
rounds.
At this point it makes sense to switch to a fixed p and a different analysis.
For p = 1/2, the first process to access r always survives, and each subsequent
process survives with probability at most 3/4 (because it leaves if the first
process writes and it reads). So the number of “excess” processes drops
as (3/4)i , and an additional dlog4/3 (7/)e rounds are enough to reduce the
expected number of survivors from 1 + 7 to 1 +  for any fixed .4
It follows that
Theorem 23.7.2. An initial set of n processes can be reduced to 1 with
probability at least 1 −  using O(log log n + log(1/)) rounds of sifters.
Proof. Let X be the number of survivors after dlg lg ne+dlog4/3 (7/)e rounds
of sifters, with probabilities tuned as described above. We’ve shown that
E [X] ≤ 1 + , so E [X − 1] ≤ . Since X − 1 ≥ 0, from Markov’s inequality
we have Pr [X ≥ 2] = Pr [X − 1 ≥ 1] ≤ E [X − 1] /1 ≤ .
4
This argument essentially follows the proof of [Asp12a, Theorem 2], which, because of
neglecting to subtract off a 1 at one point, ends up with 8/ instead of 7/.
CHAPTER 23. RANDOMIZED CONSENSUS AND TEST-AND-SET218

23.7.1 Test-and-set using sifters

Sifters were initially designed to be used for test-and-set. For this purpose,
we treat a return value of ⊥ as “keep going” and anything else as “leave with
value 1.” Using O(log log n) rounds of sifters, we can get down to one process
that hasn’t left with probability at least 1 − log−c n for any fixed constant
c. We then need a fall-back TAS to handle the log−c n chance that we get
more than one such survivor.
Alistarh and Aspnes [AA11] used the RatRace algorithm of Alistarh et
al. [AAG+ 10] for this purpose. This is an adaptive randomized test-and-set
built from splitters and two-process consensus objects that runs in O(log k)
expected time, where k is the number of processes that access the test-and-set;
a sketch of this algorithm is given in §24.5.2. If we want to avoid appealing
to this algorithm, a somewhat simpler approach is to use an approach similar
to the Lamport’s fast-path mutual exclusion algorithm (described in §17.4.2):
any process that survives the sifters tries to rush to a two-process TAS at the
top of a tree of two-processes TASes by winning a splitter, and if it doesn’t
win the splitter, it enters at a leaf and pays O(log n) expected steps. By
setting  = 1/ log n, the overall expected cost of this final stage is O(1).
This algorithm does not guarantee linearizability. I might lose a sifter
early on only to have a later process win all the sifters (say, by writing to
each one) and return 0. A gate bit as in Algorithm 22.3 solves this problem.
The full code is given in Algorithm 23.6.

23.7.2 Consensus using sifters

With some trickery, the sifter mechanism can be adapted to solve consensus,
still in O(log log n) expected individual work [Asp12a]. The main difficulty
is that a process can no longer drop out as soon as it knows that it lost: it
still needs to figure out who won, and possible help that winner over the
finish line.
The basic idea is that when a process p loses a sifter to some other process
q, p will act like a clone of q from that point on. In order to make this work,
each process writes down at the start of the protocol all of the coin-flips it
intends to use to decide whether to read or write at each round of sifting.
Together with its input, these coin-flips make up the process’s persona.
In analyzing the progress of the sifter, we count surviving personae (with
multiple copies of the same persona counting as one) instead of surviving
processes.
Pseudocode for this algorithm is given in Algorithm 23.7. Note that the
CHAPTER 23. RANDOMIZED CONSENSUS AND TEST-AND-SET219

1 if gate 6= ⊥ then
2 return 1
3 else
4 gate ← myId
5 for i ← 1 . . . dlog log ne + dlog
 4/3
(7 log n)e do
−i+1
6 with probability min 1/2, 21−2 do
7 ri ← myId
8 else
9 w ← ri
10 if w 6= ⊥ then
11 return 1

12 if splitter() = stop then

13 return 0
14 else
15 return AWWTAS()
Algorithm 23.6: Test-and-set in O(log log n) expected time

loop body is essentially the same as the code in Algorithm 23.5, except that
the random choice is replaced by a lookup in persona.chooseWrite.
To show that this works, we need to argue that having multiple copies
of a persona around doesn’t change the behavior of the sifter. In each
round, we will call the first process with a given persona p to access ri
the representative of p, and argue that a persona survives round i in
this algorithm precisely when its representative would survive round i in
a corresponding test-and-set sifter with the schedule restricted only to the
representatives.
There are three cases:

1. The representative of p writes. Then at least one copy of p survives.

2. The representative of p reads a null value. Again at least one copy of

p survives.

3. The representative of p reads a non-null value. Then no copy of p

survives: all subsequent reads by processes carrying p also read a
non-null value and discard p, and since no process with p writes, no
other process adopts p.
CHAPTER 23. RANDOMIZED CONSENSUS AND TEST-AND-SET220

1 procedure conciliator(input)
2 Let R = dlog log ne + dlog4/3 (7/)e
3 Let chooseWrite be a vector of R independent random Boolean
variables with Pr[chooseWrite[i] = 1] = pi , where
−i+1 −i
pi = 21−2 (n)−2 for i ≤ dlog log ne and pi = 1/2 for larger i.

4 persona ← hinput, chooseWrite, myIdi

5 for i ← 1 . . . R do
6 if persona.chooseWrite[i] = 1 then
7 ri ← persona
8 else
9 v ← ri
10 if v 6= ⊥ then
11 persona ← v

12 return persona.input
Algorithm 23.7: Sifting conciliator (from [Asp12a])

From the preceding analysis for test-and-set, we have that after O(log log n+
log 1/) rounds with appropriate probabilities of writing, at most 1 +  values
survive on average. This gives a probability of at most  of disagreement. By
alternating these conciliators with adopt-commit objects, we get agreement
in O(log log n + log m/ log log m) expected time, where m is the number of
possible input values.
I don’t think the O(log log n) part of this expression is optimal, but I
don’t know how to do better.

23.7.3 A better sifter for test-and-set

A more sophisticated sifter due to Giakkoupis and Woelfel [GW12a] removes
all but O(log n) processes, on average, using two operations for each process.
Iterating this sifter reduces the expected survivors to O(1) in O(log∗ n)
rounds. A particularly nice feature of the Giakkoupis-Woelfel algorithm is
that (if you don’t care about space) it doesn’t have any parameters that
require tuning to n: this means that exactly the same structure can be used
in each round. An unfortunate feature is that it’s not possible to guarantee
that every process that leaves learns the identity of a process that stays: this
means that it can’t adapted into a consensus protocol using the persona trick
CHAPTER 23. RANDOMIZED CONSENSUS AND TEST-AND-SET221

described in §23.7.2.
Pseudocode is given in Algorithm 23.8. In this simplified version, we
assume an infinitely long array A[1 . . . ], so that we don’t need to worry
about n. Truncating the array at log n also works, but the analysis requires
handling the last position as a special case, which I am too lazy to do here.

1 Choose r ∈ Z+ such that Pr [r = i] = 2−i

2 A[r] ← 1
3 if A[r + 1] = 0 then
4 stay
5 else
6 leave
Algorithm 23.8: Giakkoupis-Woelfel sifter [GW12a]

Lemma 23.7.3. In any execution of Algorithm 23.8 with an oblivious ad-

versary and n processes, at least one process stays, and the expected number
of processes that stay is O(log n).

Proof. For the first part, observe that any process that picks the largest
value of r among all processes will survive; since the number of processes is
finite, there is at least one such survivor.
For the second part, let Xi be the number of survivors with r = i. Then
E [Xi ] is bounded by n · 2−i , since no process survives with r = i without
first choosing r = i. But we can also argue that E [Xi ] ≤ 3 for any value of
n, by considering the sequence of write operations in the execution.
Because the adversary is oblivious, the location of these writes is uncor-
related with their ordering. If we assume that the adversary is trying to
maximize the number of survivors, its best strategy is to allow each process
to read immediately after writing, as delaying this read can only increase the
probability that A[r + 1] is nonzero. So in computing Xi , we are counting
the number of writes to A[i] before the first write to A[i + 1]. Let’s ignore
all writes to other registers; then the j-th write to either of A[i] or A[i + 1]
has a conditional probability of 2/3 of landing on A[i] and 1/3 on A[i + 1].
We are thus looking at a geometric distribution with parameter 1/3, which
has expectation 3.
CHAPTER 23. RANDOMIZED CONSENSUS AND TEST-AND-SET222

Combining these two bounds gives E [Xi ] ≤ min(3, 2−i ). So then

∞
min(3, n · 2−i )
X
E [survivors] ≤
i=1
= 3 lg n + O(1),

because once n · 2−i drops below 3, the remaining terms form a geometric
series.

Like square root, logarithm is concave, so Jensen’s inequality applies here

as well. So O(log∗ n) rounds of Algorithm 23.8 reduces us to an expected
constant number of survivors, which can then be fed to RatRace.
With an adaptive adversary, all of the sifter-based test-and-sets fail badly:
in this particular case, an adaptive adversary can sort the processes in order
of increasing write location so that every process survives. The best known
n-process test-and-set for an adaptive adversary is still a tree of 2-process
randomized test-and-sets, as in the Afek et al. [AWW93] algorithm described
in §22.2. Whether O(log n) expected steps is in fact necessary is still open
(as is the exact complexity of test-and-set with an oblivious adversary).

23.8 Space bounds

√
A classic result of Fich, Herlihy, and Shavit [FHS98] showed that Ω( n)
registers are needed to solve consensus even under the very weak require-
ment of nondeterministic solo termination, which says that for every
reachable configuration and every process p, there exists some continuation
of the execution in which the protocol terminates with only p running. The
best known upper bound is the trivial bound of n—one single-writer register
per process—since any algorithm that uses multi-writer registers can be
translated into one that uses only single-writer registers, and (assuming wide
enough registers) multiple registers of a single process can be combined into
one.
For many years, there was very little progress in closing the gap between
these two bounds. In 2013, we got a hint that FHS might be tight when
√
Giakkoupis et al. [GHHW13] gave a surprising O( n)-space algorithm for
the closely related problem of obstruction-free one-shot test-and-set.
But then Gelashvili [Gel15] showed an n/20 space lower bound for con-
sensus for anonymous processes, and Zhu quickly followed this with a lower
bound for non-anonymous processes [Zhu16], showing that at least n − 1
registers are required, using a clever combination of bivalence and covering
CHAPTER 23. RANDOMIZED CONSENSUS AND TEST-AND-SET223

arguments. Around the same time, Giakkoupis et al. [GHHW15] further

improved the space complexity of obstruction-free test-and-set to O(log n),
using a deterministic obstruction-free implementation of a sifter. So the brief
√ √
coincidence of the Ω( n) lower bound on consensus and the O( n) upper
bound on test-and-set turned out to be an accident.
For consensus, there is still a gap, but it’s a very small gap. Whether
the actual space needed is n − 1 or n remains open.
Chapter 24

Renaming

We will start by following the presentation in [AW04, §16.3]. This mostly

describes results of the original paper of Attiya et al. [ABND+ 90] that defined
the renaming problem and gave a solution for message-passing; however, it’s
now more common to treat renaming in the context of shared-memory, so we
will follow Attiya and Welch’s translation of these results to a shared-memory
setting.

24.1 Renaming
In the renaming problem, we have n processes, each starts with a name
from some huge namespace, and we’d like to assign them each unique names
from a much smaller namespace. The main application is allowing us to run
algorithms that assume that the processes are given contiguous numbers,
e.g., the various collect or atomic snapshot algorithms in which each process
is assigned a unique register and we have to read all of the registers. With
renaming, instead of reading a huge pile of registers in order to find the few
that are actually used, we can map the processes down to a much smaller
set.
Formally, we have a decision problem where each process has input xi
(its original name) and output yi , with the requirements:

Termination Every nonfaulty process eventually decides.

Uniqueness If pi 6= pj , then yi 6= yj .

Anonymity The code executed by any process depends only on its input
xi : for any execution of processes p1 . . . pn with inputs x1 . . . xn , and

224
CHAPTER 24. RENAMING 225

any permutation π of [1 . . . n], there is a corresponding execution of

processes pπ(1) . . . pπ(n) with inputs x1 . . . xn in which pπ(i) performs
exactly the same operations as pi and obtains the same output yi .

The last condition is like non-triviality for consensus: it excludes algo-

rithms where pi just returns i in all executions. Typically we do not have
to do much to prove anonymity other than observing that all processes are
running the same code.
We will be considering renaming in a shared-memory system, where we
only have atomic registers to work with.

24.2 Performance
Conventions on counting processes:

• N = number of possible original names.

• n = maximum number of processes.

• k = number of processes that actually execute the algorithm.

Ideally, we’d like any performance measures we get to depend on k alone

if possible (giving an adaptive algorithm). Next best would be something
polynomial in n and k. Anything involving N is bad.
We’d also like to minimize the size of the output namespace. How well
we can do this depends on what assumptions we make. For deterministic
algorithms using only read-write registers, a lower bound due to Herlihy and
Shavit [HS99] shows that we can’t get fewer than 2n − 1 names for general
n.1 Our target thus will be exactly 2n − 1 output names if possible, or 2k − 1
if we are trying to be adaptive. For randomized algorithms, it is possible to
solve strong or tight renaming, where the size of the namespace is exactly
k; we’ll see how to do this in §24.5.
A small note on bounds: There is a lot of variation in the literature on
how bounds on the size of the output namespace are stated. The original
Herlihy-Shavit lower bound [HS99] says that there is no general renaming
algorithm that uses 2n names for n + 1 processes; in other words, any n-
process algorithm uses at least 2n − 1 names. Many subsequent papers
1
This lower bound was further refined by Castañeda and Rajsbaum [CR08], who show
that 2n − 2 (but no less!) is possible for certain special values of n; all of these lower
bounds make extensive use of combinatorial topology, so we won’t try to present them
here.
CHAPTER 24. RENAMING 226

discussing lower bounds on the namespace follow the approach of Herlihy and
Shavit and quote lower bounds that are generally 2 higher than the minimum
number of names needed for n processes. This requires a certain amount of
translation when comparing these lower bounds with upper bounds, which
use the more natural convention.

24.3 Order-preserving renaming

Before we jump into upper bounds, let’s do an easy lower bound from
the Attiya et al. paper [ABND+ 90]. This bound works on a variant of
renaming called order-preserving renaming, where we require that yi <
yj whenever xi < xj . Unfortunately, this requires a very large output
namespace: with t failures, any asynchronous algorithm for order-preserving
renaming requires 2t (n − t + 1) − 1 possible output names. This lower bound
applies regardless of the model, as long as some processes may start after
other processes have already been assigned names.
For the wait-free case, we have t = n − 1, and the bound becomes just
n
2 − 1. This is a simpler case than the general t-failure case, but the essential
idea is the same: if I’ve only seen a few of the processes, I need to leave room
for the others.

Theorem 24.3.1. There is no order-preserving renaming algorithm for n

processes using fewer than 2n − 1 names.

Proof. By induction on n. For n = 1, we use 21 − 1 = 1 names; this is the

base case. For larger n, suppose we use m names, and consider an execution
in which one process pn runs to completion first. This consumes one name
yn and leaves k names less than yn and m − k − 1 names greater than yn . By
setting all the inputs xi for i < n either less than xn or greater than xn , we
can force the remaining processes to choose from the remaining k or m − k − 1
names. Applying the induction hypothesis, this gives k ≥ 2n−1 − 1 and
m−k −1 ≥ 2n−1 −1, so m = k +(m−k −1)+1 ≥ 2(2n−1 −1)+1 = 2n −1.

24.4 Deterministic renaming

In deterministic renaming, we can’t use randomization, and may or may
not have any primitives stronger than atomic registers. With just atomic
registers, we can only solve loose renaming; with test-and-set, we can solve
tight renaming. In this section, we describe some basic algorithms for
deterministic renaming.
CHAPTER 24. RENAMING 227

24.4.1 Wait-free renaming with 2n − 1 names

Here we use Algorithm 55 from [AW04], which is an adaptation to shared
memory of the message-passing renaming algorithm of [ABND+ 90]. One odd
feature of the algorithm is that, as written, it is not anonymous: processes
communicate using an atomic snapshot object and use their process ids to
select which component of the snapshot array to write to. But if we think of
the process ids used in the algorithm as the inputs xi rather than the actual
process ids i, then everything works. The version given in Algorithm 24.1
makes this substitution explicit, by treating the original name i as the input.

1 procedure getName()
2 s←1
3 while true do
4 a[i] ← s
5 view ← snapshot(a)
6 if view[j] = s for some j then
7 r ← |{j : view[j] 6= ⊥ ∧ j ≤ i}|
8 s ← r-th positive integer not in
{view[j] : j 6= i ∧ view[j] = ⊥}
9 else
10 return s

Algorithm 24.1: Wait-free deterministic renaming

The array a holds proposed names for each process (indexed by the
original names), or ⊥ for processes that have not proposed a name yet. If a
process proposes a name and finds that no other process has proposed the
same name, it takes it; otherwise it chooses a new name by first computing
its rank r among the active processes and then choosing the r-th smallest
name that hasn’t been proposed by another process. Because the rank is at
most n and there are at most n − 1 names proposed by the other processes,
this always gives proposed names in the range [1 . . . 2n − 1]. But it remains
to show that the algorithm satisfies uniqueness and termination.
For uniqueness, consider two process with original names i and j. Suppose
that i and j both decide on s. Then i sees a view in which a[i] = s and
a[j] 6= s, after which it no longer updates a[i]. Similarly, j sees a view in
which a[j] = s and a[i] 6= s, after which it no longer updates a[j]. If i’s view
is obtained first, then j can’t see a[i] 6= s, but the same holds if j’s view is
obtained first. So in either case we get a contradiction, proving uniqueness.
CHAPTER 24. RENAMING 228

Termination is a bit trickier. Here we argue that no process can run

forever without picking a name, by showing that if we have a set of processes
that are doing this, the one with smallest original name eventually picks a
name. More formally, call a process trying if it runs for infinitely many steps
without choosing a name. Then in any execution with at least one trying
process, eventually we reach a configuration where all processes have either
finished or are trying. In some subsequent configuration, all the processes
have written to the a array at least once; from this point on, the set of non-
null positions in a—and thus the rank each process computes for itself—is
stable.
Starting from some such stable configuration, look at the trying process
i with the smallest original name, and suppose it has rank r. Let F =
{z1 < z2 . . . } be the set of “free names” that are not proposed in a by any
of the finished processes. Observe that no trying process j 6= i ever proposes
a name in {z1 . . . zr }, because any such process has rank greater than r.
This leaves zr open for i to claim, provided the other names in {z1 . . . zr }
eventually become free. But this will happen, because only trying processes
may have proposed these names (early on in the execution, when the finished
processes hadn’t finished yet), and the trying processes eventually propose
new names that are not in this range. So eventually process i proposes zr ,
sees no conflict, and finishes, contradicting the assumption that it is trying.
Note that we haven’t proved any complexity bounds on this algorithm at
all, but we know that the snapshot alone takes at least Ω(N ) time and space.
Brodksy et al. [BEW11] cite a paper of Bar-Noy and Dolev [BND89] as giving
a shared-memory version of [ABND+ 90] with complexity O(n · 4n ); they also
give algorithms and pointers to algorithms with much better complexity.

24.4.2 Long-lived renaming

In long-lived renaming a process can release a name for later use by other
processes (or the same process, if it happens to run choose-name again).
Now the bound on the number of names needed is 2k − 1, where k is the
maximum number of concurrently active processes. Algorithm 24.1 can be
converted to a long-lived renaming algorithm by adding the releaseName
procedure given in Algorithm 24.2. This just erases the process’s proposed
name, so that some other process can claim it.
Here the termination requirement is weakened slightly, to say that some
process always makes progress in getName. It may be, however, that there
is some process that never successfully obtains a name, because it keeps
getting stepped on by other processes zipping in and out of getName and
CHAPTER 24. RENAMING 229

1 procedure releaseName()
2 a[i] ← ⊥
Algorithm 24.2: Releasing a name

releaseName.

24.4.3 Renaming without snapshots

Moir and Anderson [MA95] give a renaming protocol that is somewhat easier
to understand and doesn’t require taking snapshots over huge arrays. A
downside is that the basic version requires k(k + 1)/2 names to handle k
active processes.

24.4.3.1 Splitters
The Moir-Anderson renaming protocol uses a network of splitters, which
we last saw providing a fast path for mutual exclusion in §17.4.2. Each
splitter is a widget, built from a pair of atomic registers, that assigns to
each processes that arrives at it the value right, down, or stop. As discussed
previously, the useful properties of splitters are that if at least one process
arrives at a splitter, then (a) at least one process returns right or stop; and
(b) at least one process returns down or stop; (c) at most one process returns
stop; and (d) any process that runs by itself returns stop.
We proved the last two properties in §17.4.2; we’ll prove the first two here.
Another way of describing these properties is that of all the processes that
arrive at a splitter, some process doesn’t go down and some process doesn’t
go right. By arranging splitters in a grid, this property guarantees that every
row or column that gets at least one process gets to keep it—which means
that with k processes, no process reaches row k + 1 or column k + 1.
Algorithm 24.3 gives the implementation of a splitter (it’s identical to
Algorithm 17.5, but it will be convenient to have another copy here).

Lemma 24.4.1. If at least one process completes the splitter, at least one
process returns stop or right.

Proof. Suppose no process returns right; then every process sees open in door,
which means that every process writes its id to race before any process closes
the door. Some process writes its id last: this process will see its own id in
race and return stop.
CHAPTER 24. RENAMING 230

shared data:
1 atomic register race, big enough to hold an id, initially ⊥
2 atomic register door, big enough to hold a bit, initially open
3 procedure splitter(id)
4 race ← id
5 if door = closed then
6 return right
7 door ← closed
8 if race = id then
9 return stop
10 else
11 return down

Algorithm 24.3: Implementation of a splitter

Lemma 24.4.2. If at least one process completes the splitter, at least one
process returns stop or down.
Proof. First observe that if no process ever writes to door, then no process
completes the splitter, because the only way a process can finish the splitter
without writing to door is if it sees closed when it reads door (which must
have been written by some other process). So if at least one process finishes,
at least one process writes to door. Let p be any such process. From the
code, having written door, it has already passed up the chance to return
right; thus it either returns stop or down.

24.4.3.2 Splitters in a grid

Now build an m-by-m triangular grid of splitters, arranged as rows 0 . . . m − 1
and columns 0 . . . m − 1, where a splitter appears in each position (r, c) with
r+c ≤ m−1 (see Figure 24.1 for an example; this figure is taken from [Asp10]).
Assign a distinct name to each of the m


2 splitters in this grid. To obtain
a name, a process starts at (r, c) = (0, 0), and repeatedly executes the
splitter at its current position (r, c). If the splitter returns right, it moves
to (r, c + 1); if down, it moves to (r + 1, c); if stop, it stops, and returns the
name of its current splitter. This gives each name to at most one process
(by Lemma 17.4.3); we also have to show that if at most m processes enter
the grid, every process stops at some splitter.
The argument for this is simple. Suppose some process p leaves the
grid on one of the 2m output wires. Look at the path it takes to get there
CHAPTER 24. RENAMING 231

Figure 24.1: A 6 × 6 Moir-Anderson grid

(see Figure 24.2, also taken from [Asp10]). Each splitter on this path must
handle at least two processes (or p would have stopped at that splitter, by
Lemma 17.4.4). So some other process leaves on the other output wire, either
right or down. If we draw a path from each of these wires that continues right
or down to the end of the grid, then along each of these m disjoint paths
either some splitter stops a process, or some process reaches a final output
wire, each of which is at a distinct splitter. But this gives m processes in
addition to p, for a total of m + 1 processes. It follows that:

Theorem 24.4.3. An m × m Moir-Anderson grid solves renaming for up

to m processes.

The time complexity of the algorithm is O(m): Each process spends at

most 4 operations on each splitter, and no process goes through more than
2m splitters. In general, any splitter network will take at least n steps to
stop n processes, because the adversary can run them all together in a horde
that drops only one process at each splitter.
If we don’t know k in advance, we can still guarantee names of size O(k 2
)
by carefully arranging them so that each k-by-k subgrid contains the first k2
names. This gives an adaptive renaming algorithm (although the namespace
size is pretty high). We still have to choose our grid to be large enough for
the largest k we might actually encounter; the resulting space complexity is
O(n2 ).
CHAPTER 24. RENAMING 232

Figure 24.2: Path taken by a single process through a 6 × 6 Moir-Anderson

grid (heavy path), and the 6 disjoint paths it spawns (dashed paths).
(From [Asp10].)

With a slightly more clever arrangement of the splitters, it is possible to

reduce the space complexity to O(n3/2 ) [Asp10]. Whether further reductions
are possible is an open problem. Note however that linear time complexity
makes splitter networks uncompetitive with much faster randomized algo-
rithms (as we’ll see in §24.5), so this may not be a very important open
problem.

24.4.4 Getting to 2n − 1 names in polynomial space

From before, we have an algorithm that will get 2n − 1 names for n processes
out of N possible processes when run using O(N ) space (for the enormous
snapshots). To turn this into a bounded-space algorithm, run Moir-Anderson
first to get down to Θ(k 2 ) names, then run the previous algorithm (in Θ(n2 )
space) using these new names as the original names.
Since we didn’t prove anything about time complexity of the humongous-
snapshot algorithm, we can’t say much about the time complexity of this
combined one. Moir and Anderson suggest instead using an O(N k 2 ) algo-
rithm of Borowsky and Gafni to get O(k 4 ) time for the combined algorithm.
This is close to the best known: a later paper by Afek and Merritt [AM99]
holds the current record for deterministic adaptive renaming into 2k − 1
names at O(k 2 ) individual steps. On the lower bound side, it is known that
CHAPTER 24. RENAMING 233

Ω(k) is a lower bound on the individual steps of any renaming protocol with
a polynomial output namespace [AAGG11].

24.4.5 Renaming with test-and-set

Moir and Anderson give a simple renaming algorithm based on test-and-set
that is strong (k processes are assigned exactly the names 1 . . . k), adaptive
(the time complexity to acquire a name is O(k)), and long-lived, which
means that a process can release its name and the name will be available to
processes that arrive later. In fact, the resulting algorithm gives long-lived
strong renaming, meaning that the set of names in use will always be no
larger than the set of processes that have started to acquire a name and not
yet finished releasing one; this is a little stronger than just saying that the
algorithm is strong and that it is long-lived separately.
The algorithm is simple: we have a line of test-and-set bits T [1] . . . T [n].
To acquire a name, a process starts at T [1] and attempts to win each test-
and-set until it succeeds; whichever T [i] it wins gives it name i. To release a
name, a process releases the test-and-set.
Without the releases, the same mechanism gives fetch-and-increment [AWW93].
Fetch-and-increment by itself solves tight renaming (although not long-lived
renaming, since there is no way to release a name).

24.5 Randomized renaming

With randomization, we can beat both the 2k − 1 lower bound on the size of
the output namespace from [HS99] and the Ω(k) lower bound on individual
work from [AAGG11], achieving strong renaming with O(log k) expected
individual work [AACH+ 11].
The basic idea is that we can use randomization for load balancing,
where we avoid the problem of having an army of processes marching together
with only a few peeling off at a time (as in splitter networks) by having the
processes split up based on random choices. For example, if each process
generates a random name consisting of 2dlg ne bits, then it is reasonably
likely that every process gets a unique name in a namespace of size O(n2 )
(we can’t hope for less than O(n2 ) because of the birthday paradox). But
we want all processes to be guaranteed to have unique names, so we need
some more machinery.
We also need the processes to have initial names; if they don’t, there is al-
ways some nonzero probability that two identical processes will flip their coins
CHAPTER 24. RENAMING 234

in exactly the same way and end up with the same name. This observation
was formalized by Buhrman, Panconesi, Silvestri, and Vitányi [BPSV06].

24.5.1 Randomized splitters

Attiya et al. [AKP+ 06] suggested the use of randomized splitters in the
context of another problem (adaptive collect) that is closely related to
renaming.
A randomized splitter is just like a regular splitter, except that if a process
doesn’t stop it flips a coin to decide whether to go right or down. Randomized
splitters are nice because they usually split better than deterministic splitters:
if k processes √ reach a randomized splitter, with high probability no more
than k/2 + O( k log k) will leave on either output wire.
It’s not hard to show that a binary tree of these things of depth 2dlg ne
stops all but a constant expected number of processes on average;2 processes
that don’t stop can be dropped into a backup renaming algorithm (Moir-
Anderson, for example) with only a constant increase in expected individual
work.
Furthermore, the binary tree of randomized splitters is adaptive; if only
k processes show up, we only need O(log k) levels levels on average to split
them up. This gives renaming into a namespace with expected size O(k 2 ) in
O(log k) expected individual steps.

24.5.2 Randomized test-and-set plus sampling

Subsequent work by Alistarh et al. [AAG+ 10] showed how some of the same
ideas could be used to get strong renaming, where the output namespace has
size exactly n (note this is not adaptive; another result in the same paper
gives adaptive renaming, but it’s not strong). There are two pieces to this
result: an implementation of randomized test-and-set called RatRace, and a
sampling procedure for getting names called ReShuffle.
The RatRace protocol implements a randomized test-and-set with O(log k)
expected individual work. The essential idea is to use a tree of randomized
splitters to assign names, then have processes walk back up the same tree
attempting to win a 3-process randomized test-and-set at each node (there
are 3 processes, because in addition to the winners of each subtree, we
2
The proof is to consider the expected number of pairs of processes that flip their coins
the same way for all 2dlg ne steps. This is at most n2 n−2 < 1/2, so on average at most 1


process escapes the tree, giving (by symmetry) at most a 1/n chance that any particular
process escapes. Making the tree deeper can give any polynomial fraction of escapees while
still keeping O(log n) layers.
CHAPTER 24. RENAMING 235

may also have a process that stopped on that node in the renaming step);
this test-and-set is just a very small binary tree of 2-process test-and-sets
implemented using the algorithm of Tromp and Vitányi [TV02]. A gate bit
is added at the top as in the test-and-set protocol of Afek et al. [AGTV92]
to get linearizability.
Once we have test-and-set, we could get strong renaming using a linear
array of test-and-sets as suggested by Moir and Anderson [MA95], but it’s
more efficient to use the randomization to spread the processes out. In
the ReShuffle protocol, each process chooses a name in the range [1 . . . n]
uniformly at random, and attempts to win a test-and-set guarding that name.
If it doesn’t work, it tries again. Alistarh et al. show that this method
produces unique names for everybody in O(n log4 n) total steps with high
probability. The individual step complexity of this algorithm, however, is
not very good: there is likely to be some unlucky process that needs Ω(n)
probes (at an expected cost of Θ(log n) steps each) to find an empty slot.

24.5.3 Renaming with sorting networks

A later paper by Alistarh et al. [AACH+ 11] reduces the cost of renaming still
further, getting O(log k) expected individual step complexity for acquiring a
name. The resulting algorithm is both adaptive and strong: with k processes,
only names 1 through k are used. We’ll describe the non-adaptive version
here.
The basic idea is to build a sorting network out of test-and-sets; the
resulting structure, called a renaming network, routes each process through
a sequence of test-and-sets to a unique output wire. Unlike a splitter network,
a renaming network uses the stronger properties of test-and-set to guarantee
that (once the dust settles) only the lowest-numbered output wires are chosen;
this gives strong renaming.

24.5.3.1 Sorting networks

A sorting network is a kind of parallel sorting algorithm that proceeds in
synchronous rounds, where in each round the elements of an array at certain
fixed positions are paired off and swapped if they are out of order. The
difference between a sorting network and a standard comparison-based sort
is that the choice of which positions to compare at each step is static, and
doesn’t depend on the outcome of previous comparisons; also, the only effect
of a comparison is possibly swapping the two values that were compared.
Sorting networks are drawn as in Figure 24.3. Each horizontal line or wire
CHAPTER 24. RENAMING 236

Figure 24.3: A sorting network

corresponds to a position in the array. The vertical lines are comparators

that compare two values coming in from the left and swap the larger value to
the bottom. A network of comparators is a sorting network if the sequences
of output values is always sorted no matter what the order of values on the
inputs is.
The depth of a sorting network is the maximum number of comparators
on any path from an input to an output. The width is the number of wires;
equivalently, the number of values the network can sort. The sorting network
in Figure 24.3 has depth 3 and width 4.
Explicit constructions of sorting networks with width n and depth
O(log2 n) are known [Bat68]. It is also known that sorting networks with
depth O(log n) exist [AKS83], but no explicit construction of such a network
is known.

24.5.3.2 Renaming networks

To turn a sorting network into a renaming network, we replace the compara-
tors with test-and-set bits, and allow processes to walk through the network
asynchronously. This is similar to an earlier mechanism called a counting
network [AHS94], which used certain special classes of sorting networks as
counters, but here any sorting network works.
Each process starts on a separate input wire, and we maintain the
invariant that at most one process ever traverses a wire. It follows that each
test-and-set bit is only used by two processes. The first process to reach the
test-and-set bit is sent out the lower output, while the second is sent out the
CHAPTER 24. RENAMING 237

upper output. If we imagine each process that participates in the protocol

as a one and each process that doesn’t as a zero, the test-and-set bit acts
as a comparator: if no processes show up on either input (two zeros), no
processes leave (two zeros again); if processes show up on both inputs (two
ones), processes leave on both (two ones again); and if only one process ever
shows up (a zero and a one), it leaves on the bottom output (zero and one,
sorted). Because the original sorting network sorts all the ones to the bottom
output wires, the corresponding renaming network sorts all the processes
that arrive to the bottom outputs. Label these outputs starting at 1 at the
bottom to get renaming.
Since each test-and-set involves at most two processes, we can carry them
out in O(1) expected register operations using, for example, the protocol
of Tromp and Vitányi [TV02]. The expected cost for a process to acquire
a name is then O(log n) (using an AKS [AKS83] sorting network). A more
complicated construction in the Alistarh et al. paper shows how to make
this adaptive, giving an expected cost of O(log k) instead.
The problem with using an AKS network is that the AKS result is
non-constructive: what Ajtai, Komlós, and Szemerédi show is that there
is a particular randomized construction of candidate sorting networks that
succeeds in producing a correct sorting network with nonzero (but very small)
probability. Other disturbing features of this result are that we have no
efficient way to test candidate sorting networks (determining if a network of
comparators is in fact a sorting network is co-NP-hard), and the constant
in the big-O for AKS is quite spectacularly huge. So it probably makes
more sense to think of renaming networks as giving renaming in O(log2 n)
time, since this is the most efficient practical sorting network we currently
know about. This has led to efforts to produce O(log k)-work tight renaming
algorithms that don’t depend on AKS. So far this has not worked out in the
standard shared-memory model, even allowing test-and-sets.3
The use of test-and-sets to route processes to particular names is similar to
the line of test-and-sets proposed by Moir and Anderson [MA95] as described
in §24.4.5. Some differences between that protocol and renaming networks
is that renaming networks do not by themselves give fetch-and-increment
(although Alistarh et al. show how to build fetch-and-increment on top of
3
The closest to this so far is an algorithm of Berenbrink et al. [BBE+ 15], who use an
extended model that incorporates an extra primitive called a τ -register, which is basically
a collection of 2 log n test-and-set objects that are restricted so that at most τ < 2 log n of
them can be set at a time. Adding this primitive to the model is not entirely cheating, as
the authors make a case that it could be plausibly implemented in hardware. But it does
mean that we don’t know what happens if we don’t have this additional primitive.
CHAPTER 24. RENAMING 238

renaming networks at a small additional cost), and renaming networks do

not provide any mechanism for releasing names. The question of whether it
is possible to get cheap long-lived strong renaming is still open.

24.5.4 Randomized loose renaming

Loose renaming should be easier than strong renaming, and using a random-
ized algorithm it essentially reduces to randomized load balancing. A basic
approach is to use 2n names, and guard each with a test-and-set; because
less than half of the names are taken at any given time, each process gets a
name after O(1) tries and the most expensive renaming operation over all n
processes takes O(log n) expected steps.
A more sophisticated version of this strategy, which appears in [AAGW13],
uses n(1 + ) output names to get O(log log n) maximum steps. The intuition
for why this works is if n processes independently choose one of cn names
uniformly at random, then the expected number of collisions—pairs of
processes that choose the same name—is n2 /cn, or about n/2c. This may

seem like only a constant-factor improvement, but if we instead look at the

ratio between the survivors n/2c and the number of allocated names cn, we
have now moved from 1/c to 1/2c2 . The 2 gives us some room to reduce the
number of names in the next round, to cn/2, say, while still keeping a 1/c2
ratio of survivors to names.
So the actual renaming algorithm consists of allocating cn/2i names to
round i, and squaring the ratio of survivors to names in each rounds. It only
takes O(log log n) rounds to knock the ratio of survivors to names below
1/n, so at this point it is likely that all processes will have finished. At the
same time, the sum over all rounds of the allocated names forms a geometric
series, so only O(n) names are needed altogether.
Swept under the carpet here is a lot of careful analysis of the probabilities.
Unlike what happens with sifters (see §23.7), Jensen’s inequality goes the
wrong way here, so some additional technical tricks are needed (see the paper
for details). But the result is that only O(log log n) rounds are to assign
every process a name with high probability, which is the best value currently
known.
There is a rather weak lower bound in the Alistarh et al. paper that shows
that Ω(log log n) steps are needed for some process in the worst case, under
the assumption that the renaming algorithm uses only test-and-set objects
and that a process acquires a name as soon as it wins some test-and-set
object. This does not give a lower bound on the problem in general, and
indeed the renaming-network based algorithms discussed previously do not
CHAPTER 24. RENAMING 239

have this property. So the question of the exact complexity of randomized

loose renaming is still open.
Chapter 25

Software transactional
memory

Last updated 2011. Some material may be out of date. If you are interested
in software transactional memory from a theoretical perspective, there is a
more recent survey on this material by Attiya [Att14], available at http:
// www. eatcs. org/ images/ bulletin/ beatcs112. pdf .

Software transactional memory, or STM for short, goes back to

Shavit and Touitou [ST97] based on earlier proposals for hardware support
for transactions by Herlihy and Moss [HM93]. Recently very popular in
programming language circles. We’ll give a high-level description of the
Shavit and Touitou results; for full details see the actual paper.
We start with the basic idea of a transaction. In a transaction, I read a
bunch of registers and update their values, and all of these operations appear
to be atomic, in the sense that the transaction either happens completely
or not at all, and serializes with other transactions as if each occurred
instantaneously. Our goal is to implement this with minimal hardware
support, and use it for everything.
Generally we only consider static transactions where the set of memory
locations accessed is known in advance, as opposed to dynamic transac-
tions where it may vary depending on what we read (for example, maybe
we have to follow pointers through some data structure). Static transactions
are easier because we can treat them as multi-word read-modify-write.
Implementations are usually non-blocking: some infinite stream of
transactions succeed, but not necessarily yours. This excludes the simplest
method based on acquiring locks, since we have to keep going even if a

240
CHAPTER 25. SOFTWARE TRANSACTIONAL MEMORY 241

lock-holder crashes, but is weaker than wait-freedom since we can have

starvation.

25.1 Motivation
Some selling points for software transactional memory:

1. We get atomic operations without having to use our brains much.

Unlike hand-coded atomic snapshots, counters, queues, etc., we have a
universal construction that converts any sequential data structure built
on top of ordinary memory into a concurrent data structure. This is
useful since most programmers don’t have very big brains. We also
avoid burdening the programmer with having to remember to lock
things.

2. We can build large shared data structures with the possibility of

concurrent access. For example, we can implement atomic snapshots so
that concurrent updates don’t interfere with each other, or an atomic
queue where enqueues and dequeues can happen concurrently so long
as the queue always has a few elements in it to separate the enqueuers
and dequeuers.

3. We can execute atomic operations that span multiple data structures,

even if the data structures weren’t originally designed to work together,
provided they are all implemented using the STM mechanism. This
is handy in classic database-like settings, as when we want to take $5
from my bank account and put it in yours.

On the other hand, we now have to deal with the possibility that opera-
tions may fail. There is a price to everything.

25.2 Basic approaches

• Locking (not non-blocking). Acquire either a single lock for all of
memory (doesn’t allow much concurrency) or a separate lock for each
memory location accessed. The second approach can lead to deadlock
if we aren’t careful, but we can prove that if every transaction acquires
locks in the same order (e.g., by increasing memory address), then
we never get stuck: we can order the processes by the highest lock
acquired, and somebody comes out on top. Note that acquiring locks in
CHAPTER 25. SOFTWARE TRANSACTIONAL MEMORY 242

increasing order means that I have to know which locks I want before
I acquire any of them, which may rule out dynamic transactions.

• Single-pointer compare-and-swap (called ”Herlihy’s method” in [ST97],

because of its earlier use for constructing concurrent data structures
by Herlihy [Her93]). All access to the data structure goes through
a pointer in a CAS. To execute a transaction, I make my own copy
of the data structure, update it, and then attempt to redirect the
pointer. Advantages: trivial to prove that the result is linearizable (the
pointer swing is an atomic action) and non-blocking (somebody wins
the CAS); also, the method allows dynamic transactions (since you can
do anything you want to your copy). Disadvantages: There’s a high
overhead of the many copies,1 and the single-pointer bottleneck limits
concurrency even when two transactions use disjoint parts of memory.

• Multiword RMW: This is the approach suggested by Shavit and Touitou,

which most subsequent work follows. As usually implemented, it only
works for static transactions. The idea is that I write down what
registers I plan to update and what I plan to do to them. I then
attempt to acquire all the registers. If I succeed, I update all the values,
store the old values, and go home. If I fail, it’s because somebody else
already acquired one of the registers. Since I need to make sure that
somebody makes progress (I may be the only process left alive), I’ll
help that other process finish its transaction if possible. Advantages:
allows concurrency between disjoint transactions. Disadvantages: re-
quires implementing multi-word RMW—in particular, requires that
any process be able to understand and simulate any other process’s
transactions. Subsequent work often simplifies this to implementing
multi-word CAS, which is sufficient to do non-blocking multi-word
RMW since I can read all the registers I need (without any locking)
and then do a CAS to update them (which fails only if somebody else
succeeded).

25.3 Implementing multi-word RMW

We’ll give a sketchy description of Shavit and Touitou’s method [ST97],
which essentially follows the locking approach but allows other processes to
help dead ones so that locks are always released.
1
This overhead can be reduced in many cases by sharing components, a subject that
has seen much work in the functional programming literature. See for example [Oka99].
CHAPTER 25. SOFTWARE TRANSACTIONAL MEMORY 243

The synchronization primitive used is LL/SC: LL (load-linked) reads

a register and leaves our id attached to it, SC (store-conditional) writes a
register only if our id is still attached, and clears any other id’s that might
also be attached. It’s easy to build a 1-register CAS (CAS1) out of this,
though Shavit and Touitou exploit some additional power of LL/SC.

25.3.1 Overlapping LL/SC

The particular trick that gets used in the Shavit-Touitou protocol is to
use two overlapping LL/SC pairs to do a CAS-like update on one memory
location while checking that another memory location hasn’t changed. The
purpose of this is to allow multiple processes to work on the same transaction
(which requires the first CAS to avoid conflicts with other transactions) while
making sure that slow processes don’t cause trouble by trying to complete
transactions that have already finished (the second check).
To see this in action, suppose we have a register r that we want to do
a CAS on, while checking that a second register status is ⊥ (as opposed to
success or failure). If we execute the code fragment in Algorithm 25.1, it will
succeed only if nobody writes to status between its LL and SC and similarly
for r; if this occurs, then at the time of LL(r), we know that status = ⊥, and
we can linearize the write to r at this time if we restrict all access to r to go
through LL/SC.

1 if LL(status) = ⊥ then
2 if LL(r) = oldValue then
3 if SC(status, ⊥) = true then
4 SC(r, newValue)

Algorithm 25.1: Overlapping LL/SC

25.3.2 Representing a transaction

Transactions are represented by records rec. Each such record consists of a
status component that describes how far the transaction has gotten (needed
to coordinate cooperating processes), a version component that distinguishes
between versions that may reuse the same space (and that is used to shut
down the transaction when complete), a stable component that indicates
when the initialization is complete, an Op component that describes the
RMW to be performance, an array addresses[] of pointers to the arguments
CHAPTER 25. SOFTWARE TRANSACTIONAL MEMORY 244

to the RMW, and an array oldValues[] of old values at these addresses (for
the R part of the RMW). These are all initialized by the initiator of the
transaction, who will be the only process working on the transaction until it
starts acquiring locks.

25.3.3 Executing a transaction

Here we give an overview of a transaction execution:

1. Initialize the record rec for the transaction. (Only the initiator does
this.)

2. Attempt to acquire ownership of registers in addresses[]. See the

AcquireOwnerships code in the paper for details. The essential idea is
that we want to set the field owner[r] for each memory location r that
we need to lock; this is done using an overlapping LL/SC as described
above so that we only set owner[r] if (a) r is currently unowned, and
(b) nothing has happened to rec.status or rec.version. Ownership is
acquired in order of increasing memory address; if we fail to acquire
ownership for some r, our transaction fails. In case of failure, we set
rec.status to failure and release all the locks we’ve acquired (checking
rec.version in the middle of each LL/SC so we don’t release locks for
a later version using the same record). If we are the initiator of this
transaction, we will also go on to attempt to complete the transaction
that got in our way.

3. Do a LL on rec.status to see if AcquireOwnerships succeeded. If so,

update the memory, store the old results in oldValues, and release the
ownerships. If it failed, release ownership and help the next transaction
as described above.

Note that only an initiator helps; this avoids a long chain of helping and
limits the cost of each attempted transaction to the cost of doing two full
transactions, while (as shown below) still allowing some transaction to finish.

25.3.4 Proof of linearizability

Intuition is:

• Linearizability follows from the linearizability of the locking protocol:

acquiring ownership is equivalent to grabbing a lock, and updates occur
only when all registers are locked.
CHAPTER 25. SOFTWARE TRANSACTIONAL MEMORY 245

• Complications come from (a) two or more processes trying to complete

the same transaction and (b) some process trying to complete an old
transaction that has already terminated. For the first part we just
make sure that the processes don’t interfere with each other, e.g. I am
happy when trying to acquire a location if somebody else acquires it for
the same transaction. For the second part we have to check rec.status
and rec.version before doing just about anything. See the pseudocode
in the paper for details on how this is done.

25.3.5 Proof of non-blockingness

To show that the protocol is non-blocking we must show that if an unbounded
number of transactions are attempted, one eventually succeeds. First observe
that in order to fail, a transaction must be blocked by another transaction
that acquired ownership of a higher-address location than it did; eventually
we run out of higher-address locations, so there is some transaction that
doesn’t fail. Of course, this transaction may not succeed (e.g., if its initiator
dies), but either (a) it blocks some other transaction, and that transaction’s
initiator will complete it or die trying, or (b) it blocks no future transactions.
In the second case we can repeat the argument for the n − 1 surviving
processes to show that some of them complete transactions, ignoring the
stalled transaction from case (b).

25.4 Improvements
One downside of the Shavit and Touitou protocol is that it uses LL/SC very
aggressively (e.g., with overlapping LL/SC operations) and uses non-trivial
(though bounded, if you ignore the ever-increasing version numbers) amounts
of extra space. Subsequent work has aimed at knocking these down; for
example a paper by Harris, Fraser, and Pratt [HFP02] builds multi-register
CAS out of single-register CAS with O(1) extra bits per register. The proof
of these later results can be quite involved; Harris et al., for example, base
their algorithm on an implementation of 2-register CAS whose correctness
has been verified only by machine (which may be a plus in some views).

25.5 Limitations
There has been a lot of practical work on STM designed to reduce overhead
on real hardware, but there’s still a fair bit of overhead. On the theory side,
CHAPTER 25. SOFTWARE TRANSACTIONAL MEMORY 246

a lower bound of Attiya, Hillel, and Milani [AHM09] shows that any STM
system that guarantees non-interference between non-overlapping RMW
transactions has the undesirable property of making read-only transactions
as expensive as RMW transactions: this conflicts with the stated goals
of many practical STM implementations, where it is assumed that most
transactions will be read-only (and hopefully cheap). So there is quite a bit
of continuing research on finding the right trade-offs.
Chapter 26

Obstruction-freedom

Last updated 2011. Some material may be out of date. In particular: §26.3 has
not been updated to include some more recent results [ACHS13, GHHW13];
and §26.4 mostly follows the conference version [FHS05] of the Ellen-Hendler-
Shavit paper and omits stronger results from the journal version [EHS12].

The gold standard for shared-memory objects is wait-freedom: I can

finish my operation in a bounded number of steps no matter what anybody
else does. Like the gold standard in real life, this can be overly constraining.
So researchers have developed several weaker progress guarantees that are
nonetheless useful. The main ones are:

Lock-freedom An implementation is lock-free if infinitely many opera-

tions finish in any infinite execution. In simpler terms, somebody always
makes progress, but maybe not you. (Also called non-blocking.)

Obstruction-freedom An implementation is obstruction-free if, starting

from any reachable configuration, any process can finish in a bounded
number of steps if all of the other processes stop. This definition was
proposed in 2003 by Herlihy, Luchangco, and Moir [HLM03]. In lower
bounds (e.g., the Jayanti-Tan-Toueg bound described in Chapter 20)
essentially the same property is often called solo-terminating.

Both of these properties exclude traditional lock-based algorithms, where

some process grabs a lock, updates the data structure, and then release the
lock; if this process halts, no more operations finish. Both properties are
also weaker than wait-freedom. It is not hard to show that lock-freedom is a

247
CHAPTER 26. OBSTRUCTION-FREEDOM 248

stronger condition that obstruction-freedom: given a lock-free implementa-

tion, if we can keep some single process running forever in isolation, we get
an infinite execution with only finitely many completed operations. So we
have a hierarchy: wait-free > lock-free > obstruction-free > locking.

26.1 Why build obstruction-free algorithms?

The pitch is similar to the pitch for building locking algorithms: an obstruction-
free algorithm might be simpler to design, implement, and reason about
than a more sophisticated algorithm with stronger properties. Unlike locking
algorithms, an obstruction-free algorithm won’t fail because some process
dies holding the lock; instead, it fails if more than one process runs the
algorithm at the same time. This possibility may be something we can
avoid by building a contention manager, a high-level protocol that detects
contention and delays some processes to avoid it (say, using randomized
exponential back-off).

26.2 Examples
26.2.1 Lock-free implementations
Pretty much anything built using compare-and-swap or LL/SC ends up
being lock-free. A simple example would be a counter, where an increment
operation does

1 x ← LL(C)
2 SC(C, x + 1)

This is lock-free (the only way to prevent a store-conditional from suc-

ceeding is if some other store-conditional succeeds, giving infinitely many
successful increments) but not wait-free (I can starve). It’s also obstruction-
free, but since it’s already lock-free we don’t care about that.

26.2.2 Double-collect snapshots

Similarly, suppose we are doing atomic snapshots. We know that there
exist wait-free implementations of atomic snapshots, but they are subtle and
confusing. So we want to do something simpler, and hope that we at least
get obstruction-freedom.
CHAPTER 26. OBSTRUCTION-FREEDOM 249

If we do double-collects, that is, we have updates just write to a register

and have snapshots repeatedly collect until they get two collects in a row
with the same values, then any snapshot that finishes is correct (assuming no
updaters ever write the same value twice, which we can enforce with nonces).
This isn’t wait-free, because we can keep a snapshot going forever by doing
a lot of updates. It is lock-free, because we have to keep doing updates to
make this happen.
We can make this merely obstruction-free if we work hard (there is no rea-
son to do this, but it illustrates the difference between lock-freedom—good—and
obstruction-freedom—not so good). Suppose that every process keeps a count
of how many collects it has done in a register that is included in other pro-
cess’s collects (but not its own). Then two concurrent scans can stall each
other forever (the implementation is not lock-free), but if only one is running
it completes two collects in O(n) operations without seeing any changes (it
is obstruction-free).

26.2.3 Software transactional memory

Similar things happen with software transactional memory (see Chapter 25).
Suppose that I have an implementation of multiword compare-and-swap, and
I want to carry out a transaction. I read all the values I need, then execute
an MCAS operation that only updates if these values have not changed. The
resulting algorithm is lock-free (if my transaction fails, it’s because some
update succeeded). If however I am not very clever and allow some values to
get written outside of transactions, then I might only be obstruction-free.

26.2.4 Obstruction-free test-and-set

Algorithm 26.1 gives an implementation of 2-process test-and-set from atomic
registers that is obstruction-free; this demonstrates that obstruction-freedom
lets us evade the wait-free impossibility results implied by the consensus
hierarchy ([Her91b], discussed in Chapter 18).
The basic idea goes back to the racing counters technique used in
consensus protocols starting with Chor, Israeli, and Li [CIL94], and there is
some similarity to a classic randomized wait-free test-and-set due to Tromp
and Vitányi [TV02]. Each process keeps a position x in memory that it also
stores from time to time in its register a[i]. If a process gets 2 steps ahead
of the other process (as observed by comparing x to a[i − 1], it wins the
test-and-set; if a process falls one or more steps behind, it (eventually) loses.
To keep space down and guarantee termination in bounded time, all values
CHAPTER 26. OBSTRUCTION-FREEDOM 250

are tracked modulo 5.

1 x←0
2 while true do
3 δ ← x − a[1 − i]
4 if δ = 2 (mod 5) then
5 return 0
6 else if δ = −1 (mod 5) do
7 return 1
8 else
9 x ← (x + 1) mod 5
10 a[i] ← x

Algorithm 26.1: Obstruction-free 2-process test-and-set

Why this works: observe that whenever a process computes δ, x is equal

to a[i]; so δ is always an instantaneous snapshot of a[i] − a[1 − i]. If I observe
δ = 2 and return 0, your next read will either show you δ = −2 or δ = −1
(depending on whether you increment a[1 − i] after my read). In the latter
case, you return 1 immediately; in the former, you return after one more
increment (and more importantly, you can’t return 0). Alternatively, if I ever
observe δ = −1, your next read will show you either δ = 1 or δ = 2; in either
case, you will eventually return 0. (We chose 5 as a modulus because this is
the smallest value that makes the cases δ = 2 and δ = −2 distinguishable.)
We can even show that this is linearizable, by considering a solo execution
in which the lone process takes two steps and returns 0 (with two processes,
solo executions are the only interesting case for linearizability).
However, Algorithm 26.1 is not wait-free or even lock-free: if both
processes run in lockstep, they will see δ = 0 forever. But it is obstruction-
free. If I run by myself, then whatever value of δ I start with, I will see −1
or 2 after at most 6 operations.1
This gives an obstruction-free step complexity of 6, where the
obstruction-free step complexity is defined as the maximum number of
operations any process can take after all other processes stop. Note that our
usual wait-free measures of step complexity don’t make a lot of sense for
obstruction-free algorithms, as we can expect a sufficiently cruel adversary
to be able to run them up to whatever value he likes.
1
The worst case is where an increment by my fellow process leaves δ = −1 just before
my increment.
CHAPTER 26. OBSTRUCTION-FREEDOM 251

Building a tree of these objects as in §22.2 gives n-process test-and-set

with obstruction-free step complexity O(log n).

26.2.5 An obstruction-free deque

(We probably aren’t going to do this in class.)
So far we don’t have any good examples of why we would want to be
obstruction-free if our algorithm is based on CAS. So let’s describe the case
Herlihy et al. suggested.
A deque is a generalized queue that supports push and pop at both ends
(thus it can be used as either a queue or a stack, or both). A classic problem
in shared-memory objects is to build a deque where operations at one end of
the deque don’t interfere with operations at the other end. While there exist
lock-free implementation with this property, there is a particularly simple
implementation using CAS that is only obstruction-free.
Here’s the idea: we represent the deque as an infinitely-long array of
compare-and-swap registers (this is a simplification from the paper, which
gives a bounded implementation of a bounded deque). The middle of the
deque holds the actual contents. To the right of this region is an infinite
sequence of right null (RN) values, which are assumed never to appear
as a pushed value. To the left is a similar infinite sequence of left null
(LN) values. Some magical external mechanism (called an oracle in the
paper) allows processes to quickly find the first null value at either end of
the non-null region; the correctness of the protocol does not depend on the
properties of the oracle, except that it has to point to the right place at least
some of the time in a solo execution. We also assume that each cell holds a
version number whose only purpose is to detect when somebody has fiddled
with the cell while we aren’t looking (if we use LL/SC, we can drop this).
Code for rightPush and rightPop is given in Algorithm 26.2 (the code
for leftPush and leftPop is symmetric).
It’s easy to see that in a solo execution, if the oracle doesn’t lie, either
operation finishes and returns a plausible value after O(1) operations. So
the implementation is obstruction-free. But is it also correct?
To show that it is, we need to show that any execution leaves the deque
in a sane state, in particular that it preserves the invariant that the deque
consists of left-nulls followed by zero or more values followed by right-nulls,
and that the sequence of values in the queue is what it should be.
This requires a detailed case analysis of which operations interfere with
each other, which can be found in the original paper. But we can give some
intuition here. The two CAS operations in rightPush or rightPop succeed
CHAPTER 26. OBSTRUCTION-FREEDOM 252

1 procedure rightPush(v)
2 while true do
3 k ← oracle(right)
4 prev ← a[k − 1]
5 next ← a[k]
6 if prev.value 6= RN and next.value = RN then
7 if CAS(a[k − 1], prev, [prev.value, prev.version + 1]) then
8 if CAS(a[k], next, [v, next.version + 1]) then
9 we win, go home

10 procedure rightPop()
11 while true do
12 k ← oracle(right)
13 cur ← a[k − 1]
14 next ← a[k]
15 if cur.value 6= RN and next.value = RN then
16 if cur.value = LN and A[k − 1] = cur then
17 return empty
18 else if CAS(a[k], next, [RN, next.version + 1]) do
19 if CAS(a[k − 1], cur, [RN, cur.version + 1]) then
20 return cur.value

Algorithm 26.2: Obstruction-free deque

CHAPTER 26. OBSTRUCTION-FREEDOM 253

only if neither register was modified between the preceding read and the
CAS. If both registers are unmodified at the time of the second CAS, then
the two CAS operations act like a single two-word CAS, which replaces the
previous values (top, RN) with (top, value) in rightPush or (top, value) with
(top, RN) in rightPop; in either case the operation preserves the invariant.
So the only way we get into trouble is if, for example, a rightPush does a
CAS on a[k − 1] (verifying that it is unmodified and incrementing the version
number), but then some other operation changes a[k − 1] before the CAS on
a[k]. If this other operation is also a rightPush, we are happy, because it
must have the same value for k (otherwise it would have failed when it saw
a non-null in a[k − 1]), and only one of the two right-pushes will succeed
in applying the CAS to a[k]. If the other operation is a rightPop, then it
can only change a[k − 1] after updating a[k]; but in this case the update to
a[k] prevents the original right-push from changing a[k]. With some more
tedious effort we can similarly show that any interference from leftPush or
leftPop either causes the interfering operation or the original operation to
fail. This covers 4 of the 16 cases we need to consider. The remaining cases
will be brushed under the carpet to avoid further suffering.

26.3 Boosting obstruction-freedom to wait-freedom

Naturally, having an obstruction-free implementation of some object is
not very helpful if we can’t guarantee that some process eventually gets
its unobstructed solo execution. In general, we can’t expect to be able
to do this without additional assumptions; for example, if we could, we
could solve consensus using a long sequence of adopt-commit objects with
no randomization at all.2 So we need to make some sort of assumption
about timing, or find somebody else who has already figured out the right
assumption to make.
Those somebodies turn out to be Faith Ellen Fich, Victor Luchangco,
Mark Moir, and Nir Shavit, who give an algorithm for boosting obstruction-
freedom to wait-freedom [FLMS05]. The timing assumption is unknown-
bound semisynchrony, which means that in any execution there is some
maximum ratio R between the shortest and longest time interval between
any two consecutive steps of the same non-faulty process, but the processes
2
This fact was observed by Herlihy et al. [HLM03] in their original obstruction-free
paper; it also implies that there exists a universal obstruction-free implementation of
anything based on Herlihy’s universal construction.
CHAPTER 26. OBSTRUCTION-FREEDOM 254

don’t know what this ratio is.3 In particular, if I can execute more than R
steps without you doing anything, I can reasonably conclude that you are
dead—the semisynchrony assumption thus acts as a failure detector.
The fact that R is unknown might seem to be an impediment to using
this failure detector, but we can get around this. The idea is to start with
a small guess for R; if a process is suspected but then wakes up again, we
increment the guess. Eventually, the guessed value is larger than the correct
value, so no live process will be falsely suspected after this point. Formally,
this gives an eventually perfect (♦P ) failure detector, although the algorithm
does not specifically use the failure detector abstraction.
To arrange for a solo execution, when a process detects a conflict (because
its operation didn’t finish quickly), it enters into a “panic mode” where pro-
cesses take turns trying to finish unmolested. A fetch-and-increment register
is used as a timestamp generator, and only the process with the smallest
timestamp gets to proceed. However, if this process is too sluggish, other
processes may give up and overwrite its low timestamp with ∞, temporarily
ending its turn. If the sluggish process is in fact alive, it can restore its low
timestamp and kill everybody else, allowing it to make progress until some
other process declares it dead again.
The simulation works because eventually the mechanism for detecting
dead processes stops suspecting live ones (using the technique described
above), so the live process with the winning timestamp finishes its operation
without interference. This allows the next process to proceed, and eventually
all live processes complete any operation they start, giving the wait-free
property.
The actual code is in Algorithm 26.3. It’s a rather long algorithm but
most of the details are just bookkeeping.
The preamble before entering PANIC mode is a fast-path computation
that allows a process that actually is running in isolation to skip testing
any timestamps or doing any extra work (except for the one register read of
PANIC). The assumption is that the constant B is set high enough that any
process generally will finish its operation in B steps without interference. If
there is interference, then the timestamp-based mechanism kicks in: we grab
a timestamp out of the convenient fetch-and-add register and start slugging
it out with the other processes.
(A side note: while the algorithm as presented in the paper assumes
a fetch-and-add register, any timestamp generator that delivers increasing
3
This is a much older model, which goes back to a famous paper of Dwork, Lynch, and
Stockmeyer [DLS88].
CHAPTER 26. OBSTRUCTION-FREEDOM 255

1 if ¬PANIC then
2 execute up to B steps of the underlying algorithm
3 if we are done then return
4 PANIC ← true // enter panic mode
5 myTimestamp ← fetchAndIncrement()
6 A[i] ← 1 // reset my activity counter
7 while true do
8 T [i] ← myTimestamp
9 minTimestamp ← myTimestamp; winner ← i
10 for j ← 1 . . . n, j 6= i do
11 otherTimestamp ← T [j]
12 if otherTimestamp < minTimestamp then
13 T [winner] ← ∞ // not looking so winning any more
14 minTimestamp ← otherTimestamp; winner ← j
15 else if otherTimestamp < ∞ do
16 T [j] ← ∞

17 if i = winner then
18 repeat
19 execute up to B steps of the underlying algorithm
20 if we are done then
21 T [i] ← ∞
22 PANIC ← false
23 return
24 else
25 A[i] ← A[i] + 1
26 PANIC ← true
27 until T [i] = ∞
28 repeat
29 a ← A[winner]
30 wait a steps
31 winnerTimestamp ← T [winner]
32 until a = A[winner] or winnerTimestamp 6= minTimestamp
33 if winnerTimestamp = minTimestamp then
34 T [winner] ← ∞ // kill winner for inactivity

Algorithm 26.3: Obstruction-freedom booster from [FLMS05]

CHAPTER 26. OBSTRUCTION-FREEDOM 256

values over time will work. So if we want to limit ourselves to atomic registers,
we could generate timestamps by taking snapshots of previous timestamps,
adding 1, and appending process ids for tie-breaking.)
Once I have a timestamp, I try to knock all the higher-timestamp processes
out of the way (by writing ∞ to their timestamp registers). If I see a smaller
timestamp than my own, I’ll drop out myself (T [i] ← ∞), and fight on behalf
of its owner instead. At the end of the j loop, either I’ve decided I am the
winner, in which case I try to finish my operation (periodically checking T [i]
to see if I’ve been booted), or I’ve decided somebody else is the winner, in
which case I watch them closely and try to shut them down if they are too
slow (T [winner] ← ∞). I detect slow processes by inactivity in A[winner];
similarly, I signal my own activity by incrementing A[i]. The value in A[i]
is also used as an increasing guess for the time between increments of A[i];
eventually this exceeds the R(B + O(1)) operations that I execute between
incrementing it.
We still need to prove that this all works. The essential idea is to show
that whatever process has the lowest timestamp finishes in a bounded number
of steps. To do so, we need to show that other processes won’t be fighting it
in the underlying algorithm. Call a process active if it is in the loop guarded
by the “if i = winner” statement. Lemma 1 from the paper states:

Lemma 26.3.1 ([FLMS05, Lemma 1]). If processes i and j are both active,
then T [i] = ∞ or T [j] = ∞.

Proof. Assume without loss of generality that i last set T [i] to myTimestamp
in the main loop after j last set T [j]. In order to reach the active loop, i
must read T [j]. Either T [j] = ∞ at this time (and we are done, since only j
can set T [j] < ∞), or T [j] is greater than i’s timestamp (or else i wouldn’t
think it’s the winner). In the second case, i sets T [j] = ∞ before entering
the active loop, and again the claim holds.

The next step is to show that if there is some process i with a minimum
timestamp that executes infinitely many operations, it increments A[i] in-
finitely often (thus eventually making the failure detector stop suspecting it).
This gives us Lemma 2 from the paper:

Lemma 26.3.2 ([FLMS05, Lemma 2]). Consider the set of all processes that
execute infinitely many operations without completing an operation. Suppose
this set is non-empty, and let i hold the minimum timestamp of all these
processes. Then i is not active infinitely often.
CHAPTER 26. OBSTRUCTION-FREEDOM 257

Proof. Suppose that from some time on, i is active forever, i.e., it never
leaves the active loop. Then T [i] < ∞ throughout this interval (or else i
leaves the loop), so for any active j, T [j] = ∞ by the preceding lemma. It
follows that any active T [j] leaves the active loop after B + O(1) steps of j
(and thus at most R(B + O(1)) steps of i). Can j re-enter? If j’s timestamp
is less than i’s, then j will set T [i] = ∞, contradicting our assumption. But
if j’s timestamp is greater than i’s, j will not decide it’s the winner and
will not re-enter the active loop. So now we have i alone in the active loop.
It may still be fighting with processes in the initial fast path, but since i
sets PANIC every time it goes through the loop, and no other process resets
PANIC (since no other process is active), no process enters the fast path after
some bounded number of i’s steps, and every process in the fast path leaves
after at most R(B + O(1)) of i’s steps. So eventually i is in the loop alone
forever—and obstruction-freedom means that it finishes its operation and
leaves. This contradicts our initial assumption that i is active forever.

So now we want to argue that our previous assumption that there exists
a bad process that runs forever without winning leads to a contradiction, by
showing that the particular i from Lemma 26.3.2 actually finishes (note that
Lemma 26.3.2 doesn’t quite do this—we only show that i finishes if it stays
active long enough, but maybe it doesn’t stay active).
Suppose i is as in Lemma 26.3.2. Then i leaves the active loop infinitely
often. So in particular it increments A[i] infinitely often. After some finite
number of steps, A[i] exceeds the limit R(B + O(1)) on how many steps some
other process can take between increments of A[i]. For each other process j,
either j has a lower timestamp than i, and thus finishes in a finite number of
steps (from the premise of the choice of i), or j has a higher timestamp than
i. Once we have cleared out all the lower-timestamp processes, we follow the
same logic as in the proof of Lemma 26.3.2 to show that eventually (a) i sets
T [i] < ∞ and PANIC = true, (b) each remaining j observes T [i] < ∞ and
PANIC = true and reaches the waiting loop, (c) all such j wait long enough
(since A[i] is now very big) that i can finish its operation. This contradicts
the assumption that i never finishes the operation and completes the proof.

26.3.1 Cost
If the parameters are badly tuned, the potential cost of this construction is
quite bad. For example, the slow increment process for A[i] means that the
time a process spends in the active loop even after it has defeated all other
processes can be as much as the square of the time it would normally take
CHAPTER 26. OBSTRUCTION-FREEDOM 258

to complete an operation alone—and every other process may pay R times

this cost waiting. This can be mitigated to some extent by setting B high
enough that a winning process is likely to finish in its first unmolested pass
through the loop (recall that it doesn’t detect that the other processes have
reset T [i] until after it makes its attempt to finish). An alternative might
be to double A[i] instead of incrementing it at each pass through the loop.
However, it is worth noting (as the authors do in the paper) that nothing
prevents the underlying algorithm from incorporating its own contention
management scheme to ensure that most operations complete in B steps
and PANIC mode is rarely entered. So we can think of the real function of
the construction as serving as a backstop to some more efficient heuristic
approach that doesn’t necessarily guarantee wait-free behavior in the worst
case.

26.4 Lower bounds for lock-free protocols

So far we have seen that obstruction-freedom buys us an escape from the
impossibility results that plague wait-free constructions, while still allowing
practical implementations of useful objects under plausible timing assump-
tions. Yet all is not perfect: it is still possible to show non-trivial lower
bounds on the costs of these implementations in the right model. We will
present one of these lower bounds, the linear-contention lower bound of Ellen,
Hendler, and Shavit [EHS12].4 First we have to define what is meant by
contention.

26.4.1 Contention
A limitation of real shared-memory systems is that physics generally won’t
permit more than one process to do something useful to a shared object
at a time. This limitation is often ignored in computing the complexity of
a shared-memory distributed algorithm (and one can make arguments for
ignoring it in systems where communication costs dominate update costs in
the shared-memory implementation), but it is useful to recognize it if we
can’t prove lower bounds otherwise. Complexity measures that take the cost
of simultaneous access into account go by the name of contention.
The particular notion of contention used in the Ellen et al. paper is an
adaptation of the contention measure of Dwork, Herlihy, and Waarts [DHW97].
4
The result first appeared in FOCS in 2005 [FHS05], with a small but easily fixed bug in
the definition of the class of objects the proof applies to. We’ll use the corrected definition
from the journal version.
CHAPTER 26. OBSTRUCTION-FREEDOM 259

The idea is that if I access some shared object, I pay a price in memory
stalls for all the other processes that are trying to access it at the same time
but got in first. In the original definition, given an execution of the form
Aφ1 φ2 . . . φk φA0 , where all operations φi are applied to the same object as φ,
and the last operation in A is not, then φk incurs k memory stalls. Ellen et
al. modify this to only count sequences of non-trivial operations, where an
operation is non-trivial if it changes the state of the object in some states
(e.g., writes, increments, compare-and-swap—but not reads). Note that this
change only strengthens the bound they eventually prove, which shows that
in the worst case, obstruction-free implementations of operations on objects
in a certain class incur a linear number of memory stalls (possibly spread
across multiple base objects).

26.4.2 The class G

The Ellen et al. bound is designed to be as general as possible, so the
authors define a class G of objects to which it applies. As is often the case
in mathematics, the underlying meaning of G is “a reasonably large class
of objects for which this particular proof works,” but the formal definition
is given in terms of when certain operations of the implemented object are
affected by the presence or absence of other operations—or in other words,
when those other operations need to act on some base object in order to let
later operations know they occurred.
An object is in class G if it has some operation Op and initial state s
such that for any two processes p and q and every sequence of operations
AφA0 , where

1. φ is an instance of Op executed by p,

2. no operation in A or A0 is executed by p,

3. no operation in A0 is executed by q, and

4. no two operations in A0 are executed by the same process;

then there exists a sequence of operations Q by q such that for every sequence
HφH 0 where

1. HH 0 is an interleaving of Q and the sequences AA0 |r for each process

r,

2. H 0 contains no operations of q, and

CHAPTER 26. OBSTRUCTION-FREEDOM 260

3. no two operations in H 0 are executed by the same process;

then the return value of φ to p changes depending on whether it occurs after

Aφ or Hφ.
This is where “makes the proof work” starts looking like a much simpler
definition. The intuition is that deep in the guts of the proof, we are going to
be injecting some operations of q into an existing execution (hence adding Q),
and we want to do it in a way that forces q to operate on some object that p
is looking at (hence the need for Aφ to return a different value from Hφ),
without breaking anything else that is going on (all the rest of the conditions).
The reason for pulling all of these conditions out of the proof into a separate
definition is that we also want to be able to show that particular classes of
real objects satisfy the conditions required by the proof, without having to
put a lot of special cases into the proof itself.

Lemma 26.4.1. A mod-m fetch-and-increment object, with m ≥ n, is in G.

Proof. This is a classic proof-by-unpacking-the-definition. Pick some ex-

ecution AφA0 satisfying all the conditions, and let a be the number of
fetch-and-increments in A and a0 the number in A0 . Note a0 ≤ n − 2, since
all operations in A0 are by different processes.
Now let Q be a sequence of n − a0 − 1 fetch-and-increments by q, and let
HH 0 be an interleaving of Q and the sequences AA0 |r for each r, where H 0
includes no two operation of the same process and no operations at all of
q. Let h, h0 be the number of fetch-and-increments in H, H 0 , respectively.
Then h + h0 = a + a0 + (n − a0 − 1) = n + a − 1 and h0 ≤ n − 2 (since H 0
contains at most one fetch-and-increment for each process other than p and
q). This gives h ≥ (n + a + 1) − (n − 2) = a + 1 and h ≤ n + a − 1, and the
return value of φ after Hφ is somewhere in this range mod m. But none of
these values is equal to a mod m (that’s why we specified m ≥ n, although
as it turns out m ≥ n − 1 would have been enough), so we get a different
return value from Hφ than from Aφ.

As a corollary, we also get stock fetch-and-increment registers, since we

can build mod-m registers from them by taking the results mod m.
A second class of class-G objects is obtained from snapshot:

Lemma 26.4.2. Single-writer snapshot objects are in G.5

5
For the purposes of this lemma, “single-writer” means that each segment can be
written to by only one process, not that there is only one process that can execute update
operations.
CHAPTER 26. OBSTRUCTION-FREEDOM 261

Proof. Let AφA0 be as in the definition, where φ is a scan operation. Let Q

consist of a single update operation by q that changes its segment. Then
in the interleaved sequence HH 0 , this update doesn’t appear in H 0 (it’s
forbidden), so it must be in H. Nobody can overwrite the result of the
update (single-writer!), so it follows that Hφ returns a different snapshot
from Aφ.

26.4.3 The lower bound proof

Theorem 26.4.3 ([EHS12, Theorem 5.2]). For any obstruction-free imple-
mentation of some object in class G from RMW base objects, there is an
execution in which some operation incurs n − 1 stalls.
We can’t do better than n − 1, because it is easy to come up with
implementations of counters (for example) that incur at most n − 1 stalls.
Curiously, we can even spread the stalls out in a fairly arbitrary way over
multiple objects, while still incurring at most n − 1 stalls. For example, a
counter implemented using a single counter (which is a RMW object) gets
exactly n − 1 stalls if n − 1 processes try to increment it at the same time,
delaying the remaining process. At the other extreme, a counter implemented
by doing a collect over n − 1 single-writer registers (also RMW objects) gets
at least n − 1 stalls—distributed as one per register—if each register has a
write delivered to it while the reader waiting to read it during its collect. So
we have to allow for the possibility that stalls are concentrated or scattered
or something in between, as long as the total number adds up at least n − 1.
The proof supposes that the theorem is not true and then shows how to
boost an execution with a maximum number k < n − 1 stalls to an execution
with k + 1 stalls, giving a contradiction. (Alternatively, we can read the
proof as giving a mechanism for generating an (n − 1)-stall execution by
repeated boosting, starting from the empty execution.)
This is pretty much the usual trick: we assume that there is a class of
bad executions, then look for an extreme member of this class, and show that
it isn’t as extreme as we thought. In doing so, we can restrict our attention
to particularly convenient bad executions, so long as the existence of some
bad execution implies the existence of a convenient bad execution.
Formally, the authors define a k-stall execution for process p as an
execution Eσ1 . . . σi where E and σi are sequence of operations such that:
1. p does nothing in E,
2. Sets of processes Sj , j = 1 . . . i, whose union S = ij=1 Sj has size k, are
S

each covering objects Oj after E with pending non-trivial operations,

CHAPTER 26. OBSTRUCTION-FREEDOM 262

3. Each σj consists of p applying events by itself until it is about to apply

an event to Oj , after which each process in Sj accesses Oj , after which
p accesses Oj .

4. All processes not in S are idle after E,

5. p starts at most one operation of the implemented object in σ1 . . . σi ,

and

6. In every extension of E in which p and the processes in S don’t take

steps, no process applies a non-trivial event to any base object accessed
in σ1 . . . σi . (We will call this the weird condition below.)

So this definition includes both the fact that p incurs k stalls and some
other technical details that make the proof go through. The fact that p
incurs k stalls follows from observing that it incurs |Sj | stalls in each segment
σj , since all processes in Sj access Oj just before p does.
Note that the empty execution is a 0-stall execution (with i = 0) by the
definition. This shows that a k-stall execution exists for some k.
Note also that the weird condition is pretty strong: it claims not only
that there are no non-trivial operation on O1 . . . Oi in τ , but also that there
are no non-trivial operations on any objects accessed in σ1 . . . σi , which may
include many more objects accessed by p.6
We’ll now show that if a k-stall execution exists, for k ≤ n − 2, then a
(k +k 0 )-stall execution exists for some k 0 > 0. Iterating this process eventually
produces an (n − 1)-stall execution.
Start with some k-stall execution Eσ1 . . . σi . Extend this execution by
a sequence of operations σ in which p runs in isolation until it finishes its
operation φ (which it may start in σ if it hasn’t done so already), then each
process in S runs in isolation until it completes its operation. Now linearize
the high-level operations completed in Eσ1 . . . σi σ and factor them as AφA0
as in the definition of class G.
Let q be some process not equal to p or contained in any Sj (this is where
we use the assumption k ≤ n − 2). Then there is some sequence of high-level
operations Q of q such that Hφ does not return the same value as Aφ for
any interleaving HH 0 of Q with the sequences of operations in AA0 satisfying
the conditions in the definition. We want to use this fact to shove at least
one more memory stall into Eσ1 . . . σi σ, without breaking any of the other
conditions that would make the resulting execution a (k + k 0 )-stall execution.
6
And here is where I screwed up in class on 2011-11-14, by writing the condition as the
weaker requirement that nobody touches O1 . . . Oi .
CHAPTER 26. OBSTRUCTION-FREEDOM 263

Consider the extension τ of E where q runs alone until it finishes every

operation in Q. Then τ applies no nontrivial events to any base object
accessed in σ1 . . . σk , (from the weird condition on k-stall executions) and
the value of each of these base objects is the same after E and Eτ , and thus
is also the same after Eσ1 . . . σk and Eτ σ1 . . . σk .
Now let σ 0 be the extension of Eτ σ1 . . . σk defined analogously to σ:
p finishes, then each process in each Sj finishes. Let HφH 0 factor the
linearization of Eτ σ1 . . . σi σ 0 . Observe that HH 0 is an interleaving of Q and
the high-level operations in AA0 , that H 0 contains no operations by q (they
all finished in τ , before φ started), and that H 0 contains no two operations by
the same process (no new high-level operations start after φ finishes, so there
is at most one pending operation per process in S that can be linearized
after φ).
Now observe that q does some non-trivial operation in τ to some base
object accessed by p in σ. If not, then p sees the same responses in σ 0 and in
σ, and returns the same value, contradicting the definition of class G.
So does q’s operation in τ cause a stall in σ? Not necessarily: there
may be other operations in between. Instead, we’ll use the existence of q’s
operation to demonstrate the existence of at least one operation, possibly by
some other process we haven’t even encountered yet, that does cause a stall.
We do this by considering the set F of all finite extensions of E that are free
of p and S operations, and look for an operation that stalls p somewhere in
this infinitely large haystack.
Let Oi+1 be the first base object accessed by p in σ that is also accessed
by some non-trivial event in some sequence in F . We will show two things:
first, that Oi+1 exists, and second, that Oi+1 is distinct from the objects
O1 . . . Oi . The first part follows from the fact that τ is in F , and we have
just shown that τ contains a non-trivial operation (by q) on a base object
accessed by p in σ. For the second part, we use the weird condition on k-stall
executions again: since every extension of E in F is ({p} ∪ S)-free, no process
applies a non-trivial event to any base object accessed in σ1 . . . σi , which
includes all the objects O1 . . . Oi .
You’ve probably guessed that we are going to put our stalls in on Oi+1 .
We choose some extension X from F that maximizes the number of processes
with simultaneous pending non-trivial operations on Oi+1 (we’ll call this set
of processes Si+1 and let |Si+1 | be the number k 0 > 0 we’ve been waiting for),
and let E 0 be the minimum prefix of X such that these pending operations
are still pending after EE 0 .
We now look at the properties of EE 0 . We have:
CHAPTER 26. OBSTRUCTION-FREEDOM 264

• EE 0 is p-free (follows from E being p-free and E 0 ∈ F , since everything

in F is p-free).

• Each process in Sj has a pending operation on Oj after EE 0 (it did

after E, and didn’t do anything in E 0 ).

This means that we can construct an execution EE 0 σ1 . . . σi σi+1 that

includes k + k 0 memory stalls, by sending in the same sequences σ1 . . . σi as
before, then appending a new sequence of events where (a) p does all of its
operations in σ up to its first operation on Oi+1 ; then (b) all the processes in
the set Si+1 of processes with pending events on Oi+1 execute their pending
events on Oi+1 ; then (c) p does its first access to Oi+1 from σ. Note that in
addition to giving us k + k 0 memory stalls, σi+1 also has the right structure
for a (k + k 0 )-stall execution. But there is one thing missing: we have to
show that the weird condition on further extensions still holds.
Specifically, letting S 0 = S ∪Si+1 , we need to show that any ({p}∪S 0 )-free
extension α of EE 0 includes a non-trivial access to a base object accessed
in σ1 . . . σi+1 . Observe first that since α is ({p} ∪ S 0 )-free, then E 0 α is
({p} ∪ S)-free, and so it’s in F : so by the weird condition on Eσ1 . . . σi , E 0 α
doesn’t have any non-trivial accesses to any object with a non-trivial access
in σ1 . . . σi . So we only need to squint very closely at σi+1 to make sure it
doesn’t get any objects in there either.
Recall that σi+1 consists of (a) a sequence of accesses by p to objects
already accessed in σ1 . . . σi (already excluded); (b) an access of p to Oi+1 ;
and (c) a bunch of accesses by processes in Si+1 to Oi+1 . So we only need
to show that α includes no non-trivial accesses to Oi+1 . Suppose that it
does: then there is some process that eventually has a pending non-trivial
operation on Oi+1 somewhere in α. If we stop after this initial prefix α0 of α,
we get k 0 + 1 processes with pending operations on Oi+1 in EE 0 α0 . But then
E 0 α0 is an extension of E with k 0 + 1 processes with a simultaneous pending
operation on Oi+1 . This contradicts the choice of X to maximize k 0 . So if
our previous choice was in fact maximal, the weird condition still holds, and
we have just constructed a (k + k 0 )-stall execution. This concludes the proof.

26.4.4 Consequences
We’ve just shown that counters and snapshots have (n − 1)-stall executions,
because they are in the class G. A further, rather messy argument (given in
the Ellen et al. paper) extends the result to stacks and queues, obtaining a
slightly weaker bound of n total stalls and operations for some process in
CHAPTER 26. OBSTRUCTION-FREEDOM 265

the worst case.7 In both cases, we can’t expect to get a sublinear worst-case
bound on time under the reasonable assumption that both a memory stall
and an actual operation takes at least one time unit. This puts an inherent
bound on how well we can handle hot spots for many practical objects, and
means that in an asynchronous system, we can’t solve contention at the
object level in the worst case (though we may be able to avoid it in our
applications).
But there might be a way out for some restricted classes of objects. We saw
in Chapter 21 that we could escape from the Jayanti-Tan-Toueg [JTT00] lower
bound by considering bounded objects. Something similar may happen here:
the Fich-Herlihy-Shavit bound on fetch-and-increments requires executions
with n(n − 1)d + n increments to show n − 1 stalls for some fetch-and-
increment if each fetch-and-increment only touches d objects, and even for
d = log n this is already superpolynomial. The max-register construction
of a counter [AACH12] doesn’t help here, since everybody hits the switch
bit at the top of the max register, giving n − 1 stalls if they all hit it at the
same time. But there might be some better construction that avoids this.

26.4.5 More lower bounds

There are many more lower bounds one can prove on lock-free implementa-
tions, many of which are based on previous lower bounds for stronger models.
We won’t present these in class, but if you are interested, a good place to
start is [AGHK06].

26.5 Practical considerations

Also beyond the scope of what we can do, there is a paper by Fraser
and Harris [FH07] that gives some nice examples of the practical trade-
offs in choosing between multi-register CAS and various forms of software
transactional memory in implementing lock-free data structures.

7
This is out of date: Theorem 6.2 of [EHS12] gives a stronger result than what’s in
[FHS05].
Chapter 27

BG simulation

The Borowsky-Gafni simulation [BG93], or BG simulation for short, is

a deterministic, wait-free algorithm that allows t + 1 processes to collectively
construct a simulated execution of a system of n > t processes of which t may
crash. For both the simulating and simulated system, the underlying shared-
memory primitives are atomic snapshots; these can be replaced by atomic
registers using any standard snapshot algorithm. The main consequence
of the BG simulation is that the question of what decision tasks can be
computed deterministically by an asynchronous shared-memory system that
tolerates t crash failures reduces to the question of what can be computed by
a wait-free system with exactly t + 1 processes. This is an easier problem, and
in principle can be solved exactly using the topological approach described
in Chapter 28.
The intuition for how this works is that the t + 1 simulating processes
solve a sequence of agreement problems to decide what the n simulated
processes are doing; these agreement problems are structured so that the
failure of a simulator stops at most one agreement. So if at most t of the
simulating processes can fail, only t simulated processes get stuck as well.
We’ll describe here a version of the BG simulation that appears in a follow-
up paper by Borowsky, Gafni, Lynch, and Rajsbaum [BGLR01]. This gives a
more rigorous presentation of the mechanisms of the original Borowsky-Gafni
paper, and includes a few simplifications.

27.1 High-level strategy

To avoid having to simulate specific choices of operations, the BG simulation
assumes that all simulated processes alternate between taking snapshots and

266
CHAPTER 27. BG SIMULATION 267

doing updates. This assumption is not very restrictive, because two snapshots
with no intervening update are equivalent to two snapshots separated by an
update that doesn’t change anything, and two updates with no intervening
snapshot can be replaced by just the second update, since the adversary
could choose to schedule them back-to-back anyway.
This approach means that we can determine the actions of some simulated
process by determining the sequence of snapshots that it receives. So the
goal will be to allow any of the real processes to take a snapshot on behalf
of any of the simulated processes, and then coordinate these snapshots via
weak consensus objects to enforce consistency if more than one real process
tries to simulate a step of the same simulated process. The key tool for doing
this is a safe agreement object, described in §27.2.

27.2 Safe agreement

A naive approach to simulate n processes using f + 1 processes would be
to lock each simulated process behind a mutex, and have the real processes
take turns grabbing a lock, simulating a step, and releasing the lock. If
we could somehow guarantee that processes never get stuck waiting for a
particular mutex just because some process died holding the lock, then we
could treat any blocked simulated process as dead, and charge its death to
the dead process holding the lock. This would give the mapping of at most
f simulated failures to f real failures we are hoping for. But this depends
on a lot of subtleties in how we implement the mutexes, so the standard BG
simulation goes through a weakening of consensus instead.
The safe agreement mechanism performs agreement without running
into the FLP bound, by providing a weaker termination condition. It is
guaranteed to terminate only if there are no failures by any process during
an initial, bounded, unsafe section of its execution, but if a process fails
later, it can prevent termination. Processes can detect when they leave the
unsafe section and have to wait for other processes only in the safe section.
This means that they can dovetail spinning in the safe sections of multiple
safe agreement objects without getting stuck entirely, even if dead processes
in the unsafe sections are blocking some of the objects.
Each process i starts the agreement protocol with a proposei (v) event
for its input value v. At some point during the execution of the protocol, the
process receives a notification safei , followed later (if the protocol finishes)
by a second notification agreei (v 0 ) for some output value v 0 . It is guaranteed
that the protocol terminates as long as all processes continue to take steps
CHAPTER 27. BG SIMULATION 268

until they receive the safe notification, and that the usual validity (all
outputs equal some input) and agreement (all outputs equal each other)
conditions hold. There is also a wait-free progress condition that the safei
notices do eventually arrive for any process that doesn’t fail, no matter what
the other processes do (so nobody gets stuck in their unsafe section).
Pseudocode for a safe agreement object is given in Algorithm 27.1. This
is a translation of the description of the algorithm in [BGLR01], which is
specified at a lower level using I/O automata.1

// proposei (v)
1 A[i] ← hv, 1i
2 if snapshot(A) contains hj, 2i for some j 6= i then
// Back off
3 A[i] ← hv, 0i
4 else
// Advance
5 A[i] ← hv, 2i
// safei
6 repeat
7 s ← snapshot(A)
8 until s does not contain hj, 1i for any j
// agreei
9 return s[j].value where j is smallest index with s[j].level = 2
Algorithm 27.1: Safe agreement (adapted from [BGLR01])

The communication mechanism is a snapshot object containing a pair

A[i] = hvaluei , leveli i for each process i, initially h⊥, 0i. When a process
carries out proposei (v), it sets A[i] to hv, 1i, advancing to level 1. It then
looks around to see if anybody else is at level 2; if so, it backs off to 0, and if
not, it advances to 2. In either case it then spins until it sees a snapshot with
nobody at level 1, and agrees on the level-2 value with the smallest index i.
The safei transition occurs when the process leaves level 1 (no matter
which way it goes). This satisfies the progress condition, since there is no
loop before this, and guarantees termination if all processes leave their unsafe
interval, because no process can then wait forever for the last 1 to disappear.
To show agreement, observe that at least one process advances to level 2
(because the only way a process doesn’t is if some other process has already
1
The I/O automaton model is described in Appendix G.
CHAPTER 27. BG SIMULATION 269

advanced to level 2), so any process i that terminates observes a snapshot s

that contains at least one level-2 tuple and no level-1 tuples. This means
that any process j whose value is not already at level 2 in s can at worst
reach level 1 after s is taken. But then j sees a level-2 tuples and backs
off. It follows that any other process i0 that takes a later snapshot s0 that
includes no level-1 tuples sees the same level-2 tuples as i, and computes the
same return value. (Validity also holds, for the usual trivial reasons.)

27.3 The basic simulation algorithm

The basic BG simulation uses a single snapshot object A with t+1 components
and an infinite array of safe agreement objects Sjr .
Each component A[i] of A belongs to one of the t + 1 simulating processes,
and is a vector of values A[i][j] that process i believes process j will have
written at some point during the simulated execution. These values are
tagged with round numbers: each A[i][j] holds a tuple hv, ri representing the
value v that process i determines process j would have written after taking
r snapshots.
The contents of these snapshots are obtained from the Sjr objects. The
inputs to Sjr are simulated snapshots, and the output sjr of Sjr represents
the value of the r-th snapshot performed by simulated process j.
Each simulating process i cycles through all simulated processes j. Simu-
lating one round of a particular process j involves four phases:

1. Make an initial guess for sjr by taking a snapshot of A and taking the
value with the largest round number for each component A[−][k].

2. Initiate the safe agreement protocol Sjr using this guess. It continues
to run Sjr until it leaves the unsafe interval.

3. Attempt to finish Sjr , by performing one iteration of the loop from

Algorithm 27.1. If this iteration doesn’t succeed, move on to simulating
j + 1 (but come back to this phase for j eventually).

4. If Sjr terminates, compute a new value vjr for j to write based on the
simulated snapshot returned by Sjr , and update A[i][j] with hvjr , ri.

Actually implementing this while maintaining an abstraction barrier

around safe agreement is tricky. One approach might be to have each process
i manage a separate thread for each simulated process j, and wrap the unsafe
part of the safe agreement protocol inside a mutex just for threads of i. This
CHAPTER 27. BG SIMULATION 270

guarantees that i enters the unsafe part of any safe agreement object on
behalf of only one simulated j at a time, while preventing delays in the safe
part of Sjr from blocking it from finishing some other Sj 0 r0 .

27.4 Effect of failures

So now what happens if a simulating process i fails? This won’t stop any
other process i0 from taking snapshots on behalf of j, or from generating its
own values to put in A[i0 ][j]. What it may do is prevent some safe agreement
object Sjr from terminating. The termination property of Sjr means that
this can only occur if the failure occurs while i is in the unsafe interval for
Sjr —but since i is only in the unsafe interval for at most one Sjr at a time,
this stalls only one simulated process j. It doesn’t block any i0 , because any
other i0 is guaranteed to leave its own unsafe interval for Sjr after finitely
many steps, and though it may waste some effort waiting for Sjr to finish,
once it is in the safe interval it doesn’t actually wait for it before moving on
to other simulated j 0 .
It follows that each failure of a simulating process knocks out at most
one simulated process. So a wait-free system with t + 1 processes—and thus
at most t failures in the executions we care about—will produces at most t
failures inside the simulation.

27.5 Inputs and outputs

Two details not specified in the description above are how i determines
j’s initial input and how i determines its own outputs from the outputs
of the simulated processes. For the basic BG simulation, this is pretty
straightforward: we use the safe agreement objects Sj0 to agree on j’s input,
after each i proposes its own input vector for all j based on its own input to
the simulator protocol. For outputs, i waits for at least n − t of the simulated
processes to finish, and computes its own output based on what it sees.
One issue that arises here is that we can only use the simulation to
solve colorless tasks, which are decision problems where any process can
return the output of any other process without causing trouble.2 This works
for consensus or k-set agreement, but fails pretty badly for renaming. The
extended BG simulation, due to Gafni [Gaf09], solves this problem by
2
The term “colorless” here comes from use of colors to represent process ids in the
topological approach described in Chapter 28. These colors aren’t really colors, but
topologists like coloring nodes better than assigning them ids.
CHAPTER 27. BG SIMULATION 271

mapping each simulating process p to a specific simulated process qp , and

using a more sophisticated simulation algorithm to guarantee that qp doesn’t
crash unless p does; details can be found in Gafni’s paper. There is also a
later paper by Imbs and Raynal [IR09] that simplifies some details of the
construction. Here, we will limit ourselves to the basic BG simulation.

27.6 Correctness of the simulation

To show that the simulation works, observe that we can extract a simulated
execution by applying the following rules:

1. The round-r write operation of j is represented by the first write tagged

with round r performed for j.

2. The round-r snapshot operation of j is represented by whichever snap-

shot operation wins Sjr .

The simulated execution then consists of a sequence of write and snapshot

operations, with order of the operations determined by the order of their
representatives in the simulating execution, and the return values of the
snapshots determined by the return values of their representatives.
Because all processes that simulate a write for j in round r use the same
snapshots to compute the state of j, they all write the same value. So the only
way we get into trouble is if the writes included in our simulated snapshots
are inconsistent with the ordering of the simulated operations defined above.
Here the fact that each simulated snapshot corresponds to a real snapshot
makes everything work: when a process performs a snapshot for Sjr , then it
includes all the simulated write operations that happen before this snapshot,
since the s-th write operation by k will be represented in the snapshot if and
only if the first instance of the s-th write operation by k occurs before it.
The only tricky bit is that process i’s snapshot for Sjr might include some
operations that can’t possibly be included in Sjr , like j’s round-r write or
some other operation that depends on it. But this can only occur if some
other process finished Sjr before process i takes its snapshot, in which case
i’s snapshot will not win Sjr and will be discarded.

27.7 BG simulation and consensus

BG simulation was originally developed to attack k-set agreement, but (as
pointed out by Gafni [Gaf09]) it gives a particularly simple proof of the
CHAPTER 27. BG SIMULATION 272

impossibility of consensus with one faulty process. Suppose that we had a

consensus protocol that solved consensus for n > 1 processes with one crash
failure, using only atomic registers. Then we could use BG simulation to get
a wait-free consensus protocol for two processes. But it’s easy to show that
atomic registers can’t solve wait-free consensus, because (following [LAA87]),
we only need to do the last step of FLP that gets a contradiction when
moving from a bivalent C to 0-valent Cx or 1-valent Cy. We thus avoid the
complications that arise in the original FLP proof from having to deal with
fairness.
More generally, BG simulation means that increasing the number of
processes while keeping the same number of crash failures doesn’t let us
compute anything we couldn’t before. This gives a formal justification for
the slogan that the difference between distributed computing and parallel
computing is that in a distributed system, more processes can only make
things worse.
Chapter 28

Topological methods

Here we’ll describe some results applying topology to distributed computing,

mostly following a classic paper of Herlihy and Shavit [HS99]. This was
one of several papers [BG93, SZ00] that independently proved lower bounds
on k-set agreement [Cha93], which is a relaxation of consensus where we
require only that there are at most k distinct output values (consensus is
1-set agreement). These lower bounds had failed to succumb to simpler
techniques.

28.1 Basic idea

The basic idea is to use tools from combinatorial topology to represent
indistinguishability proofs. We’ve seen a lot of indistinguishability proofs that
involving showing that particular pairs of executions are indistinguishable to
some process, which means that that process produces the same output in
both executions. In a typical proof of this kind, we then construct a chain of
executions Ξ1 , . . . , Ξk such that for each i, there is some p with Ξi |p = Ξi+1 |p.
We’ve generally been drawing these with the executions as points and the
indistinguishability relation as an edge between two executions. In the
topological method, we use the dual of this picture: each process’s view (the
restriction of some execution to events visible to that process) is represented
as a point, and an execution Ξ is represented as a simplex connecting all of
the points corresponding to views of Ξ by particular processes.
A simplex is a generalization to arbitrary dimension of the sequence
that starts with a point (a 0-simplex), an edge (a 1-simplex), a triangle
(a 2-simplex), or a tetrahedron (a 3-simplex). In general, an n-simplex is
a solid n-dimensional object with n + 1 vertices and n + 1 faces that are

273
CHAPTER 28. TOPOLOGICAL METHODS 274

(n − 1)-simplexes. As a combinatorial object, this is a fancy way of depicting

the power set of the set of vertices: each subset corresponds to a facet of the
original simplex. A simplicial complex consists of a bunch of simplexes pasted
together by identifying vertices: this is similar to the technique in graphics
of representing the surface of a three-dimensional object by decomposing
it into triangles. Topologists use these to model continuous surfaces, and
have many tools for deriving interesting properties of those surfaces from a
description of the simplicial complex.
For distributed computing, the idea is that some of these topological prop-
erties, when computed for the simplicial complex resulting from some protocol
or problem specification may sometimes useful to determine properties of
the underlying protocol or problem.

28.2 k-set agreement

The motivating problem for much of this work was getting impossibility
results for k-set agreement, proposed by Chaudhuri [Cha93]. The k-set
agreement problem is similar to consensus, where each process starts with
an input and eventually returns a decision value that must be equal to some
process’s input, but the agreement condition is relaxed to require only that
the set of decision values include at most k values.
With k − 1 crash failures, it’s easy to build a k-set agreement algorithm:
wait until you have seen n − k + 1 input values, then choose the smallest one
you see. This works because any value a process returns is necessarily among
the k smallest input values (including the k − 1 it didn’t see). Chaudhuri
conjectured that k-set agreement was not solvable with k failures, and gave
a proof of a partial result (analogous to the existence of an initial bivalent
configuration for consensus) based on Sperner’s Lemma [Spe28]. This is a
classic result in topology that says that certain colorings of the vertices of a
graph in the form of a triangle that has been divided into smaller triangles
necessarily contain a small triangle with three different colors on its corners.
This connection between k-set renaming and Sperner’s Lemma became the
basic idea behind each the three independent proofs of the conjecture that
appeared shortly thereafter [HS99, BG93, SZ00].
Our plan is to give a sufficient high-level description of the topological
approach that the connection between k-set agreement and Sperner’s Lemma
becomes obvious. It is possible to avoid this by approaching the problem
purely combinatorially, as is done, for example, in Section 16.3 of [AW04].
The presentation there is obtained by starting with a topological argument
CHAPTER 28. TOPOLOGICAL METHODS 275

and getting rid of the topology (in fact, the proof in [AW04] contains a proof
of Sperner’s Lemma with the serial numbers filed off). The disadvantage of
this approach is that it obscures what is really going in and makes it harder to
obtain insight into how topological techniques might help for other problems.
The advantage is that (unlike these notes) the resulting text includes actual
proofs instead of handwaving.

28.3 Representing distributed computations using

topology
Topology is the study of properties of shapes that are preserved by continuous
functions between their points that have continuous inverses, which get the
rather fancy name of homeomorphisms. A continuous function1 is one
that maps nearby points to nearby points. A homeomorphism is continuous
in both directions: this basically means that you can stretch and twist and
otherwise deform your object however you like, as long as you don’t tear
it (which would map nearby points on opposite sides of the tear to distant
points) or glue bits of it together (which turns into tearing when we look
at the inverse function). Topologists are particularly interested in showing
when there is no homeomorphism between two objects; the classic example
is that you can’t turn a sphere into a donut without damaging it, but you
can turn a donut into a coffee mug (with a handle).
Working with arbitrary objects embedded in umpteen-dimensional spaces
is messy, so topologists invented a finite way of describing certain well-behaved
objects combinatorially, by replacing ugly continuous objects like spheres
and coffee mugs with simpler objects pasted together in complex ways. The
simpler objects are simplexes, and the more complicated pasted-together
objects are called simplicial complexes. The nifty thing about simplicial
complexes is that they give a convenient tool for describing what states or
outputs of processes in a distributed algorithm are “compatible” in some
sense, and because topologists know a lot about simplicial complexes, we
can steal their tools to describe distributed algorithms.
1
Strictly speaking, this is the definition a continuous function between metric spaces,
which are spaces that have a consistent notion of distance. There is an even more general
definition of continuity that holds for spaces that are too strange for this.
CHAPTER 28. TOPOLOGICAL METHODS 276

28.3.1 Simplicial complexes and process states

The formal definition of a k-dimensional simplex is the convex closure of
(k + 1) points {x1 . . . xk+1 } in general position; the convex closure part means
the set of all points ai xi where ai = 1 and each ai ≥ 0, and the general
P P

position part means that the xi are not all contained in some subspace of
dimension (k −1) or smaller (so that the simplex isn’t squashed flat somehow).
What this gives us is a body with (k + 1) corners and (k + 1) faces, each of
which is a (k − 1)-dimensional simplex (the base case is that a 0-dimensional
simplex is a point). Each face includes all but one of the corners, and each
corner is on all but one of the faces. So we have:

• 0-dimensional simplex: point.2

• 1-dimensional simplex: line segment with 2 endpoints (which are both

corners and faces).

• 2-dimensional simplex: triangle (3 corners with 3 1-dimensional sim-

plexes for sides).

• 3-dimensional simplex: tetrahedron (4 corners, 4 triangular faces).

• 4-dimensional simplex: 5 corners, 5 tetrahedral faces. It’s probably

best not to try to visualize this.

A simplicial complex is a bunch of simplexes stuck together; formally,

this means that we pretend that some of the corners (and any faces that
include them) of different simplexes are identical points. There are ways to
do this right using equivalence relations. But it’s easier to abstract out the
actual geometry and go straight to a combinatorial structure.
An (abstract) simplicial complex is just a collection of sets with the
property that if A is a subset of B, and B is in the complex, then A is also
in the complex (this means that if some simplex is included, so are all of
its faces, their faces, etc.). This combinatorial version is nice for reasoning
about simplicial complexes, but is not so good for drawing pictures.
The trick to using this for distributed computing problems is that we
are going to build simplicial complexes by letting points be process states
(or sometimes process inputs or outputs), each labeled with a process id,
2
For consistency, it’s sometimes convenient to define a point as having a single (−1)-
dimensional face defined to be the empty set. We won’t need to bother with this, since
0-dimensional simplicial complexes correspond to 1-process distributed systems, which are
amply covered in almost every other Computer Science class you have ever taken.
CHAPTER 28. TOPOLOGICAL METHODS 277

and letting the sets that appear in the complex be those collections of
states/inputs/outputs that are compatible with each other in some sense.
For states, this means that they all appear in some global configuration
in some admissible execution of some system; for inputs and outputs, this
means that they are permitted combinations of inputs or outputs in the
specification of some problem.
Example: For 2-process binary consensus with processes 0 and 1, the
input complex, which describes all possible combinations of inputs, consists
of the sets

{{}, {p0}, {q0}, {p1}, {q1}, {p0, q0}, {p0, q1}, {p1, q0}, {p1, q1}} ,

which we might draw like this:

p0 q0

q1 p1

Note that there are no edges from p0 to p1 or q0 to q1: we can’t have

two different states of the same process in the same global configuration.
The output complex, which describes the permitted outputs, is

{{}, {p0}, {q0}, {p1}, {q1}, {p0, q0}, {p1, q1}} .

As a picture, this omits two of the edges (1-dimensional simplexes) from the
input complex:

p0 q0

q1 p1

One thing to notice about this output complex is that it is not connected:
there is no path from the p0–q0 component to the q1–p1 component.
Here is a simplicial complex describing the possible states of two processes
p and q, after each writes 1 to its own bit then reads the other process’s bit.
Each node in the picture is labeled by a sequence of process ids. The first id
in the sequence is the process whose view this node represents; any other
CHAPTER 28. TOPOLOGICAL METHODS 278

process ids are processes this first process sees (by seeing a 1 in the other
process’s register). So p is the view of process p running by itself, while pq
is the view of process p running in an execution where it reads q’s register
after q writes it.
p qp pq q

The edges express the constraint that if we both write before we read,
then if I don’t see your value you must see mine (which is why there is no
p–q edge), but all other combinations are possible. Note that this complex
is connected: there is a path between any two points.
Here’s a fancier version in which each process writes its input (and
remembers it), then reads the other process’s register (i.e., a one-round full-
information protocol). We now have final states that include the process’s
own id and input first, then the other process’s id and input if it is visible.
For example, p1 means p starts with 1 but sees a null and q0p1 means q starts
with 0 but sees p’s 1. The general rule is that two states are compatible if p
either sees nothing or q’s actual input and similarly for q, and that at least
one of p or q must see the other’s input. This gives the following simplicial
complex:
p0 q0p0 p0q0 q0

q1p0 p1q0

p0q1 q0p1

q1 p1q1 q1p1 p1

Again, the complex is connected.

The fact that this looks like four copies of the p–qp–pq–q complex pasted
into each edge of the input complex is not an accident: if we fix a pair of
inputs i and j, we get pi–qjpi–piqj–qj, and the corners are pasted together
because if p sees only p0 (say), it can’t tell if it’s in the p0/q0 execution or
the p0/q1 execution.
The same process occurs if we run a two-round protocol of this form,
where the input in the second round is the output from the first round. Each
round subdivides one edge from the previous round into three edges:
CHAPTER 28. TOPOLOGICAL METHODS 279

p−q

p − qp − pq − q

p − (qp)p − p(qp) − qp − (pq)(qp) − (qp)(pq) − pq − q(pq) − (pq)q − q

Here (pq)(qp) is the view of p after seeing pq in the first round and seeing
that q saw qp in the first round.

28.3.2 Subdivisions
In the simple write-then-read protocol above, we saw a single input edge turn
into 3 edges. Topologically, this is an example of a subdivision, where we
represent a simplex using several new simplexes pasted together that cover
exactly the same points.
Certain classes of protocols naturally yield subdivisions of the input
complex. The iterated immediate snapshot (IIS) model, defined by
Borowsky and Gafni [BG97], considers executions made up of a sequence
of rounds (the iterated part) where each round is made up of one or more
mini-rounds in which some subset of the processes all write out their current
views to their own registers and then take snapshots of all the registers (the
immediate snapshot part). The two-process protocols of the previous section
are special cases of this model.
Within each round, each process p obtains a view vp that contains the
previous-round views of some subset of the processes. We can represent the
views as a subset of the processes, which we will abbreviate in pictures by
putting the view owner first: pqr will be the view {p, q, r} as seen by p, while
qpr will be the same view as seen by q. The requirements on these views
are that (a) every process sees its own previous view: p ∈ vp for all p; (b)
all views are comparable: vp ⊆ vq or vq ⊆ vp ; and (c) if I see you, then I see
everything you see: q ∈ vp implies vq ⊆ vp . This last requirement is called
immediacy and follows from the assumption that writes and snapshots are
done in the same mini-round: if I see your write, then I see all the values
you do, because your snapshot is either in an earlier mini-round than mine
or in the same mini-round. Note this depends on the peculiar structure of
the mini-rounds, where all the writes precede all the snapshots.
CHAPTER 28. TOPOLOGICAL METHODS 280

The IIS model does not correspond exactly to a standard shared-memory

model (or even a standard shared-memory model augmented with cheap
snapshots). There are two reasons for this: standard snapshots don’t provide
immediacy, and standard snapshots allow processes to go back and perform
more than one snapshot on the same object. The first issue goes away if
we are looking at impossibility proofs, because the adversary can restrict
itself only to those executions that satisfy immediacy; alternatively, we can
get immediacy from the participating set protocol of [BG97], which we
will describe in §28.6.1. The second issue is more delicate, but Borowsky
and Gafni demonstrate that any decision protocol that runs in the standard
model can be simulated in the IIS model, using a variant of the BG simulation
algorithm described in Chapter 27.
For three processes, one round of immediate snapshots gives rise to the
simplicial complex depicted in Figure 28.1. The corners of the big triangle
are the solo views of processes that do their snapshots before anybody else
shows up. Along the edges of the big triangle are views corresponding to
2-process executions, while in the middle are complete views of processes that
run late enough to see everything. Each little triangle corresponds to some
execution. For example, the triangle with corners p, qp, rpq corresponds to
a sequential execution where p sees nobody, q sees p, and r sees both p and
q. The triangle with corners pqr, qpr, and rpq is the maximally-concurrent
execution where all three processes write before all doing their snapshots:
here everybody sees everybody. It is not terribly hard to enumerate all
possible executions and verify that the picture includes all of them. In higher
dimension, the picture is more complicated, but we still get a subdivision
that preserves the original topological structure [BG97].
Figure 28.2 shows (part of) the next step of this process: here we have
done two iterations of immediate snapshot, and filled in the second-round
subdivisions for the p–qpr–rpq and pqr–qpr–rpq triangles. (Please imagine
similar subdivisions of all the other triangles that I was too lazy to fill in
by hand.) The structure is recursive, with each first-level triangle mapping
to an image of the entire first-level complex. As in the two-process case,
adjacent triangles overlap because the relevant processes don’t have enough
information; for example, the points on the qpr–rpq edge correspond to views
of q or r that don’t include p in round 2 and so can’t tell whether p saw p or
pqr in round 1.
The important feature of the round-2 complex (and the round-k complex
in general) is that it’s a triangulation of the original outer triangle: a
partition into little triangles where each corner aligns with corners of other
little triangles.
CHAPTER 28. TOPOLOGICAL METHODS 281

rp qp

qpr rpq
pr pq

pqr

r q
qr rq

Figure 28.1: Subdivision corresponding to one round of immediate snapshot

CHAPTER 28. TOPOLOGICAL METHODS 282

rp qp

qpr rpq
pr pq

pqr

r q
qr rq

Figure 28.2: Subdivision corresponding to two rounds of immediate snapshot

CHAPTER 28. TOPOLOGICAL METHODS 283

(Better pictures of this process in action can be found in Figures 25 and

26 of [HS99].)

28.4 Impossibility of k-set agreement

Now let’s show that there is no way to do k-set agreement with n = k + 1
processes in the IIS model.
Suppose that after some fixed number of rounds, each process chooses an
output value. This output can only depend on the view of the process, so is
fixed for each vertex in the subdivision. Also, the validity condition means
that a process can only choose an output that it can see among the inputs in
its view. This means that at the corners of the outer triangle (corresponding
to views where the process thinks it’s alone), a process must return its input,
while along the outer edges (corresponding to views where two processes may
see each other but not the third), a process must return one of the two inputs
that appear in the corners incident to the edge. Internal corners correspond
to views that include—directly or indirectly—the inputs of all processes, so
these can be labeled arbitrarily. An example is given in Figure 28.3, for a
one-round protocol with three processes.
We now run into Sperner’s Lemma [Spe28], which says that, for any
subdivision of a simplex into smaller simplexes, if each corner of the original
simplex has a different color, and each corner that appears on some face of
the original simplex has a color equal to the color of one of the corners of
that face, then within the subdivision there are an odd number of simplexes
whose corners are all colored differently.3
3
The proof of Sperner’s Lemma is not hard, and is done by induction on the dimension
k. For k = 0, any subdivision consists of exactly one zero-dimensional simplex whose single
corner covers all k + 1 = 1 colors. For k + 1, suppose that the colors are {1, . . . , k + 1},
and construct a graph with a vertex for each little simplex in the subdivision and an
extra vertex for the region outside the big simplex. Put an edge in this graph between
each pair of regions that share a k-dimensional face with colors {1, . . . , k}. The induction
hypothesis tells us that there are an odd number of edges between the outer-region vertex
and simplexes on the {1, . . . , k}-colored face of the big simplex. The Handshaking Lemma
from graph theory says that the sum of the degrees of all the nodes in the graph is even.
But this can only happen if there are an even number of nodes with odd degree, implying
that the are are an odd number of simplexes in the subdivision with an odd number of
faces colored {1, . . . , k}, because the extra node for the outside region has exactly one
face colored {1, . . . , k}. Since zero is even, this means there is at least one simplex in the
subdivision with an odd number of faces colored {1, . . . , k}.
Now suppose we have a simplex with an odd number of faces colored {1, . . . , k}. Let f
be one such face. If the corner v not contained in f is colored c 6= k + 1, then our simplex
has exactly two faces colored {1, . . . , k}: f , and the face that replaces f ’s c-colored corner
CHAPTER 28. TOPOLOGICAL METHODS 284

2 1

1 1
1 3

2 3 3 3

Figure 28.3: An attempt at 2-set agreement

CHAPTER 28. TOPOLOGICAL METHODS 285

How this applies to k-set agreement: Suppose we have n = k + 1 processes

in a wait-free system (corresponding to allowing up to k failures). With
the cooperation of the adversary, we can restrict ourselves to executions
consisting of ` rounds of iterated immediate snapshot for some ` (termination
comes in here to show that ` is finite). This gives a subdivision of a simplex,
where each little simplex corresponds to some particular execution and each
corner some process’s view. Color all the corners of the little simplexes in
this subdivision with the output of the process holding the corresponding
view. Validity means that these colors satisfy the requirements of Sperner’s
Lemma. Sperner’s Lemma then says that some little simplex has all k + 1
colors, giving us a bad execution with more than k distinct output values.
The general result says that we can’t do k-set agreement with k failures
for any n > k. This follows immediately from the n = k + 1 version using
BG simulation (Chapter 27).

28.5 Simplicial maps and specifications

Let’s step back and look at consensus again.
One thing we could conclude from the fact that the output complex for
consensus was not connected but the ones describing our simple protocols
were was that we can’t solve consensus (non-trivially) using these protocols.
The reason is that to solve consensus using such a protocol, we would need
to have a mapping from states to outputs (this is just whatever rule tells
each process what to decide in each state) with the property that if some
collection of states are consistent, then the outputs they are mapped to are
consistent.
In simplicial complex terms, this means that the mapping from states
to outputs is a simplicial map, a function f from points in one simplicial
complex C to points in another simplicial complex D such that for any
simplex A ∈ C, f (A) = {f (x)|x ∈ A} gives a simplex in D. (Recall that
consistency is represented by including a simplex, in both the state complex
and the output complex.) A mapping from states to outputs that satisfies
the consistency requirements encoded in the output complex s always a
simplicial map, with the additional requirement that it preserves process ids
(we don’t want process p to decide the output for process q). Conversely,
any id-preserving simplicial map gives an output function that satisfies the
consistency requirements.
with v. So the only way to get an odd number of {1, . . . , k}-colored faces is to have all
k + 1 colors. It follows that there are an odd number of (k + 1)-colored simplexes.
CHAPTER 28. TOPOLOGICAL METHODS 286

Simplicial maps are examples of continuous functions, which have all

sorts of nice topological properties. One nice property is that a continuous
function can’t separate a path-connected space ( one in which there is a
path between any two points) into path-disconnected components. We can
prove this directly for simplicial maps: if there is a path of 1-simplexes
{x1 , x2 }, {x2 , x3 }, . . . {xk−1 , xk } from x1 to xk in C, and f : C → D is a
simplicial map, then there is a path of 1-simplexes {f (x1 ), f (x2 )}, . . . from
f (x1 ) to f (xk ). Since being path-connected just means that there is a path
between any two points, if C is connected we’ve just shown that f (C) is as
well.
Getting back to our consensus example, it doesn’t matter what simplicial
map f you pick to map process states to outputs; since the state complex C
is connected, so is f (C), so it lies entirely within one of the two connected
components of the output complex. This means in particular that everybody
always outputs 0 or 1: the protocol is trivial.

28.5.1 Mapping inputs to outputs

For general decision tasks, it’s not enough for the outputs to be consistent
with each other. They also have to be consistent with the inputs. This can
be expressed by a relation ∆ between input simplexes and output simplexes.
Formally, a decision task is modeled by a triple (I, O, ∆), where I is the
input complex, O is the output complex, and (A, B) ∈ ∆ if and only if B is
a permissible output given input I. Here there are no particular restrictions
on ∆ (for example, it doesn’t have to be a simplicial map or even a function),
but it probably doesn’t make sense to look at decision tasks unless there is
at least one permitted output simplex for each input simplex.

28.6 The asynchronous computability theorem

Given a decision task specified in this way, there is a topological characteriza-
tion of when it has a wait-free solution. This is given by the Asynchronous
Computability Theorem (Theorem 3.1 in [HS99]), which says:
Theorem 28.6.1. A decision task (I, O, ∆) has a wait-free protocol using
shared memory if and only if there exists a chromatic subdivision σ of I and
a color-preserving simplicial map µ : σ(I) → O such that for each simplex s
in σ(I), µ(S) ∈ ∆(carrier(S, I)).
To unpack this slightly, a chromatic subdivision is a subdivision where
each vertex is labeled by a process id (a color), and no simplex has two
CHAPTER 28. TOPOLOGICAL METHODS 287

vertices with the same color. A color-preserving simplicial map is a simplicial

map that preserves ids. The carrier of a simplex in a subdivision is whatever
original simplex it is part of. So the theorem says that I can only solve a
task if I can find a simplicial map from a subdivision of the input complex
to the output complex that doesn’t do anything strange to process ids and
that is consistent with ∆.
Looking just at the theorem, one might imagine that the proof consists
of showing that the protocol complex defined by the state complex after
running the protocol to completion is a subdivision of the input complex,
followed by the same argument we’ve seen already about mapping the state
complex to the output complex. This is almost right, but it’s complicated by
two inconvenient facts: (a) the state complex generally isn’t a subdivision of
the input complex, and (b) if we have a map from an arbitrary subdivision
of the input complex, it is not clear that there is a corresponding protocol
that produces this particular subdivision.
So instead the proof works like this:

Protocol implies map Even though we don’t get a subdivision with the
full protocol, there is a restricted set of executions that does give a
subdivision. So if the protocol works on this restricted set of execu-
tions, an appropriate map exists. There are two ways to prove this:
Herlihy and Shavit do so directly, by showing that this restricted set
of executions exists, and Borowksy and Gafni [BG97] do so indirectly,
by showing that the IIS model (which produces exactly the standard
chromatic subdivision used in the ACT proof) can simulate an ordinary
snapshot model. Both methods are a bit involved, so we will skip over
this part.

Map implies protocol This requires an algorithm. The idea here is that
that participating set algorithm, originally developed to solve k-set
agreement [BG93], produces precisely the standard chromatic subdivi-
sion used in the ACT proof. In particular, it can be used to solve the
problem of simplex agreement, the problem of getting the processes
to agree on a particular simplex contained within the subdivision of
their original common input simplex. This is a little easier to explain,
so we’ll do it.

28.6.1 The participating set protocol

Algorithm 28.1 depicts the participating set protocol; this first appeared
in [BG93], although the presentation here is heavily influenced by the version
CHAPTER 28. TOPOLOGICAL METHODS 288

in Elizabeth Borowsky’s dissertation [Bor95]. The shared data consists of

a snapshot object level, and processes start at a high level and float down
until they reach a level i such that there are already i processes at this level
or below. The set returned by a process consists of all processes it sees at
its own level or below, and it can be shown that this in fact implements a
one-shot immediate snapshot. Since immediate snapshots yield a standard
subdivision, this gives us what we want for converting a color-preserving
simplicial map to an actual protocol.

1 Initially, level[i] = n + 2 for all i.

2 repeat
3 level[i] ← level[i] − 1
4 v ← snapshot(level)
5 S ← {j | v[j] ≤ level[i]}
6 until |S| ≥ level[i]
7 return S
Algorithm 28.1: Participating set

The following theorem shows that the return values from participating
set have all the properties we want for iterated immediate snapshot:
Theorem 28.6.2. Let Si be the output of the participating set algorithm for
process i. Then all of the following conditions hold:
1. For all i, i ∈ Si . (Self-containment.)
2. For all i, j, Si ⊆ Sj or Sj ⊆ Si . (Atomic snapshot.)
3. For all i, j, if i ∈ Sj , then Si ⊆ Sj . (Immediacy.)
Proof. Self-inclusion is trivial, but we will have to do some work for the other
two properties.
The first step is to show that Algorithm 28.1 neatly sorts the processes
out into levels, where each process that returns at level ` returns precisely
the set of processes at level ` and below.
For each process i, let Si be the set of process ids that i returns, let `i
be the final value of level[i] when i returns, and let Si0 = {j | `j ≤ `i }. Our
goal is to show that Si0 = Si , justifying the above claim.
Because no process ever increases its level, if process i observes level[j] ≤ `i
in its last snapshot, then `j ≤ level[j] ≤ `i . So Si0 is a superset of Si . We
thus need to show only that no extra processes sneak in; in particular, we
will to show that |Si | = |Si0 |, by showing that both equal `i .
CHAPTER 28. TOPOLOGICAL METHODS 289

The first step is to show that |Si0 | ≥ |Si | ≥ `i . The first inequality follows
from the fact that Si0 ⊇ Si ; the second follows from the code (if not, i would
have stayed in the loop).
The second step is to show that |Si0 | ≤ `i . Suppose not; that is, suppose
that |Si0 | > `i . Then there are at least `i + 1 processes with level `i or less, all
of which take a snapshot on level `i + 1. Let i0 be the last of these processes
to take a snapshot while on level `i + 1. Then i0 sees at least `i + 1 processes
at level `i + 1 or less and exits, contradicting the assumption that it reaches
level `i . So |Si0 | ≤ `i .
The atomic snapshot property follows immediately from the fact that if
`i ≤ `j , then `k ≤ `i implies `k ≤ `j , giving Si = Si0 ⊆ Sj0 = Sj . Similarly, for
immediacy we have that if i ∈ Sj , then `i ≤ `j , giving Si ≤ Sj by the same
argument.

The missing piece for turning this into IIS is that in Algorithm 28.1, I
only learn the identities of the processes I am supposed to include but not
their input values. This is easily dealt with by the usual trick of adding an
extra register for each process, to which it writes its input before executing
participating set.

28.7 Proving impossibility results

To show something is impossible using the ACT, we need to show that
there is no color-preserving simplicial map from a subdivision of I to O
satisfying the conditions in ∆. This turns out to be equivalent to showing
that there is no continuous function from I to O with the same properties,
because any such simplicial map can be turned into a continuous function
(on the geometric version of I, which includes the intermediate points in
addition to the corners). Fortunately, topologists have many tools for proving
non-existence of continuous functions.

28.7.1 k-connectivity
Define the m-dimensional disk to be the set of all points at most 1 unit away
from the origin in Rm , and the m-dimensional sphere to be the surface of
the (m + 1)-dimensional disk (i.e., all points exactly 1 unit away from the
origin in Rm+1 ). Note that what we usually think of as a sphere (a solid
body), topologists call a disk, leaving the term sphere for just the outside
part.
CHAPTER 28. TOPOLOGICAL METHODS 290

An object is k-connected if any continuous image of an m-dimensional

sphere can be extended to a continuous image of an (m + 1)-dimensional
disk, for all m ≤ k.4 This is a roundabout way of saying that if we can draw
something that looks like a deformed sphere inside our object, we can always
include the inside as well: there are no holes that get in the way. The punch
line is that continuous functions preserve k-connectivity: if we want to map
an object with no holes continuously into some other object, the image had
better not have any holes either.
Ordinary path-connectivity is the special case when k = 0; here, the
0-sphere consists of two points and the 1-disk is the path between them. So
0-connectivity says that for any two points, there is a path between them.
For 1-connectivity, if we draw a loop (a path that returns to its origin),
we can include the interior of the loop somewhere. One way to thinking
about this is to say that we can shrink the loop to a point without leaving
the object (the technical term for this is that the path is null-homotopic,
where a homotopy is a way to transform one thing continuously into another
thing over time and the null path sits on a single point). An object that is
1-connected is also called simply connected.
For 2-connectivity, we can’t contract a sphere (or box, or the surface of a
2-simplex, or anything else that looks like a sphere) to a point.
The important thing about k-connectivity is that it is possible to prove
that any subdivision of a k-connected simplicial complex is also k-connected
(sort of obvious if you think about the pictures, but it can also be proved
formally), and that k-connectivity is preserved by simplicial maps (if not,
somewhere in the middle of all the k-simplexes representing our surface is a
(k + 1)-simplex in the domain that maps to a hole in the range, violating the
rule that simplicial maps map simplexes to simplexes). So a quick way to
show that the Asynchronous Computability Theorem implies that something
is not asynchronously computable is to show that the input complex is
k-connected and the output complex isn’t.

28.7.2 Impossibility proofs for specific problems

Here are some applications of the Asynchronous Computability Theorem
and k-connectivity:

Consensus There is no nontrivial wait-free consensus protocol for n ≥ 2

4
This definition is for the topological version of k-connectivity. It is not related in any
way to the definition of k-connectivity in graph theory, where a graph is k-connected if
there are k disjoint paths between any two points.
CHAPTER 28. TOPOLOGICAL METHODS 291

b1 c2 a1 b2 c1 a2 b1

c3 a4 b3 c4 a3 b4 c3

a1 b2 c1 a2 b1 c2 a1

Figure 28.4: Output complex for renaming with n = 3, m = 4. Each vertex

is labeled by a process id (a, b, c) and a name (1, 2, 3, 4). Observe that the
left and right edges of the complex have the same sequence of labels, as do
the top and bottom edges; the complex thus folds up into a (twisted) torus.
(This is a poor imitation of part of [HS99, Figure 9].)

processes. Proof: The input complex is 1-connected, but the output

complex is not, and we need a map that covers the entire output
complex (by nontriviality).

k-set agreement There is no wait-free k-set agreement for n ≥ k + 1

processes. Proof: The output complex for k-set agreement is not
k-connected, because buried inside it are lots of (k + 1)-dimensional
holes corresponding to missing simplexes where all k + 1 processes
choose different values. But these holes aren’t present in the input
complex—it’s OK if everybody starts with different inputs—and the
validity requirements for k-set agreement force us to map the surfaces
of these non-holes around holes in the output complex. (This proof
actually turns into the Sperner’s Lemma proof if we fully expand the
claim about having to map the input complex around the hole.)

Renaming There is no wait-free renaming protocol with less than 2n − 1

output names for all n. The general proof of this requires showing that
with fewer names we get holes that are too big (and ultimately reduces
to Sperner’s Lemma); for the special case of n = 3 and m = 4, see
Figure 28.4, which shows how the output complex of renaming folds
up into the surface of a torus. This means that renaming for n = 3
and m = 4 is exactly the same as trying to stretch a basketball into an
inner tube.
Chapter 29

Approximate agreement

Last updated 2011. Some material may be out of date.

The approximate agreement [DLP+ 86] or -agreement problem is

another relaxation of consensus where input and output values are real
numbers, and a protocol is required to satisfy modified validity and agreement
conditions.
Let xi be the input of process i and yi its output. Then a protocol
satisfies approximate agreement if it satisfies:

Termination Every nonfaulty process eventually decides.

Validity Every process returns an output within the range of inputs. For-
mally, for all i, it holds that (minj xj ) ≤ yi ≤ (maxj xj ).

-agreement For all i and j, |i − j| ≤ .

Unlike consensus, approximate agreement has wait-free algorithms for

asynchronous shared memory, which we’ll see in §29.1). But a curious
property of approximate agreement is that it has no bounded wait-free
algorithms, even for two processes (see §29.2)

29.1 Algorithms for approximate agreement

Not only is approximate agreement solvable, it’s actually easily solvable, to
the point that there are many known algorithms for solving it.

292
CHAPTER 29. APPROXIMATE AGREEMENT 293

We’ll use the algorithm of Moran [Mor95], mostly as presented in [AW04,

Algorithm 54] but with a slight bug fix;1 pseudocode appears in Algo-
rithm 29.1.2
The algorithm carries out a sequence of asynchronous rounds in which
processes adopt new values, such that the spread of the vector of all values
Vr in round r, defined as spread Vr = max Vr − min Vr , drops by a factor of
2 per round. This is done by having each process choose a new value in each
round by taking the midpoint (average of min and max) of all the values it
sees in the previous round. Slow processes will jump to the maximum round
they see rather than propagating old values up from ancient rounds; this is
enough to guarantee that latecomer values that arrive after some process
writes in round 2 are ignored.
The algorithm uses a single snapshot object A to communicate, and
each process stores its initial input and a round number along with its
current preference. We assume that the initial values in this object all have
round number 0, and that log2 0 = −∞ (which avoids a special case in the
termination test).

1 A[i] ← hxi , 1, xi i
2 repeat
3 hx01 , r1 , v1 i . . . hx0n , rn , vn i ← snapshot(A)
4 rmax ← maxj rj
5 v ← midpoint{vj | rj = rmax }
6 A[i] ← hxi , rmax + 1, vi
7 until rmax ≥ 2 and rmax ≥ log2 (spread({x0j })/)
8 return v
Algorithm 29.1: Approximate agreement

To show this works, we want to show that the midpoint operation

guarantees that the spread shrinks by a factor of 2 in each round. Let Vr
1
The original algorithm from [AW04] does not include the test rmax ≥ 2. This allows
for bad executions in which process 1 writes its input of 0 in round 1 and takes a snapshot
that includes only its own input, after which process 2 runs the algorithm to completion
with input 1. Here process 2 will see 0 and 1 in round 1, and will write (1/2, 2, 1) to
A[2]; on subsequent iterations, it will see only the value 1/2 in the maximum round, and
after dlog2 (1/)e rounds it will decide on 1/2. But if we now wake process 1 up, it will
decided 0 immediately based on its snapshot, which includes only its own input and gives
spread(x) = 0. Adding the extra test prevents this from happening, as new values that
arrive after somebody writes round 2 will be ignored.
2
Showing that this particular algorithm works takes a lot of effort. If I were to do this
over, I’d probably go with a different algorithm due to Schenk [Sch95].
CHAPTER 29. APPROXIMATE AGREEMENT 294

be the set of all values v that are ever written to the snapshot object with
round number r. Let Ur ⊆ Vr be the set of values that are ever written to
the snapshot object with round number r before some process writes a value
with round number r + 1 or greater; the intuition here is that Ur includes only
those values that might contribute to the computation of some round-(r + 1)
value.

Lemma 29.1.1. For all r for which Vr+1 is nonempty,

spread(Vr+1 ) ≤ spread(Ur )/2.

Proof. Let Uri be the set of round-r values observed by a process i in the
iteration in which it sees rmax = r in some iteration, if such an iteration
exists. Note that Uri ⊆ Ur , because if some value with round r + 1 or greater
is written before i’s snapshot, then i will compute a larger value for rmax .
Given two processes i and j, we can argue from the properties of snapshot
that either Uri ⊆ Urj or Urj ⊆ Uri . The reason is that if i’s snapshot comes
first, then j sees at least as many round-r values as i does, because the only
way for a round-r value to disappear is if it is replaced by a value in a later
round. But in this case, process j will compute a larger value for rmax and
will not get a view for round r. The same holds in reverse if j’s snapshot
comes first.
Observe that if Uri ⊆ Urj , then

midpoint(Uri ) − midpoint(Urj ) ≤ spread(Urj )/2.

This holds because midpoint(Uri ) lies within the interval min Urj , max Urj ,
 

and every point in this interval is within spread(Urj )/2 of midpoint(Urj ). The
same holds if Urj ⊆ Uri . So any two values written in round r + 1 are within
spread(Ur )/2 of each other.
In particular, the minimum and maximum values in Vr+1 are within
spread(Ur )/2 of each other, so spread(Vr+1 ) ≤ spread(Ur )/2.

Corollary 29.1.2. For all r ≥ 2 for which Vr is nonempty,

spread(Vr ) ≤ spread(U1 )/2r−1 .

Proof. By induction on r. For r = 2, this is just Lemma 29.1.1. For larger

r, use the fact that Ur−1 ⊆ Vr−1 and thus spread(Ur−1 ) ≤ spread(Vr−1 ) to
CHAPTER 29. APPROXIMATE AGREEMENT 295

compute

spread(Vr ) ≤ spread(Ur−1 )/2

≤ spread(Vr−1 )/2
≤ (spread(U1 )/2r−2 )/2
= spread(U1 )/2r−1 .

Let i be some process that finishes in the fewest number of rounds. Process
i can’t finish until it reaches round rmax +1, where rmax ≥ log2 (spread({x0j })/)
for a vector of input values x0 that it reads after some process writes round
2 or greater. We have spread({x0j }) ≥ spread(U1 ), because every value in
U1 is included in x0 . So rmax ≥ log2 (spread(U1 )/) and spread(Vrmax +1 ) ≤
spread(U1 )/2rmax ≤ spread(U1 )/(spread(U1 )/) = . Since any value re-
turned is either included in Vrmax +1 or some later Vr0 ⊆ Vrmax +1 , this gives
us that the spread of all the outputs is less than : Algorithm 29.1 solves
approximate agreement.
The cost of Algorithm 29.1 depends on the cost of the snapshot operations,
on , and on the initial input spread D. For linear-cost snapshots, this works
out to O(n log(D/)).

29.2 Lower bound on step complexity

The dependence on D/ is necessary, at least for deterministic algorithms.
Here we give a lower bound due to Herlihy [Her91a], which shows that any
deterministic approximate agreement algorithm takes at least log3 (D/) total
steps even with just two processes.
Define the preference of a process in some configuration as the value it
will choose if it runs alone starting from this configuration. The preference
of a process p is well-defined because the process is deterministic; it also can
only change as a result of a write operation by another process q (because
no other operations are visible to p, and p’s own operations can’t change
its preference). The validity condition means that in an initial state, each
process’s preference is equal to its input.
Consider an execution with two processes p and q, where p starts with
preference p0 and q starts with preference q0 . Run p until it is about to
perform a write that would change q’s preference. Now run q until it is about
to change p’s preference. If p’s write no longer changes q’s preference, start p
CHAPTER 29. APPROXIMATE AGREEMENT 296

again and repeat until both p and q have pending writes that will change the
other process’s preference. Let p1 and q1 be the new preferences that result
from these operations. The adversary can now choose between running P only
and getting to a configuration with preferences p0 and q1 , Q only and getting
p1 and q0 , or both and getting p1 and q1 ; each of these choices incurs at least
one step. By the triangle inequality, |p0 − q0 | ≤ |p0 − q1 |+|q1 − p1 |+|p1 − q0 |,
so at least on of these configurations has a spread between preferences that is
at least 1/3 of the initial spread. It follows that after k steps the best spread
we can get is D/3k , requiring k ≥ log3 (D/) steps to get -agreement.
Herlihy uses this result to show that there are decisions problems that have
wait-free but not bounded wait-free deterministic solutions using registers.
Curiously, the lower bound says nothing about the dependence on the number
of processes; it is conceivable that there is an approximate agreement protocol
with running time that depends only on D/ and not n.
 Part III

Other communication models

297
Chapter 30

Overview

In this part, we consider models that don’t fit well into the standard message-
passing or shared-memory models. These includes models where processes
can directly observe the states of nearby processes (Chapter 31); where
computation is inherently local and the emphasis is on computing information
about the communication graph (Chapter 32); where processes wander about
and exchange information only with processes they physically encounter
(Chapter 33); where processes (in the form of robots) communicate only by
observing each others’ locations and movements (Chapter 34); and where
processes can transmit only beeps, and are able to observe only whether at
least one nearby process beeped (Chapter 35).
Despite the varying communication mechanisms, these models all share
the usual features of distributed systems, where processes must contend with
nondeterminism and incomplete local information.

298
Chapter 31

Self-stabilization

A self-stabilizing algorithm has the property that, starting from any ar-
bitrary configuration, it eventually reaches a legal configuration, and this
property is stable in the sense that it remains in a legal configuration
thereafter. The notion of which configurations are legal depends on what
problem we are trying to solve, but the overall intuition is that an algorithm
is self-stabilizing if it can recover from arbitrarily horrible errors, and will
stay recovered as long as no new errors occur.
It’s generally not possible to detect whether the algorithm is in a legal
configuration from the inside: if a process has a bit that says that everything
is OK, the adversary can set that bit in the initial configuration, even if
everything is in fact broken. So self-stabilizing algorithms don’t actually
terminate: at best, they eventually converge to a configuration where the
necessary ongoing paranoid consistency checks produce no further changes
to the configuration (a property called silent self-stabilization.
The idea of self-stabilization first appeared in a paper by Edsger Dijk-
stra [Dij74], where he considered the problem of building robust token-ring
networks. In a token-ring network, there are n nodes arranged in a directed
cycle, and we want a single token to circulate through the nodes, as a mech-
anism for enforcing mutual exclusion: only the node currently possessing the
token can access the shared resource.
The problem is: how do you get the token started? Dijkstra worried both
about the possibility of starting with no tokens or with more than one token,
and he wanted an algorithm that would guarantee that, from any starting
state, eventually we would end up with exactly one token that would circulate
as desired. He called such an algorithm self-stabilizing, and gave three
examples, the simplest of which we will discuss in §31.2 below. These became

299
CHAPTER 31. SELF-STABILIZATION 300

the foundation for the huge field of self-stabilization, which spans thousands
of papers, at least one textbook [Dol00], a specialized conference (SSS, the
International Symposium on Stabilization, Safety, and Security in Distributed
Systems), and its own domain name http://www.selfstabilization.org/.
We won’t attempt to summarize all of this, but will highlight a few results
to give a sampling of what self-stabilizing algorithms look like.

31.1 Model
Much of the work in this area, dating back to Dijkstra’s original paper, does
not fit well in either the message-passing or shared-memory models that we
have been considering in this class, both of which were standardized much
later. Instead, Dijkstra assumed that processes could, in effect, directly
observe the states of their neighbors. A self-stabilizing program would
consist of a collection of what he later called guarded commands [Dij75],
statements of the form “if [some condition is true] then [update my state in
this way].” In any configuration of the system, one or more of these guarded
commands might have the if part (the guard) be true; these commands are
said to be enabled.
A step consists of one or more of these enabled commands being executed
simultaneously, as chosen by an adversary scheduler, called the distributed
daemon. The usual fairness condition applies: any process that has an
enabled command eventually gets to execute it. If no commands are enabled,
nothing happens. With the central daemon variant of the model, only one
step can happen at a time. With the synchronous daemon, every enabled
step happens at each time. Note that both the central and synchronous
daemons are special cases of the distributed daemon.
More recent work has tended to assume a distinction between the part
of a process’s state that is visible to its neighbors and the part that isn’t.
This usually takes the form of explicit communication registers or link
registers, which allow a process to write a specific message for a specific
neighbor. This is still not quite the same as standard message-passing or
shared-memory, because a process is often allowed to read and write multiple
link registers atomically.

31.2 Token ring circulation

For example, let us consider Dijkstra’s token ring circulation algorithm.
There are several versions of this in Dijkstra’s paper [Dij74]. We will do the
CHAPTER 31. SELF-STABILIZATION 301

unidirectional (n + 1)-state version, which is the simplest to describe and

analyze.
For this algorithm, the processes are numbered as elements 0 . . . n − 1 of
Zn , with all arithmetic on process ids being done modulo n.1 Each process i
can observe both its own state that that of its left neighbor at (i − 1) mod m.
Process 0 has a special role and has different code from the others, but
the rest of the processes are symmetric. Each process i has a variable `i that
takes on values in the range 0 . . . m − 1, interpreted as elements of Zm . The
algorithm is given in Algorithm 31.1.

1 Code for process 0:

2 if `0 = `n−1 then `00 ← (`n−1 + 1) mod (n + 1)
3 Code for process i 6= 0:
4 if `i 6= `i−1 then `0i ← `i−1
Algorithm 31.1: Dijkstra’s large-state token ring algorithm [Dij74]

In this algorithm, the nonzero processes just copy the state of the process
to their left. The zero process increments its state if it sees the same state
to its left. Note that the nonzero processes have guards on their commands
that might appear useless at first glance, but these are there ensure that the
adversary can’t waste steps getting nonzero processes to carry out operations
that have no effect.
What does this have to with tokens? The algorithm includes an additional
interpretation of the state, which says that:

1. If `0 = `n−1 , then 0 has a token, and

2. If `i 6= `i−1 , for i 6= 0, then i has a token.

This instantly guarantees that there is at least one token: if none of the
nonzero processes have a token, then all the `i variables are equal. But then
0 has a token. It remains though to show that we eventually converge to a
configuration where at most one process has a token.
Define a configuration ` as legal if there is some value j such that `i = `j
for all i ≤ j and `i = `j − 1 (mod n + 1) for all i > j. When j = n − 1,
this makes all `i equal, and 0 has the only token. When j < n − 1, then
`0 6= `n−1 (so 0 does not have a token), `j 6= `j+1 (so j + 1 has a token), and
`i = `i+1 for all i 6∈ j, n − 1 (so nobody else has a token). That each legal
1
In Dijkstra’s paper, there are n + 1 processes numbered 0 . . . n, but this doesn’t really
make any difference.
CHAPTER 31. SELF-STABILIZATION 302

configuration has exactly one token partially justifies our definition of legal
configurations.
If a configuration ` is legal, then when j = n − 1, the only enabled step
is `00 ← (`n−1 + 1) mod (n + 1); when j < n − 1, the only enabled step is
`0j+1 ← `j . In either case, we get a new legal configuration `0 . So the property
of being a legal configuration is stable, which is the other half of justifying
our definition.
Now we want to show that we eventually converge to a legal configuration.
Fix some initial configuration `0 , and let c be some value such that `0i 6= c
for all i. (There is at least one such c by the Pigeonhole Principle.) We will
argue that there is a sequence of configurations with c as a prefix of the
values that forms a bottleneck forcing us into a legal configuration:

Lemma 31.2.1. Let `0 , `1 , . . . be the sequence of configurations in some

execution of Dijkstra’s token ring circulation algorithm. Let 0 ≤ c ≤ n be
such that `0i 6= c for all i. Then for any configuration `t , either t is legal, or
there is some 0 ≤ j < n such that `ti = c if and only if i < j.

Proof. By induction on t. For the base case, `0 satisfies `0i = c if and only if
i < j when j = 0.
If `t is legal, `t+1 is also legal. So the interesting case is when `t is not
legal. In this case, there is some 0 ≤ j < n such that `ti = c if and only if
i < j.
If j = 0, then `ti 6= c for all i. Then the only way to get `t+1 i = c is if
i = 0. But then `t+1 satisfies the condition with j 0 = 1.
If 0 < j < n, then `ti = c for at least one i < j, and `tn−1 = 6 c since
n − 1 6< j. So we may get a transition that sets `t+1 j = `t
j−1 = c, giving a
new configuration ` t+1 0
that satisfies the induction hypothesis with j = j + 1,
or we may get a transition that does not create or remove any copies of c.
In either case the induction goes through.

To show that we eventually hit this bottleneck, we use a potential function.

Starting in some initial configuration `0 , let c be some missing value in `0 as
defined above. For any configuration `, define g(`) = (c − `0 ) mod (n + 1)
to be the gap between `0 and c. For each i ∈ {0, . . . , n − 2}, define ui (`) =
[`i 6= `i+1 ] to be the indicator variable for whether i is unhappy with its
successor, because its successor has not yet agreed to adopt its value.2 The
idea is that unhappiness moves right when some i 6= 0 copies its predecessor
2
The notation [P ], where P is some logical predicate, is called an Iverson bracket
and means the function that is 1 when P is true and 0 when P is false.
CHAPTER 31. SELF-STABILIZATION 303

and that the gap drops when 0 increments its value. By weighting these
values appropriately, we can arrange for a function that always drops.
Let
n−2
X
Φ(`) = ng(`) + (n − 1 − i)ui (`). (31.2.1)
i=0

Most of the work here is being done by the first two terms. The g term
tracks the gap between `0 and c, weighted by n. The sum tracks unhappiness,
weighted by distance to position n − 1.
In the initial configuration `0 , g is at most n, and each ui is at most 1,
so Φ(`0 ) = O(n2 ). We also have that Φ ≥ 0 always; if Φ = 0, then g = 0 and
ui = 0 for all i implies we are in an all-c configuration, which is legal. So we’d
like to argue that every step of the algorithm in a non-legal configuration
reachable from `0 reduces Φ by at least 1, forcing us into a legal configuration
after O(n2 ) steps.
Consider any step of the algorithm starting from a non-legal configuration
` with Φ(`t ) > 0 that satisfies the condition in Lemma 31.2.1:
t

6 0, then ui−1 changes from 1 to 0, reducing Φ

• If it is a step by i =
by (n − 1 − (i − 1)) = n − i. It may be that ui changes from 0 to 1,
increasing Φ by n − i − 1, but the sum of these changes is at most −1.

• If it is a step by 0, then u0 may increase from 0 to 1, increasing Φ by

n − 1. But g drops by 1 as long as `t0 6= c, reducing Φ by n, for a total
change of at most −1. The case `t = c is excluded by the assumption
that `t is non-legal and satisfies the conditions of Lemma 31.2.1, as the
only way for 0 to change its value away from c is if `tn−1 is also c.

Since the condition of Lemma 31.2.1 holds for any reachable `t , as long
as we are in a non-legal configuration, Φ drops by at least 1 per step. If we
do not reach a legal configuration otherwise, Φ can only drop O(n2 ) times
before hitting 0, giving us a legal configuration. Either way, the configuration
stabilizes in O(n2 ) steps.

31.3 Synchronizers
Self-stabilization has a curious relationship with failures: the arbitrary initial
state corresponds to an arbitrarily bad initial disruption of the system, but
once we get past this there are no further failures. So it is not surprising that
many of the things we can do in a failure-free distributed system we can also
CHAPTER 31. SELF-STABILIZATION 304

do in a self-stabilizing system. One of these is to implement a synchronizer,

which will allow us to pretend that our system is synchronous even if it isn’t.
The synchronizer we will describe here, due to Awerbuch et al. [AKM+ 93,
AKM+ 07], is a variant of the alpha synchronizer. It assumes that each
process can observe the states of its neighbors and that we have a central
daemon (meaning that one process takes a step at a time).
To implement this synchronizer in a self-stabilizing system, each process
v has a variable P (v), its current pulse. We also give each process a rule for
adjusting P (v) when it takes a step. Our goal is to arrange for every v to
increment its pulse infinitely often while staying at most one ahead of its
neighbors N (v). Awerbuch et al. give several possible rules for achieving
this goal, and consider the effectiveness of each.
The simplest rule is taken directly from the alpha synchronizer. When
activated, v sets
P (v) ← min (P (u) + 1)
u∈N (v)

This rule works find as long as every process starts synchronized. But
it’s not self-stabilizing. A counterexample, given in the paper, assumes we
have 10 processes organized in a ring. By carefully choosing which processes
are activated at each step, we can go through the following sequence of
configurations, where in each configuration the updated node is shown in
boldface:

1234312343
1234212343
1234232343
1234234343
1234234543
1234234542
3234234542
3434234542
3454234542

Here the final configuration is identical to the original if we increment each

value by one and shift the values to the left one position. So we can continue
this process indefinitely. But at the same time, each configuration has at
least one pair of adjacent nodes whose values differ by more than one.
The problem that arises in the counterexample is that sometimes values
can go backwards. A second rule proposed by Awerbuch et al. avoids this
CHAPTER 31. SELF-STABILIZATION 305

problem by preventing this, using the rule:

!
P (v) ← max P (v), min (P (u) + 1)
u∈N (v)

This turns out to be self-stabilizing, but the time to stabilize is unbounded

even in small networks. One counterexample is a network consisting of just
three nodes:

1 1 1050

If we run the nodes in round-robin order, the left two nodes will eventually
catch up to the rightmost, but it will take a while.
After some further tinkering, the authors present their optimal rule, which
they call max minus one:
(
minu∈N (v) (P (u) + 1) if P (v) looks legal,
P (v) ←
maxu∈N (v) (P (u) − 1) otherwise.

Here P (v) looks legal if it is within ±1 of all of its neighbors.

The intuition for why this works is that the most backward node pulls
the rest down to its level in O(D) time using the max-minus-one rule, after
which we just get the alpha synchronizer since everybody’s local values look
legal.
The actual proof uses a potential function at each node v given by

φ(v) = max(P (u) − P (v) − d(u, v)),

u

where d(u, v) is the distance between u and v in the graph. This is zero if
the skew between any pair of nodes is equal to the distance, which is the
most we can expect from a synchronizer. The proof shows that applying
the max-minus-one rule never increases φ(v), and decreases it by at least 1
whenever a node v with positive φ(v) changes P (v). Because this only gives
P
a bound of φ(v), the rest of the proof using a second potential function

Φ(v) = min {d(u, v) | P (u) − P (v) − d(u, v) = φ(v)} ,

u

which measures the distance from v to the nearest node u that supplies the
maximum in φ(v). It is shown that Φ(v) drops by 1 per time unit. When it
reaches 0, then φ(v) = P (v) − P (v) − d(v, v) = 0.
CHAPTER 31. SELF-STABILIZATION 306

We’ll skip over the details of the argument here. If you are interested,
the presentation in the paper is not too hard to follow.
The important part is that once we have a synchronizer, we can effectively
assume synchrony in other self-stabilizing algorithms. We just run the
synchronizer underneath our main protocol, and when the synchronizer
stabilizes, that gives us the initial starting point for the main protocol.
Because the main protocol itself should stabilize starting from an arbitrary
configuration, any insanity produced while waiting for the synchronizer to
converge is eventually overcome.

31.4 Spanning trees

The straightforward way to construct a spanning tree in a graph is to use
Bellman-Ford [Bel58, For56] to compute a breadth-first search tree rooted
at the node with lowest id. This has a natural implementation in the self-
stabilizing model: each process maintains root and dist, and when a process
takes a step, it sets root to the minimum of its own id and the minimum
root among its neighbors, and sets dist to 0 if it has the minimum id, or to
one plus the minimum distance to the root among its neighbors otherwise.
It is not hard to show that in the absence of errors, this converges to a
configuration where every node knows the id of the root and its distance to
the root in O(D) time, where D is the diameter of the network. A spanning
tree can then be extracted by the usual trick of having each node select as
parent some neighbor closer to the root.
But it fails badly if the system starts in an arbitrary state, because of
the ghost root problem. Suppose that some process wakes up believing
in the existence of a distant, fake root with lower id than any real process.
This fake root will rapidly propagate through the other nodes, with distances
increasing without bound over time. For most graphs, the algorithm will
never converge to a single spanning tree.
Awerbuch et al. [AKM+ 93] solve this problem by assuming a known
upper bound on the maximum diameter of the graph. Because the distance
to a ghost root will steadily increase over time, eventually only real roots
will remain, and the system will converge to a correct BFS tree.
It’s easiest to show this if we assume synchrony, or at least some sort
of asynchronous round structure. Define a round as the minimum time for
every node to update at least once. Then the minimum distance for any
ghost root rises by at least one per round, since any node with the minimum
distance either has no neighbor with the ghost root (in which case it picks
CHAPTER 31. SELF-STABILIZATION 307

a different root), or any neighbor that has the ghost root has at least the
same distance (in which case it increases its distance) Once the minimum
distance exceeds the upper bound D0 , all the ghost roots will have been
eliminated, and only real distances will remain. This gives a stabilization
time (in rounds) linear in the upper bound on the diameter.

31.5 Self-stabilization and local algorithms

In Chapter 32, we will look at algorithms in the LOCAL model, where
named processes in a synchronous network, organized as an unknown graph,
can send a polynomial-sized message to each neighbor in each round and
perform arbitrary computation locally. The goal is usually to compute some
property of the graph quickly, often in significantly fewer rounds than the
diameter of the graph.
There is a close connection between self-stabilizing algorithms and the
LOCAL model. The idea is that if we have a local algorithm that runs in
f (n) rounds, each process can propagate its information in a self-stabilizing
way to all nodes at distance at most f (n), and we can reconstruct the output
of the local algorithm whenever this information changes.
For each node u, let xu be its input value; we assume that this is fixed
once the system stabilizes. The state of u will be a table Tu , where Tu is
a partial function from sequences of nodes of length at most f (n) to input
values. We can represent this partial function as a set of ordered pairs
Tu = hw, xi, where we write Tu (w) = x if x is the unique value such that
hw, xi ∈ Tu , or Tu (w) = ⊥ if there is no such value. We have one rule at each
node u, which we can imagine is guarded so that it fires only if it changes Tu :
[
Tu ← {hu, xu i} ∪ {huw, xi | |uw| ≤ f (n), Tv (w) = x} (31.5.1)
v∈δ(u)

We can now argue that, after stabilization, this process eventually con-
verges to Tu consisting precisely of the set of all pairs hw, xv i where w is a
u–v path of length at most f (n) and xv is the input to v. Indeed, this works
under almost any reasonable assumption about scheduling. The relevant
lemma:

Lemma 31.5.1. Starting from any initial configuration, for any sequence
w of at most f (n) vertices starting at u and ending at v, (31.5.1) fires for
each node in w in reverse order, then Tu (w) = xv if w is a u–v path, and
Tu (w) = ⊥ otherwise.
CHAPTER 31. SELF-STABILIZATION 308

Proof. The proof is by induction on the length of w. The base case is when
|w| = 1, implying w = u = v. Here rule (31.5.1) writes hu, xu i to Tu , giving
Tu (u) = xu as claimed.
For a sequence w = uw0 where w0 is a nonempty path from some node u0
to v, if u0 is a neighbor of u, then firing rule (31.5.1) at u after firing the rule
for each node in w0 has Tu (uw0 ) ← Tu0 (w) = xv by the induction hypothesis.
If uw0 is not a path from u to v, then either u0 is not a neighbor of u, or w0
is not a path from u0 to v and Tu0 (w0 ) = ⊥ by the induction hypothesis. In
either case, Tu (uw0 ) ← ⊥.

What does this buy us? Suppose we have a deterministic synchronous

algorithm that runs in f (n) rounds. Starting from a stable configuration,
Lemma 31.5.1 tells us that any fair daemon will eventually leave us in a
configuration where each node u stores in Tu both the inputs of all nodes
within distance f (n) and enough information to reconstruct how they are
connected. So u can simulate the execution of any node at distance d for
up to f (n) − d rounds. In particular, it can simulate its own execution for
f (n) rounds, computing the same output as it would produce in the LOCAL
model.
Chapter 32

Distributed graph algorithms

In Chapter 31, we saw that certain classes of “local” algorithms have a

straightforward conversion to self-stabilizing algorithms. In this chapter,
we’ll look more closely at what kinds of problems can be solved with this
kind of locality.

32.1 The LOCAL and CONGEST models

The LOCAL and CONGEST models were defined by Peleg [Pel00] to for-
malize the idea of local distributed computation. Similar models had been
considered previously without being specifically named [Lin92], but these
names are now standard.
The LOCAL model is a synchronous message-passing model where the
processes are organized into a graph, all run the same code, and can com-
municate only with their neighbors in the graph. To break symmetry, each
process starts with a unique id that is polynomial in the number of processes
n. The processes may also start with local inputs, but often we are interested
simply in computing some property of the graph itself. There is no bound
on the size of messages.
The CONGEST model is like the local model, but messages are lim-
ited to O(log n) bits. More generally, the CONGEST(b) model allows
messages of size b, making LOCAL = CONGEST(∞) and CONGEST =
CONGEST(O(log n)).
In both models, we usually assume that the processes do not know the
structure of the graph or their place in it. But for specific problems, we
might require the graph to be from some restricted class (e.g., rings, trees,
cliques).

309
CHAPTER 32. DISTRIBUTED GRAPH ALGORITHMS 310

32.2 Local graph coloring

One of the first problems studied in local distributed graph algorithms is
graph coloring [Lin92], where we wish to assign each node in the graph a
small label distinct from its neighbors. Because the nodes initially start
with large distinct labels, graph coloring in the LOCAL model shares some
similarities with renaming, since we will use the unique ids as a starting
point for generating the colors.

32.2.1 Coloring graphs with out-degree 1

Let us start by describing a classic local algorithm for 3-coloring a directed
graph with maximum out-degree 1, a class of graphs that includes both cycles
and rooted trees. The algorithm we will use is ultimately due to Cole and
Vishkin [CV86], although the application to local graph coloring was given
by Linial [Lin92], and the version given here incorporates some additional
features from Peleg’s textbook [Pel00].
The idea from the Cole and Vishkin algorithm is to treat each identity
x as a long bit-string xk xk−1 . . . x0 , where k = blg N c and x = 2i xi and
P

repeatedly apply an operation that to shorten these ids while maintaining

the property that neighbors have distinct ids.
At each synchronous round, each process adopts a new identity based
on its old identity x and the identity x0 of its successor. We look for the
smallest index i for which xi 6= x0i . We then generate a new identity 2i + xi ;
this is the same as taking the bit-vector representation of i and tacking xi
on to the end of it.
In the case of a node with no successor, we pretend that it has a successor
with x00 6= x0 . This effectively knocks x down to its last bit x0 .
We now argue that this never produces two adjacent identities that are
the same. Let i0 be the smallest index for which my successor’s x0i0 differs
from their successor’s x00i0 . If i0 =
6 i, then my successor’s new identity 2i0 + xi0
will not equal my new identity 2i + xi , because the initial bits will be different.
But if i0 = i, then my successor’s new identity is 2i + x0i = 6 2i + xi because
x0i 6= xi .
Assuming that the largest initial color is N , the largest possible value
for i is blg N c, and so the largest possible value for 2i + 1 is 2blg N c + 1.
This converges to at most 5 after O(log∗ N ) rounds, which gives us 6 colors:
0, . . . , 5, where no two adjacent processes have the same color.
To reduce this to three colors, add a phase for each c ∈ {3, 4, 5} to
eliminate c. In each phase, we carry out a two-stage process: first, we shift
CHAPTER 32. DISTRIBUTED GRAPH ALGORITHMS 311

all colors down, by having each node switch its color to that of its successor
(or some new color chosen from {0, 1, 2} if it doesn’t have a successor). The
reason for doing this is that it guarantees that each node’s predecessors
will all share the same color, meaning that that node now has at most two
colors represented among its predecessors and successor. At the same time,
it doesn’t create any new pair of adjacent nodes with the same color. So
for the second stage, each node v that currently has color c can choose a
new color from {0, 1, 2} that is the smallest color that doesn’t appear in its
neighborhood. Since none of v’s neighbors change color (they don’t have
color c), this replaces all instances of c with a color form {0, 1, 2} while
keeping all edges two-colored. After doing this for all c ∈ {3, 4, 5}, the only
colors left are in {0, 1, 2}.
Doing the 6 to 3 reduction in the obvious way takes an additional 6
rounds, which is dominated by the O(log∗ n) of reducing from N initial ids.

32.2.2 Lower bound for rings

Using a Ramsey-theoretic argument, Linial [Lin92] showed that Ω(log∗ n)
is a lower bound on the time to color a directed ring with n nodes in the
LOCAL model, which implies that the algorithm of the previous section is
optimal up to constants, since a directed ring is a special case of a graph with
out-degree 1. We’ll describe here a simplified version of Linial’s original proof
given by Laurinharju and Suomela [LS14]. (The Laurinharju and Suomela
paper is only two pages long, so it may be worth skipping the rest of this
section and just reading it in the original.)
The idea is that any coloring algorithm in the local model that runs in
time T assigns a color to each node based only on the initial ids of the 2T + 1
within T hops of the node. So we can represent any possible deterministic
coloring algorithm by specifying the mapping from these 2T + 1 ids to colors.
Define a k-ary c-coloring function as a function A : [n]k → [c] where
[n] = {1, dots, n} is the id space and [c] = {1, dots, c} is a set of c colors,
with the property that

A(x1 , x2 , . . . , xk ) 6= A(x2 , . . . , xk , xk+1 ) (32.2.1)

for any increasing sequence 1 ≤ x1 < x2 < · · · < xk+1 ≤ n.

The restriction to increasing sequences and values in [n] rather than
[N ] is more restrictive that a general c-coloring algorithm, but if we have
a successful 3-coloring algorithm that runs in time T , we can extract from
it a (2T + 1)-ary 3-coloring function, and condition (32.2.1) will hold given
that the original algorithm never assigns the same color to adjacent nodes.
CHAPTER 32. DISTRIBUTED GRAPH ALGORITHMS 312

Conversely, if condition (32.2.1) fails for some sequence (x1 , x2 , . . . xk+1 ),

then we can supply this sequence as the ids for the first k + 1 nodes in the
ring and show the algorithm fails. This implies that a 3-coloring algorithm
that runs in time T can exist only if there is a (2T +1)-ary 3-coloring function.
The lower bound proof works by showing that T needs to be Ω(log∗ n) for
this to be possible.
It holds trivially that any 1-ary c-coloring function requires c ≥ n. The
proof works by showing how to transform any k-ary c-coloring function into
a (k − 1)-ary 2c -coloring function, which hits the trivial bound after k − 1
steps.
Lemma 32.2.1 ([LS14, Lemma 2]). For k > 1, given a k-ary c-coloring
function A, it is possible to construct a (k − 1)-ary 2c -coloring function B.
Proof. Let B 0 (x1 , . . . , xk−1 ) = {A(x1 , x2 , . . . , xk−1 , xk ) | xk > xk−1 }. In other
words, we fill in the missing parameter xk with all possible values xk > xk−1 ,
and return the set of colors that we obtain from A. Since there are exactly
2c possible sets, we can obtain B : [N ]k−1 → [2c ] by encoding each set as a
distinct number in [2c ] = {1, . . . , 2c }.
We will now prove that B satisfies (32.2.1) whenever A does, by showing
the contraposition that if B does not satisfy (32.2.1), then A doesn’t either.
Suppose now that (32.2.1) does not hold for B, that is, there is some
increasing sequence (x1 , . . . , xk ) such that B(x1 , . . . , xk−1 = B(x2 , . . . , xk ),
or equivalently B 0 (x1 , . . . , xk−1 = B 0 (x2 , . . . , xk ).
We will feed this bad sequence to A and see what happens. Let α =
A(x1 , . . . , xk ). Since xk is one of the possible extensions of (x1 , . . . , xk−1 )
used to generate B 0 (x1 , . . . , xk−1 ), we get α ∈ B 0 (x1 , . . . , xk−1 ). But then α
is also contained in B 0 (x2 . . . , xk ) = B 0 (x1 , . . . , xk−1 ). From the definition
of B 0 (x2 , . . . , xk ), this implies that there is some xk+1 > xk such that α =
A(x2 , . . . , xk , xk+1 ) = A(x1 , x2 , . . . , xk ). But then A is not a k-ary c-coloring
function.
To get the Ω(log∗ n) lower bound, start with a k-ary 3-coloring function
and iterate Lemma 32.2.1 to get a 1-ary f (k − 1)-coloring function where f (k)
is the result of applying the function 2x to 3 k − 1 times. Then f (k − 1) ≥ n,
which implies k = Ω(log∗ n).

32.2.3 Coloring bounded-degree graphs

The O(log∗ n)-time 3-coloring algorithm for out-degree 1 digraphs can be
used to get a simple O(∆2 + log∗ n) time for (∆ + 1)-coloring any graph with
maximum degree ∆, using an algorithm of Panconesi and Rizzi [PR01].
CHAPTER 32. DISTRIBUTED GRAPH ALGORITHMS 313

This algorithm has three steps:

1. First, partition the original graph G into ∆ directed graphs G1 , . . . , G∆ ,

each with maximum out-degree 1. We can do this in O(1) rounds:
each process collects the ids of its neighbors, and assigns each a port
number in {1, . . . , ∆} in increasing order of id, while also orienting
each edge to point to the neighbor with larger id. Each directed graph
Gi then consists of all edges for which the source node assigns port
number 1.

2. Next, use Cole-Vishkin (§32.2.1) to 3-color each Gi .

3. To color the original graph G, start with G01 = G1 and repeatedly

merge each G0i with the next unmerged Gi+1 to get G0i+1 . The merging
process consists of assigning each node a color in [3(∆ + 1)] by taking
an ordered pair of its color in Gi+1 and its color in G0i . Then for each
c ∈ {∆ + 2, . . . , 3(∆ + 1)}, have each node with color c choose the
smallest color not represented among its neighbors. (This is the same
color-reduction scheme used to go from six to three colors in §32.2.1,
except without the shifting.) This costs O(∆) rounds per merge, with
O(∆) total merges, giving ∆ + 1 colors after O(∆2 ) rounds.

This algorithm has the advantage of simplicity, but there are faster
algorithms. An algorithm of Ghaffari and Kuhn [GK20] obtains a (∆ + 1)-
coloring of a graph with maximum degree ∆ in O(log2 ∆ log n) rounds.
Chapter 33

Population protocols

Here are four mostly-equivalent models:

Population protocols A population protocol [AAD+ 06] consists of a

collection of agents with states in some state space Q. At each step, the
adversary picks two of the agents to interact, and both get to update
their state according to a joint transition function δ : Q×Q → Q×Q. A
global fairness condition requires that if some global configuration C
of the system occurs infinitely often, and there is a step that transforms
C to C 0 , then this step eventually occurs.
Computation by population protocols usually consists of computing
some function of the initial states of the population and propagating
the output of this function to all agents. As in a self-stabilizing system,
termination is not detected; instead, we hope to converge to the correct
answer eventually.
In some versions of the model, interactions between agents are limited by
an interaction graph (only adjacent agents can interact), or are assumed
to be chosen randomly instead of adversarially. These assumptions
may in some cases increase the power of the model.

Chemical reaction networks In a (CRN for short), we have a collection

of molecules representing different species. These molecules may
undergo chemical reactions that transform one or more inputs into one
or more outputs, as in this bit of classic rocketry:

H2 + O2 → H2 O + O

Computation by a chemical reaction network consists of putting some

314
CHAPTER 33. POPULATION PROTOCOLS 315

appropriate mix of molecules into a test tube, stirring it up, and hoping
to learn something from the final product.
Unlike population protocols, chemical reaction networks do not neces-
sarily conserve the number of molecules in each interaction, and (in
reality at least) require some source of energy to keep the reactions
going.

Petri nets A Petri net [Pet62] is a collection of of places and transitions,

in the form of a bipartite graph, with tokens wandering around through
the places. A transition fires by consuming one token from each place
in its in-neighborhood and adding one token to each place in its out-
neighborhood, assuming there is at least one token on each place in its
in-neighborhood. Various conditions are assumed on which transitions
fire in which order.
Petri nets were invented to model chemical reaction networks, so it’s
not surprising that they do so. Pretty much any result in population
protocols or CRNs can be translated to Petri nets or vice versa, by the
agent molecule token
mapping: state species place
transition reaction transition
Following a long-standing and probably unjustified American prejudice,
we will not talk much about Petri nets, but there has been a fair bit of
cross-pollenization between the population protocol, CRN, and Petri
net literature.

Vector addition systems You have a non-negative integer vector x. There

is a set of rules −a + b, where a and b are both non-negative integer
vectors, and you are allowed to replace x by x − a + b if x − a ≥ 0.
These are basically Petri nets without jargon.
Of these models, population protocols are currently the most popular
in the theory of distributed computing community, with chemical reaction
networks moving up fast. So we’ll talk about population protocols.

33.1 Definition of a population protocol

Let us begin by giving a formal definition of a population protocol, following
the original definition of Angluin et al. [AAD+ 06].
A population protocol is a tuple hX, Y, Q, I, O, δi, where X is the
input alphabet, Y is the output alphabet, Q is the state space, I : X → Q
CHAPTER 33. POPULATION PROTOCOLS 316

maps inputs to initial states, O : Q → Y maps states to outputs, and

δ : Q × Q → Q × Q is the transition function.
A population consists of n agents, taken to be the vertices of a directed
graph called the interaction graph. Most of the time we will assume the
interaction graph is a complete graph, but the model allows for more re-
strictive assumptions. A configuration is a mapping C from agents to
Q. A transition involves choosing two agents x and y such that xy is an
edge in the interaction D graph,E and updating the configuration C to a new
configuration C with Cx0 , Cy0 = δ(hCx , Cy i) and Cz0 = Cz for all z 6∈ {x, y}.
0

The first agent in an interaction is called the initiator and the second
agent the responder. Note that this means that the model breaks symmetry
for us.
With a complete interaction graph, we can will often not bother with the
identities of specific agents and just treat the configuration C as a multiset
of states.
The main difference between population protocols and similar models
is the input and output mappings, and the notion of stable computation,
which gets its own section.

33.2 Stably computable predicates

A predicate P on a vector of initial inputs is stably computable if there
exists a population protocol such that it eventually holds forever that every
agent correctly outputs whether P is true or false. Stably computable
functions are defined similarly.
One of the big early results on population protocols was an exact char-
acterization of stably computable predicates for the complete interaction
graph. We will give a sketch of this result below, after giving some examples
of protocols that compute particular predicates.

33.2.1 Time complexity

The original population protocol did not define a notion of time, since
the fairness condition allows arbitrarily many junk transitions before the
system makes progress. More recent work has tended to compute time
complexity by assuming random scheduling, where the pair of agents to
interact is determined by choosing an edge uniformly from the interaction
graph (which means uniformly from all possible pairs when the interaction
graph is complete).
CHAPTER 33. POPULATION PROTOCOLS 317

Assuming random scheduling (and allowing for a small probability of

error) greatly increases the power of population protocols. So when using this
time measure we have to be careful to mention whether we are also assuming
random scheduling to improve our capabilities. Most of the protocols in
this section are designed to work as long as the scheduling satisfies global
fairness—they don’t exploit random scheduling—but we will discuss running
time in the random-scheduling case as well.

33.2.2 Examples
These examples are mostly taken from the original paper of Angluin et
al. [AAD+ 06].

33.2.2.1 Leader election

Most stably computable predicates can be computed as a side-effect of
leader election, so we’ll start with a leader election protocol. The state
space consists of L (leader) and F (follower); the input map makes every
process a leader initially. Omitting transitions that have no effect, the
transition relation is given by

L, L → L, F.

It is easy to see that in any configuration with more than one leader,
there exists a transition that eventually reduces the number of leaders. So
global fairness says this happens eventually, which causes us to converge to
a single leader after some finite number of interactions.
If we assume random scheduling, the expected number of transitions to
get down to one leader is exactly
n n
X n(n − 1) X 1
= n(n − 1)
k=2
k(k − 1) k=2
k(k − 1)
n 
1 1
X 
= n(n − 1) −
k=2
k−1 k
1
 
= n(n − 1) 1 −
n
2
=n .
CHAPTER 33. POPULATION PROTOCOLS 318

33.2.2.2 Distributing the output

The usual convention in a population protocol is that we want every process
to report the output. It turns out that this is equivalent to the leader
reporting the output.
Given a protocol A with states of the form h`, xi where ` ∈ {L, F } is the
leader bit and x is whatever the protocol is computing, define a new protocol
A0 with states h`, x, yi where y = O(x) when ` = L and y is the output of
the last leader the agent met when ` = F .
Now as soon as the leader has converged on an output, it only needs to
meet each other agent once to spread it to them. This takes an additional
nHn−1 /2 = O(n2 log n) interactions on average.

33.2.2.3 Remainder mod m

We can now give an example of a protocol that stably computes a function:
we will count the number of agents in some special initial state A, modulo
a constant m. (We can’t count the exact total because the agents are
finite-state.)
Each agent has a state h`, xi, where ` ∈ {L, F } as in the leader election
protocol, and x ∈ Zm . The input mapping sends A to hL, 1i and everything
else to hL, 0i. The non-trivial transitions are given by

hL, xi , hL, yi → hL, (x + y) mod mi , hF, 0i

This protocol satisfies the invariant that the sum over all agents of
the second component, mod m, is unchanged by any transition. Since the
components for any is follower is zero, this means that when we converge to
a unique leader, it will contain the count of initial A’s mod m.

33.2.2.4 Linear threshold functions

Remainder mod m was one of two tools in [AAD+ 06] that form the foundation
for computing all stably computable predicates. The second computes linear
threshold predicates, of the form
X
ai xi ≥ b, (33.2.1)

where the xi are the counts of various possible inputs and the ai and b are
integer constants. This includes comparisons like x1 > x2 as a special case.
The idea is to compute a truncated version of the left-hand side of (33.2.1)
as a side-effect of leader election.
CHAPTER 33. POPULATION PROTOCOLS 319

Fix some k > max(|b|, maxi |ai |). In addition to the leader bit, each agent
stores an integer in the range −k through k. The input map sends each xi to
the corresponding coefficient ai , and the transition rules cancel out positive
and negative ai , and push any remaining weight to the leader as much as
possible subject to the limitation that values lie within [−k, k].
Formally, define a truncation function t(x) = max(−k, min(k, r)), and a
remainder function r(x) = x − t(x). These have the property that if |x| ≤ 2k,
then t(x) and r(x) both have their absolute value bounded by k. If we have
the stronger condition |x| ≤ k, then t(x) = x and r(x) = 0.
We can now define the transition rules:

hL, xi , h−, yi → hL, t(x + y)i , hF, r(x + y)i

hF, xi , hF, yi → hF, t(x + y)i , hF, r(x + y)i

These have the property that the sum of the second components is
preserved by all transitions. Formally, if we write yi for the second component
P
of agent i, then yi does not change through the execution of the protocol.
When agents with positive and negative values meet, we get cancellation.
This reduces the quantity |yi |. Global fairness implies that this quantity
P

will continue to drop until eventually all nonzero yi have the same sign.
Once this occurs, and there is a unique leader, then the leader will
eventually absorb as much of the total as it can. This will leave the leader
P
with y = min(k, max(−k, yi )). By comparing this quantity with b, the
leader can compute the threshold predicate.

33.2.3 Presburger arithmetic and semilinear sets

Presburger arithmetic [Pre29] is the first-order theory (in the logic sense)
of the natural numbers with addition, equality, 0, and 1. This allows
expressing ideas like “x is even:”

∃y : x = y + y

or “x is greater than y”:

∃z : x = y + z + 1

but not “x is prime” or even x = y · z.”

Presburger arithmetic has various amazing properties, including decid-
ability–there is an algorithm that will tell you if any statement in Presburger
arithmetic is true or not (in doubly-exponential time [FR98])—and quanti-
fier elimination—a formula using any combination of ∀ and ∃ quantifiers
CHAPTER 33. POPULATION PROTOCOLS 320

can be converted to a formula with no quantifiers, using the predicates <

and ≡k for constant values of k.
There is also a one-to-one correspondence between predicates in Pres-
burger arithmetic and semilinear sets, which are finite unions of linear
sets of the form {b + ai xi } where b is a non-negative integer vector, the
P

ai are non-negative integer coefficients, and the xi are non-negative integer

vectors, and there are only finitely many terms.
(We will not attempt to prove any of this.)
It turns out that Presburger arithmetic (alternatively, semilinear sets)
captures exactly what can and can’t be stably computed by a population
protocol. For example, no semilinear set contains all and only primes (because
any infinite semilinear set on one variable is an arithmetic progression), and
primes aren’t recognizable by a population protocol. An intuitive and not
entirely incorrect explanation is that in both cases we can’t do multiplication
because we can’t do nested loops. In population protocols this is because
even though we can do a single addition that turns exactly A many blank
tokens into B’s, using the rule

A, − → A0 , B

we can’t multiply by repeated addition, because we can’t detect that the

first addition step addition has ended to start the next iteration of the outer
loop.
Below we’ll describe the correspondence between semilinear sets and
stably-computable predicates. For full details see [AAD+ 06, AAE06].

33.2.3.1 Semilinear predicates are stably computable

This part is easy. We have that any Presburger formula can be represented
as a logical combination of <, +, and ≡k operators. We can implement
P
any formula of the form ai xi < b, where ai and b are integer constants,
using the linear threshold function from §33.2.2.4. We can implement any
ai xi ≡k b using a straightforward extension of the
P
formula of the form
mod-k counter from §33.2.2.3. If we run these in parallel for each predicate
in our formula, we can then apply any logical connectives to the result.
For example, if we want to express the statement that “x is an odd
number greater than 5”, we have out agents compute separately x ≡2 1 and
x > 5; if the leader computes true for both of these, it assigns true to its real
output.
CHAPTER 33. POPULATION PROTOCOLS 321

33.2.3.2 Stably computable predicates are semilinear

This is the hard direction, because we have to exclude any possible algorithm
for computing a non-semilinear set. The full proof is pretty involved, and
can be found in [AAE06]. A survey paper of Aspnes and Ruppert [AR09]
gives a simplified proof of the weaker result (modeled on an introductory
argument in [AAE06]) that any stably-computable set is a finite union of
monoids. Like linear sets, monoids are of the form {b + ai xi }, but the
P

number of terms in the sum might be infinite.

We won’t do either of these proofs.

33.3 Random interactions

An alternative to assuming worst-case scheduling is to assume random
scheduling: at each step, a pair of distinct agents is chosen uniformly at
random to interact. This gives the population protocol substantially more
power, and (with some tweaks to allow for different reactions to occur at
different rates) is the standard assumption in chemical reaction networks.
An example of an algorithm that exploits the assumption of random
scheduling is the approximate majority protocol of Angluin, Aspnes, and
Eisenstat [AAE08b]. This protocol starts with a mix of agents in states x
and y, and uses a third state b (for blank) to allow the initial majority value
to quickly take over the entire population. The non-trivial transition rules
are:

xy → xb
yx → yb
xb → xx
bx → xx
yb → yy
by → yy

If two opposite agents meet, one becomes blank, depending on which

initiates the reaction (this is equivalent to flipping a coin under the random-
scheduling assumption). These reactions produce a supply of blank agents,
drawing equally from both sides. But when a blank agent meets a non-blank
agent, it adopts the non-blank agent’s state. This is more likely to be the
majority state, since there are more agents to meet in the majority state.
So if we consider only transitions that change the net number of x agents
CHAPTER 33. POPULATION PROTOCOLS 322

minus y agents, we get a random walk biased toward the majority value with
an absorbing barrier in the state where all agents are equal. However, the
rate at which these special transitions occur depends on how quickly blank
agents are created, which in turn depends on the relative numbers of x and
y agents.
Analysis of the full process is difficult, but Angluin et al. show that with
high probability all agents end up in the initial majority state in O(n log n)
√
interactions, provided
√ the initial majority is large enough (Ω( n log n), later
improved to Ω( n log n by Condon et al. [CHKM19]). Curiously, a later
paper by Cardelli and Csikász-Nagy [CCN12] showed that the cell cycle
controlling mitosis in all living organisms uses a chemical switch that looks
suspiciously like the approximate majority algorithm, making this algorithm
roughly three billion years old.
But we can do better than this. With random scheduling, we have
much more control over how a computation evolves, and this can be used
to simulate (with high probability) a register machine, which in turn can
be used to simulate a Turing machine. The catch is that the state of a
population protocol with n agents can be described using O(log n) bits, by
counting the number of agents in each state. So the most we can simulate is
a machine that has O(log n) space.
The original population protocol paper included a simulation of an
O(log n)-space Turing machine, but the time overhead per operation was very
bad, since most operations involved a controller agent personally adjusting
the state of some other agent, which requires at least O(n) interactions before
the controller meets its target.
A better construction was given by Angluin et al. [AAE08a], under the
assumption that the population starts with a single agent in a special leader
state. The main technique used in this paper it to propagate a message m
using an epidemic protocol mb → mm. The time for an epidemic to spread
through a population of n individuals through random pairwise interactions is
well-understood, and has the property that (a) the time to infect everybody is
Θ(log n) with high probability, and (b) it’s still Θ(log n) with high probability
if we just want to infect a polynomial fraction n of the agents.
So now the idea is that if the leader, for example, wants to test if there is
a particular state x in the population, it can spread a message x? using an
epidemic, and any agent with x can respond by starting a counter-epidemic
x!. So if there is an x, the leader finds out about it in O(log n) time, the
time for the first epidemic to go out plus the time for the second epidemic
to come back.
What if there is no x agent? Then the query goes out but nothing comes
CHAPTER 33. POPULATION PROTOCOLS 323

back. If the leader can count off Θ(n) time units (with an appropriate
constant, it can detect this. But it does not have enough states by itself to
count to Θ(log n).
The solution is to take advantage of the known spreading time for epi-
demics to build a phase clock out of epidemics. The idea here is that
the leader will always be in some phase 0 . . . m − 1. Non-leader agents
try to catch up with the leader by picking up on the latest rumor of the
leader’s phase, which is implemented formally by transitions of the form
hx, ii hF, ji → hx, ii hF, ii when 0 < i − j < m/2 (mod m). The leader on
the other hand is a hipster and doesn’t like it when everybody catches up; if
it sees a follower in the same phase, it advances to the next phase to maintain
its uniqueness: hL, ii hF, ii → hL, i + 1i hF, ii.
Because the current phase spreads like an epidemic, when the leader
advances to i + 1, every agent catches up in a log n time w.h.p. This means
both that the leader doesn’t spend too much time in i + 1 before meeting
a same-phase follower and that followers don’t get too far behind. (In
particular, followers don’t get so far behind that they start pulling other
followers forward.) But we also have that it takes at least b log n time
w.h.p. before more than n followers catch up. This gives at most an
n−1  1 probability that the leader advances twice in b log n time. By
making m large enough, the chances that this happens enough to get all
the way around the clock in less than, say b(m/2) log n) time can be made
at most n−c for any fixed c. So the leader can now count of Θ(log n) time
w.h.p., and in particular can use this to time any other epidemics that are
propagating around in parallel with the phase clock.
Angluin et al. use these techniques to implement various basic arithmetic
operations such as addition, multiplication, division, etc., on the counts of
agents in various states, which gives the register machine simulation. The
simulation can fail with nonzero probability, which is necessary because
otherwise it would allow implementing non-semilinear operations in the
adversarial scheduling model.
The assumption of an initial leader can be replaced by a leader election
algorithm, but at the time of the Angluin et al. paper, no leader election
algorithm better than the Θ(n)-time fratricide protocol described §33.2.2.1
was known, and even using this protocol requires an additional polynomial-
time cleaning step before we can run the main algorithm, to be sure that
there are no leftover phase clock remnants from deposed leaders to cause
trouble. So the question of whether this could be done faster remained open.
Hopes of finding a better leader election protocol without changing the
model ended when Doty and Soloveichek [DS15] proved a matching Ω(n)
CHAPTER 33. POPULATION PROTOCOLS 324

lower bound on the expected time to convergence for any leader election
algorithm in the more general model of chemical reaction networks. This
results holds assuming constant states and a dense initial population where
any state that appears is represented by a constant fraction of the agents.
Because of this and related lower bounds, recent work on fast population
protocols has tended to assume more states. This is a fast-moving area of
research, so I will omit trying to summarize the current state of the art here.
There are good recent surveys of this work in [AG18, ER+ 18].
Chapter 34

Mobile robots

Last updated 2016. Some material may be out of date.

Mobile robots are a model of distributed computation where the agents

(robots) are located in a plane, and have no ability to communicate except
by observing each others’ positions. Typical problems are to get a group of
robots to gather on a single point, or more generally to arrange themselves in
some specified pattern. This is complicated by the usual issues of asynchrony
and failures.

34.1 Model
We will start by describing the Suzuki-Yamashita model [SY99], the
CORDA model [Pri01], and some variants. We’ll follow the naming con-
ventions used by Agmon and Peleg [AP06].
Basic idea:

• We have a bunch of robots represented by points in the plane R2 .

• Each robot executes a look-compute-move cycle:

– Look phase: obtain snapshot of relative positions of all the other

robots.
– Compute phase: pick a new point to move to.
– Move phase: move to that point.

• Robots are dumb. Various possible limitations that might apply:

325
CHAPTER 34. MOBILE ROBOTS 326

– Anonymity: any two robots that see the same view take the
same action.
– Oblivious: The output of the compute phase is base only on
results of last look phase, and not on any previous observations.
Robots have no memory!
– No absolute coordinates: Translations of the space don’t change
the behavior of the robots.
– No sense of direction: robots don’t know which way is north.
More formally, if view v can be rotated to get view v 0 , then a
robot that sees v 0 will make the same move as in v, subject to the
same rotation.
– No sense of scale: robots don’t have a consistent linear measure.
If view v can be scaled to get view v 0 , then a robot that sees v 0
will move to the same point as in v, after applying the scaling.
– No sense of chirality: robots can’t tell counter-clockwise from
clockwise. Flipping a view flips the chosen move but has no other
effect.
– No ability to detect multiplicities: the view of other robots is a
set of points (rather than a multiset), so if two robots are on the
same point, they look like one robot.
– Fat robots: robots block the view of the robots behind them.

• Adversary can interfere in various ways:

– During move phase, robot is guaranteed to either move some

minimum distance δ > 0 or reach its target, but adversary can
stop a robot after it has moved δ.
– Look-compute-move phases are asynchronous, and adversary can
schedule robots subject to various constraints.
∗ Asynchronous model (ASYNC): The adversary can delay
a robot between look and move phases, so that robots might
be moving based on out-of-date information.
∗ Semi-synchronous model (SSYNC): Each look-compute-
move cycle is an atomic action, so moves are always based on
current information. The adversary may schedule more one
or more robots to do their look-compute-move in each round.
Also known as the ATOM model. This was the model given
by Suzuki and Yamashita [SY99].
CHAPTER 34. MOBILE ROBOTS 327

∗ Fully synchronous model (FSYNC): Like SSYNC, but

every robot is active in every round.
– But we do have fairness: the adversary must activate each (non-
faulty) robot infinitely often.
• We may also have faults:
– Byzantine faults: Byzantine robots can move anywhere they
like.
– Crash faults: crashed robots don’t move even when they are
supposed to.

The simplest goal is to gather the non-faulty robots together on a single

point despite all these possible disasters. Other goals might be formation of
particular shapes. An additional source of variation here is whether we want
exact gathering (every robot eventually gets to exactly where it should be)
or just convergence (over time, robots get closer and closer to where they
should be).
Below, we will mostly be looking at the semi-synchronous model, with the
assumption that robots are anonymous and oblivious, and have no absolute
coordinates, sense of direction, or sense of scale. However, we will generally
let robots detect multiplicity. Depending on the results we are describing,
we may or may not assume chirality.

34.2 Two robots, no faults

Suzuki and Yamashita [SY99] showed that it’s impossible to get two deter-
ministic, oblivious robots to the same point in the semi-synchronous model
assuming no absolute coordinates and no sense of direction, although they
can converge. The convergence algorithm is simple: have each robot move
to the midpoint of the two robots whenever it is activated. This always
reduces the distance between the robots by min(δ, d/2). But it doesn’t solve
gathering if only one robot moves at a time.
This turns out to be true in general [SY99, Theorem 3.1]. The idea is
this: Suppose we have an oblivious algorithm for gathering. Consider two
robots at distinct points p and q, and suppose that after one round they
both move to r. There are two cases:
1. Both robots move. By symmetry, r = (p + q)/2. So now construct a
different execution in which only one robot moves (say, the one that
moved least recently, to avoid running into fairness).
CHAPTER 34. MOBILE ROBOTS 328

2. Only one robot moves. Without loss of generality, suppose the robot
at p moves to q. Then there is a different execution where q also moves
to p and the robots switch places.
In either case the distance between the two robots in the modified
execution is at least half the original distance. In particular, it’s not zero.
Note that this works even if the adversary can’t stop a robot in mid-move.
Both obliviousness and the lack of coordinates and sense of direction
are necessary. If the robots are not oblivious, then they can try moving
to the midpoint, and if only one of them moves then it stays put until the
other one catches up. If the robots have absolute coordinates or a sense of
direction, then we can deterministically choose one of the two initial positions
as the ultimate gathering point (say, the northmost position, or the westmost
position if both are equally far north). But if we don’t have any of this we
are in trouble.
Like the 3-process impossibility result for Byzantine agreement, the 2-
process impossibility result for robot gathering extends to any even number of
robots where half of them are on one point and half on the other. Anonymity
then means that each group of robots acts the same way a single robot would
if we activate them all together. Later work (e.g., [BDT12]) refers to this as
bivalent configuration, and it turns out to be the only initial configuration
for which it is not possible to solve gathering absent Byzantine faults.

34.3 Three robots

Agmon and Peleg [AP06] show that with three robots, it is possible to
solve gathering in the SSYNC model with one crash fault but not with one
Byzantine fault. We’ll start with the crash-fault algorithm. Given a view
v = {p1 , p2 , p3 }, this sends each robot to the “goal” point pG determined
according to the following rules:
1. If v has a point p with more than one robot, set pG = p.

2. If p1 , p2 , and p3 are collinear, set pG to the middle point.

3. If p1 , p2 , and p3 form an obtuse triangle (one with a corner whose angle

is ≥ π/2, set pG to the obtuse corner.

4. If p1 , p2 , and p3 form an acute triangle (one with no angles ≥ π/2), set

pG to the intersection of the angle bisectors.
Here is a sketch of why this works. For the real proof see [AP06].
CHAPTER 34. MOBILE ROBOTS 329

1. If we are in a configuration with multiplicity > 1, any non-faulty robot

not on the multiplicity point eventually gets there.

2. If we are in a collinear configuration, we stay collinear until eventually

one of the outer robots gets to the middle point, giving a configuration
with multiplicity > 1.

3. If we are in an obtuse-triangle configuration, we stay in an obtuse-

triangle configuration until eventually one of the acute-corner robots
gets to the obtuse corner, again giving a configuration with multiplicity
> 1.

4. If we are in an acute-triangle configuration, then a somewhat messy

geometric argument shows that if at least one robot moves at least δ
toward the intersection of the angle bisectors, then the circumference
of the triangle drops by cδ for some constant c > 0. This eventually
leads either to the obtuse-triangle case (if we happen to open up one
of the angles enough) or the multiplicity > 1 case (if the circumference
drops to zero).

However, once we have a Byzantine fault, this blows up. This is shown
by considering a lot of cases, and giving a strategy for the adversary and the
Byzantine robot to cooperate to prevent the other two robots from gathering
in each case. This applies to both algorithms for gathering and convergence:
the bad guys can arrange so that the algorithm eventually makes no progress
at all.
The first trick is to observe that any working algorithm for the n =
3, f = 1 case must be hyperactive: every robot attempts to move in every
configuration with multiplicity 1. If not, the adversary can (a) activate the
non-moving robot (which has no effect); (b) stall the moving non-faulty robot
if any, and (c) move the Byzantine robot to a symmetric position relative to
the first two so that the non-moving robot become the moving robot in the
next round and vice versa. This gives an infinite execution with no progress.
The second trick is to observe that if we can ever reach a configuration
where two robots move in a way that places them further away from each
other (a diverging configuration), then we can keep those two robots at
the same or greater distance forever. This depends on the adversary being
able to stop a robot in the middle of its move, which in turn depends on the
robot moving at least δ before the adversary stops it. But if the robots have
no sense of scale, then we can scale the initial configuration so that this is
not a problem.
CHAPTER 34. MOBILE ROBOTS 330

Here is the actual argument: Suppose that from positions p0 and q0

there is a step in which the non-faulty robots move to p1 and q1 with
d(p1 , q1 ) > d(p, q). Starting from p1 and q1 , run both robots until they
are heading for states p2 and q2 with d(p2 , q2 ) ≤ d(p0 , q0 ). By continuity,
somewhere along the paths p1 p2 and q1 q2 there are intermediate points p02
and q20 with d(p02 , q20 ) = d(p0 , q0 ). Stop the robots at these points, move the
Byzantine robot r to the appropriate location to make everything look like
the initial p0 , q0 configuration, and we are back where we started.
So now we know that (a) we have to move every robot, and (b) we can’t
move any two robots away from each other. In the full version of the proof,
this is used to show by an extensive case analysis that as long as we start
with no two robots on the same point, this always either makes no progress
or reaches three distinct points on the same line. We’ll skip over this part
and just argue that once we have three hyperactive collinear robots, that
two of them are diverging. This will show that in the worst case we can’t
win, because the adversary could start everybody out on the same line, but
it is not quite as general as the full result of Agmon and Peleg.
Suppose the robots are at positions p1 < p2 < p3 . Then p2 has to move
either left or right, which means moving away from either p3 or p1 . In either
case we have a diverging pair (because the adversary can elect not to move
the robots on the end). So now the divergence argument goes through, and
we are done.

34.4 Many robots, with crash failures

It turns out that we can solve the gathering problem even if we have many
robots and some of them can crash, as long as the robots do not start on
the same line. The reason for this is that any set of non-collinear points
x1 , . . . , xn in R2 has a unique geometric median, defined as the point m
that minimizes ni=1 d(m, xi ), and the geometric median is unchanged if we
P

move any of the points towards it.

So the algorithm is for all the robots to walk toward this point. It doesn’t
matter if some of the robots don’t move, or don’t move at the same speed,
because the median doesn’t change. Eventually, all the non-faulty processes
will reach it.
There is one drawback to this approach, which is that even though
very good approximation algorithms exist [CLM+ 16], the geometric median
appears to be difficult to compute exactly. We could declare that we are
willing to assume that our robots have infinite computational power, but this
CHAPTER 34. MOBILE ROBOTS 331

is not an easy assumption to justify in this case. An alternative is to build an

algorithm that marches toward the geometric median in certain cases where
it is straightforward to compute, and does something more sophisticated
otherwise. This approach was taken by Bramas and Tixeuil [BT15], who
also supplied the idea of using the geometric median in the first place. We
will not go into detail about their algorithm.
Chapter 35

Beeping

Last updated 2016. Some material may be out of date.

The (discrete) beeping model was introduced by Cornejo and Kuhn [CK10]
to study what can be computed in a wireless network where communication
is limited to nothing but carrier sensing. According to the authors, the model
is inspired in part by some earlier work on specific algorithms based on carrier
sensing due to Scheideler et al. [SRS08] and Flury and Wattenhofer [FW10].
It has in turn spawned a significant literature, not only in its original domain
of wireless networking, but also in analysis of biological systems, which often
rely on very limited signaling mechanisms. Some of this work extends or
adjusts the capabilities of the processes in various ways, but the essential
idea of tightly limited communication remains.
In its simplest form, the model consists of synchronous processes organized
in an undirected graph. Processes wake up at arbitrary rounds chosen by the
adversary, and do not know which round they are in except by counting the
number of rounds since they woke. Once awake, a process chooses in each
round to either send (beep) or listen. A process that sends learns nothing in
that round. A process that listens learns whether any of its neighbors sends,
but not how many or which one(s).
From a practical perspective, the justification for the model is that carrier
sensing is cheap and widely available in radio networks. From a theoretical
perspective, the idea is to make the communication mechanism as restrictive
as possible while still allowing some sort of distributed computing. The
assumption of synchrony both adds to and limits the power of the model.
With no synchrony at all, it’s difficult to see how to communicate anything
with beeps, since each process will just see either a finite or infinite sequence
of beeps with not much correlation to its own actions. With continuous

332
CHAPTER 35. BEEPING 333

time, subtle changes in timing can be used to transmit arbitrarily detailed

information. So the assumption of a small number of synchronous rounds is a
compromise between these two extremes. The assumption that processes wake
at different times and have no common sense of time prevents synchronization
on rounds, for example by reserving certain rounds for beeps by processes
with particular ids. It is up to the protocol to work around these limitations.

35.1 Interval coloring

One way to get around the problem of not having a common global clock
is to solve interval coloring, the main problem considered by Cornejo
and Kuhn. This is related to TDMA multiplexing in cell phone networks,
and involves partitioning a repeating interval of T rounds in a network
with maximum degree ∆ into subintervals of length Ω(T /∆) such that each
process is assigned a subinterval and no two adjacent processes are assigned
overlapping subintervals. The idea is that these intervals can then be used to
decide when each process is allowed to use its main radio to communicate.1
Cornejo and Kuhn give an algorithm for interval coloring that assigned
a subinterval of length Ω(T /∆) to each process assuming that the size of
the interval T is known to all processes and that T is at least a constant
multiple of ∆. However, the processes do not know anything about the
structure of the graph, and in particular do not know ∆. This requires
each process to get an estimate of the size of its neighborhood (so that it
knows how large a subinterval to try to acquire) and to have a mechanism
for collision detection that keeps it from grabbing an interval that overlaps
with a neighbor’s interval. The process is complicated by the fact that my
length-T intervals and your length-T intervals may be offset from each other,
and that I can’t detect that you are beeping if you and I are beeping at the
same time.
To simplify things a bit, the presentation below will assume that the
graph is regular, so that d(v) equals the maximum degree ∆ for all nodes
in the graph. The paper [CK10] gives an analysis that does not need this
assumption. We’ll also wave our hands around a lot instead of doing actual
algebra in many places.
1
We may really want 2-hop coloring here, where no two of my neighbors get the same
color, because this is what (a) allows me to tell my neighbors apart, and (b) allows my
neighbors not to interfere with each other, but that is a subject for later papers (see, for
example, [MRZ15].
CHAPTER 35. BEEPING 334

35.1.1 Estimating the degree

The basic idea is to have each process beep once in every T consecutive slots.
Each process divides time into periods of length T , starting when it wakes
up. Because processes wake up at different times, my period might overlap
with up to two of yours. This means that if S is the set of times during my
period where I hear beeps, then S includes at most two beeps per process,
so |S| is at most twice my actual degree. This gives an upper bound on d(v),
and indeed each process will just use the maximum number of beeps it heard
in the last period as the basis for its estimate of d(v).
For the lower bound side, we want to argue that if processes choose slots
at random in a way that is more likely to avoid collisions than create them,
and there are enough slots, then we expect to get few enough collisions that
|S| is also Ω(∆). The details of this depend on the mechanism for choosing
slots, but if we imagine slots are chosen uniformly, then E [|S|] ≥ ∆(1 − ∆/T ),
which is Ω(∆) under our assumption that T ≥ c∆ for some sufficiently large
c. We can compensate by the error by inserting a fudge factor η, chosen so
that (1/η)|S| is very likely to be an upper bound on the degree.

35.1.2 Picking slots

T
Each process will try to grab a subinterval of size b = η |S|+1 , where η is the
fudge factor mentioned above. If it has not already picked a position p, then
it chooses one uniformly at random from the set of all positions p such that
S[p − b − 2, p + b + 1] from the most recent period includes no beeps. Because
this selection criterion knocks out up to (2b + 4)|S| possible choices, it does
tend to concentrate uncolored processes on a smaller range of positions than
a uniform pick from the entire interval, increasing the likelihood of collisions.
|S|
But we can choose η to make 2(b + 4)|S| = 2ηT |S|+1 a small enough fraction
of T that this is not a problem.

35.1.3 Detecting collisions

The basic idea for detecting a collision is that I will abandon my color p if I
hear any beeps in [p − b − 2, p + b + 1] during the next period. This works
great as long as nobody chooses exactly the same p as me. To avoid this,
each process flips a coin and beeps at either p or p + 1. So even if I choose
the same slot as one or more of my neighbors, there is a 1/2 chance per
period that I detect this and pick a new color next time around.
What this means is that in each round (a) I have a constant probability
of getting a good estimate of my degree (which means I set b correctly); (b)
CHAPTER 35. BEEPING 335

I have a constant probability of detecting a collision with a neighbor if there

is one (which means I pick a new position if I have to); and (c) I have a
constant probability that if I pick a new position it is a good one. If we
repeat all these constant-probability wins for O(log n) periods, then all n
processes win, and we are done.

35.2 Maximal independent set

A high-impact early result in the beeping model was a paper by Afek et
al.. [AAB+ 11] that showed that a biological mechanism used in fruit-fly
sensory organ development to choose a subset of cells that are not too close
to each other can be viewed as an implementation of maximal independent
set using beeps. As a distributed algorithm, this algorithm is not so good,
so instead we will talk about a follow-up paper [AABJ+ 11] by some of the
same authors on more effective beeping algorithms for MIS.
Recall that a subset of the vertices of a graph is indepedent if no two
vertices in the set are adjacent. A maximal independent set (MIS) is an
independent set of vertices that can’t be increased without including adjacent
vertices. Equivalently, it’s an independent set where every non-member is
adjacent to some member.
Afek et al. give a couple of algorithms for beeping MIS that require
either special knowledge of the graph or extensions to the beeping model.
The justification for this is a lower bound, which they also give, that shows
that without any knowledge p of the graph, computing an MIS in the standard
beeping model takes Ω( n/ log n) time with constant probaiblity. We’ll
describe the lower bound and then show how to compute MIS in O(log3 n)
time given a polynomial upper bound on n.

35.2.1 Lower bound

For the lower bound, the idea is to exploit the fact that the adversary can
wake up nodes over time. To avoid allowing the adversary to delay the
algorithm from finishing indefinitely by just not waking up some nodes for a
while, the running time is computed as the maximum time from when any
particular node p wakes up to when p converges to being in the MIS or not.
The cornerstone of the proof is the observation that if a process doesn’t
know the size of the graph, then it has to decide whether to beep or not
within a constant number of rounds. Specifically, for any fixed sequence of
beeps b0 , b1 , . . . , where bi is an indicator variable for whether the process
hears a beep in round i after it wakes up, either the process never beeps
CHAPTER 35. BEEPING 336

or there are constant ` and p such that the process beeps in round ` with
probability p. This follows because if the process is ever going to beep, there
is some first round ` where it might beep, and the probability that it does
so is constant because it depends only on the algorithm and the sequence b,
and not on n.
If an algorithm that hears only silence remains silent, then nobody ever
beeps, and nobody learns anything about the graph. Without knowing
anything, it’s impossible to correctly compute an MIS (consider a graph with
only two nodes that might or might not have an edge between them). This
means that in any working algorithm, there is some round ` and probability
p such that each process beeps with probability p after ` rounds of silence.
We can now beep the heck out of everybody by assembling groups of
1
Θ( p log n) processes and waking up each one ` rounds before we want them
to deliver their beeps. But we need to be a little bit careful to keep the
graph from being so connected that the algorithm finds an MIS despite this.
There are two cases, depending on what a process that hears only beeps
does:

1. If a process that hears only beeps stays silent forever, then we build
a graph with k − 1 cliques C1 , . . . , Ck−1 of size Θ( kp log n) each, and
a set of k cliques U1 , . . . , Uk of size Θ(log n) each. Here k  ` is a
placeholder that will be filled in later (foreshadowing: it’s the biggest
value that doesn’t give us more than n processes). Each Ci clique is
further partitioned into subcliques Ci1 , . . . , Cik of size Θ( p1 log n) each.
Each Cij is attached to Uj by a complete bipartite graph.
We wake up each clique Ci in round i, and wake up all the U cliques
in round `. We can prove by induction on rounds that with high
probability, at least one process in each Cij beeps in round i + `, which
means that every process in every Ui hears a beep in the first k − 1
rounds that it is awake, and remains silent, causing the later C cliques
to continue to beep.
Because each Ci is a clique, each contains at most one element of the
M IS, and so between them they contain at most k − 1 elements of
the MIS. But there are k U cliques, so one of them is not adjacent to
any MIS element in a C clique. This means that one of the Uj must
contain an MIS element.
So now we ask when this extra Uj picks an MIS element. If it’s in the
first k − 1 rounds after it wakes up, then all elements have seen the
same history, so if any of them attempt to join the MIS then all of
CHAPTER 35. BEEPING 337

them do with independent constant probability each. This implies that

we can’t converge to the MIS until at least k rounds.
Now we pick k to be as large as possible so pthat the total number of
processes Θ(k 2 log n) = n. This gives k = Ω( n/ log n) as claimed.
2. If a process starts beeping with probability p0 after hearing beeps for
m rounds, then we can’t apply the silent-beeper strategy because the
C cliques will stop hearing silence. Instead, we replace the C cliques
with new cliques S1 , . . . , Sm−1 of size Θ( p1 log n) each. We start the
process by having the S cliques shout at the U cliques for the first
m − 1 rounds. After this, we can start having the U cliques shout at
each other: each clique Uj is connected to q earlier cliques, consisting
of up to q Uj 0 for j < j and enough Si to fill out the remainder.
We now argue that if a process that hears only beeps chooses to join the
MIS with constant probability after q rounds, then every U clique gets
at least two processes joining with high probability, which is trouble.
Alternatively, if no process in a U clique tries to join the MIS for at
least q rounds, then for q = O(n/ log n), there are U cliques that are
connected only to other U cliques, which again means we don’t get an
MIS. So in this case we get a lower bound of Ω(n/ log n) on the time
for each node to converge.
The lower bound in the paper is actually a bit stronger than this, since
it allows the processes to send more detailed messages than beeps as long
as there are no collisions. Reducing this back to beeping means tuning the
constants so we get at least two messages out of every clique.

35.2.2 Upper bound with known bound on n

Algorithm 35.1 [AABJ+ 11] converges to a maximal independent set in
O(log2 N log n) rounds, from any initial state, given an upper bound N
on the number of processes n.
The proof that this works is a bit involved, so if you want to see all the
details, you should look at the paper. The intuition runs like this:
1. At least one of any two adjacent processes that both think they are in
the MIS will eventually notice the other during the final phase, causing
it to restart.
2. If I start at the beginning of the protocol, and I have a neighbor already
in the MIS, then I will hear them during my initial listening phase and
restart.
CHAPTER 35. BEEPING 338

1 Leave MIS and restart the algorithm here upon hearing a beep
2 for c lg2 N rounds do
3 listen
4 for i ← 1 to lg N do
5 for c lg N rounds do
2i
6 with probability 8N do
7 beep
8 else
9 listen

10 Join MIS
11 while I don’t hear any beeps do
12 with probability 12 do
13 beep
14 listen
15 else
16 listen
17 beep;

Algorithm 35.1: Beeping a maximal independent set (from

[AABJ+ 11]
CHAPTER 35. BEEPING 339

3. If two adjacent nodes both execute the middle phase of increasing-

probability beeps, then one of them will go through a phase where
it listens with probability at least 1/2 while the other beeps with
probability at least 1/2 (note that this might not be the same phase for
both, because the nodes might not start at the same time). This gives
at least a 1/4 chance per round that the likely listener drops out, for at
least a 1 − n−c/2 chance that it drops out during the c lg n rounds that
it listens with this probability, assuming its neighbor does not drop
out. This means that by tuning c large enough, we can make it highly
improbable that any pair of neighbors both enter the MIS (and if they
do, eventually at least one drops out). So we eventually get a set that
is independent, but maybe not maximal.

4. The hard part: After O(log2 N log n) rounds, it holds with high prob-
ability that every node is either in the MIS or has a neighbor in the
MIS. This will give us that the alleged MIS is in fact maximal.
The bad case for termination is when some node u hears a neighbor v
that is then knocked out by one of its neighbors w. So now u is not in
the MIS, but neither is its (possibly only) neighbor v. The paper gives
a rather detailed argument that this can’t happen too often, which we
will not attempt to reproduce here. The basic idea is that if one of v’s
neighbors were going to knock v shortly after v first beeps, then the
sum of the probabilities of those neighbors beeping must be pretty high
(because at least one of them has to be beeping instead of listening
when v beeps). But they don’t increase their beeping probabilities
very fast, so if this is the case, then with high probability one of them
would have beeped in the previous c log N rounds before v does. So
the most likely scenario is that v knocks out u and knocks out the
rest of its neighbors at the same time, causing it to enter the MIS and
remain there forever. This doesn’t happen always, so we might have to
have some processes go through the whole O(log2 N ) initial rounds of
the algorithm more than once before the MIS converges. But O(log n)
attempts turn out to be enough to make it work in the end.
Appendix

340
Appendix A

Assignments

Assignments should be uploaded to Canvas in PDF format.

Do not include any identifying information in your submissions.
This will allow grading to be done anonymously.
Make sure that your submissions are readable. You are strongly
advised to use LATEX, Microsoft Word, Google Docs, or similar software to
generate typeset solutions. Scanned or photographed handwritten submis-
sions often come out badly, and submissions that are difficult for the grader
to read will be penalized.
Sample solutions will appear in this appendix after the assignment is due.
To maintain anonymity of both submitters and graders, questions about
grading should be submitted through Canvas.
Questions about assignments can be sent to the instructor directly at
[email protected], or submitted to the course Piazza site at piazza.
com/yale/fall2020/cpsc465.

A.0 Assignment 0: due as soon as possible

Send me email! My address is [email protected].
In your message, include:

1. Your name.
2. Your status: whether you are an undergraduate, grad student, auditor,
etc.
3. Whether you are taking the course as CPSC 465 or CPSC 565.
4. Anything else you’d like to say.

341
APPENDIX A. ASSIGNMENTS 342

(Assignment 0 will not be graded, but you should do it anyway.)

A.1 Assignment 1: due Wednesday, 2020-09-23, at

5:00pm Eastern US time
A.1.1 A token-passing game
Suppose we have an asynchronous bidirectional message-passing network in
the form of a connected graph, where initially m of the n nodes possess a
token, represented by a local variable hasToken being set to true. We’d like
to be able to move the tokens around, while preserving the total number of
tokens.

1. Show that no algorithm that allows tokens to move can guarantee that
there are exactly m tokens in any reachable configuration.

2. Give an algorithm that satisfies the following two properties, starting

with a configuration with m tokens:

(a) Safety: In any reachable configuration, there are at most m tokens.

You should give an explicit invariant that implies this, and show
that any transition of your algorithm preserves the invariant.
(b) Liveness: From any reachable configuration C0 , for any subset S
of the processes with |S| = m, there exists an execution starting
in C0 that ends with a configuration in which every process in S
has a token.1

To keep things simple, you may assume that the processes can make
non-deterministic choices. For example, a process p might choose
arbitrarily between sending a message to a neighbor q or to a different
neighbor r, and each choice leads to a different possible execution.

Solution
1. Suppose that we are preserving total tokens. Consider some transition
between configurations C1 and C2 . If some process switches hasToken
1
Strictly speaking, this is a lot weaker than the usual definition of liveness, because it
effectively assumes that the adversary is cooperating with us. In retrospect I should have
written this as “for any admissible adversary strategy, there is a sequence of nondeterministic
choices by the algorithm that causes the execution to reach a desired configuration.” But I
didn’t write this, and so it’s fine to answer the problem I did write.
APPENDIX A. ASSIGNMENTS 343

from 1 to 0 between these configurations, then some other process must

switch hasToken from 0 to 1. But the definition of delivery events in
the asynchronous message-passing model only allows one process at a
time to change its state. It follows that no process can change hasToken
from 1 to 0 in any transition, so tokens can’t move.
2. Consider the following algorithm:
• At any time, a process with hasToken = 1 may send a message
takeThis to any of its neighbors and set hasToken = 0.
• A process that receives takeThis when hasToken = 1 sends takeThis
to any of its neighbors. A process that receives takeThis when
hasToken = 0 may either set hasToken = 1 or send takeThis to
any of its neighbors.
Let us show that this has the desired properties:
(a) Safety: Our invariant will be that the sum of the number
of processes with hasToken = 1 plus the number of takeThis
messages in transit will be m.
The invariant holds in the initial configuration because there
are exactly m processes with hasToken = 1 and no message
in transit.
It is preserved by transitions, because in each possible transi-
tion, either:
i. Some process changes hasToken = 1 to hasToken = 0 and
generates a takeThis message;
ii. Some process changes hasToken = 0 to hasToken = 1
while consuming a takeThis message; or
iii. Some process consumes a takeThis message but generates
a new takeThis message.
In each case, the total number of tokens plus messages is
preserved.
(b) Liveness: For any configuration C, let T (C) be the set of
processes with hasToken = 1. We will argue that if T (C) 6= S,
there exists a partial execution that increases |T (C) ∩ S| by
1. First pick some p ∈ S \ T (C). Now consider two cases:
i. If there is at least one takeThis message t in transit, apply
the following strategy. Deliver t. If the recipient of t is
p, set p.hasToken = 1. If not, have the recipient send
takeThis to some neighbor that is closer to p than it is.
Repeat this process until a takeThis message reaches p.
APPENDIX A. ASSIGNMENTS 344

ii. If there is no takeThis message in transit, generate a

takeThis message at some q ∈ T (C) \ S while setting
q.hasToken to 0. Then apply the previous case.
For any configuration with T (C) 6= S, at least one of these
two conditions will hold because of the safety property.
Each partial execution defined above increases |T (C) ∩ S| by
one, and we can only increase |T (C) ∩ S| at most m times
because |S| = m, so after at most m of these partial executions
we reach a configuration with T (C) = S.

A.1.2 A load-balancing problem

Consider a two-way message-passing ring with n = mk nodes, where m > 1
and k is odd. Nodes at positions 0, k, 2k, . . . , (m − 1)k are initially marked
as leaders, while nodes at other positions are followers. All nodes have a
sense of direction, and can distinguish their left neighbor from their right,
but they do not have any other id information.
Algorithm A.1 is intended to allow the leaders to recruit followers. It is
not hard to show that every follower eventually adds itself to a tree of parent
pointers rooted at some leader. We would like all of these trees to contain
roughly the same number of nodes.

1 initially do
2 if I am a leader then
3 parent ← id
4 send recruit to both neighbors
5 else
6 parent ← ⊥

7 upon receiving recruit from p do

8 if parent = ⊥ then
9 parent ← p
10 send recruit to my neighbor who is not p

Algorithm A.1: Recruiting algorithm for Problem A.1.2.

1. Suppose we run this algorithm in a synchronous system. What is the

minimum and maximum possible size of a tree?
APPENDIX A. ASSIGNMENTS 345

2. Suppose instead we run the algorithm in an asynchronous system. Now

what is the minimum and maximum possible size of a tree?

3. Give an algorithm for the asynchronous version of this model that

guarantees that all trees are the same size.

Solution
1. In a synchronous execution, we can prove by induction that for each t
with 0 ≤ t ≤ k−12 , and each 0 ≤ i ≤ m − 1, each node at position ik ± t
joins the tree rooted at ik at time t. This puts exactly k nodes in each
tree.

2. In an asynchronous execution, by rushing messages from 0, we can

recruit all nodes in the range [−k + 1, k − 1] to the 0 tree before any
other messages are delivered. Conversely, each leader ik can’t recruit
nodes (i − 1)k or (i + 1)k, because these are leaders. So the maximum
size of any tree is 2k − 1.
For the minimum size, suppose we rush all messages from nodes k and
(m − 1)k. Then nodes 1 and m − 1 are recruited into the trees rooted
at these nodes before either message from 0 is delivered. This shows
that there are executions with a minimum tree size of 1.

3. The easiest fix may be to have each leader initially send just one recruit
message to the right. For each i, this recruits all agents ik, . . . , ik+(k−1)
to a tree of size k rooted at ik.

A.2 Assignment 2: due Wednesday, 2020-10-07, at

5:00pm Eastern US time
A.2.1 Synchronous agreement with limited broadcast
Suppose that we are given a synchronous message passing system on a
complete network in which messages are replaced by k-way broadcasts, where
the recipient is replaced by a recipient list of up to k processes. Suppose
further that when a process crashes in round r, each of its round-r messages
is either delivered to all of the processes on the message’s recipient list or to
none of them. A process can send as many messages as it likes to as many
groups as it likes, but if it crashes in some round, any subset of the messages
sent in that round may be lost.
APPENDIX A. ASSIGNMENTS 346

Show that for it is possible to solve agreement in this model in O(f /k)
rounds, assuming n > f .

Solution
We’ll use the flooding algorithm of Dolev and Strong [DS83] (see §9.2), but

replace sending S to all n processes in each round with sending S to all nk
possible recipient lists. As in the original algorithm, we want to prove that
after some round with few enough failures, all the non-faulty processes have
the same set.
Let Sir be the set stored by process i after r rounds. Suppose there is
some round r + 1 in which fewer than k processes fail. Then every recipient
list in round r includes a process that does not fail in round r + 1. Let L
be the set of processes that successfully deliver a message to at least one
recipient list in round r, and let S = ∪i∈L Sir . Then for each value v ∈ S,
there is some process that receives v during round r, does not crash in round
r + 1, and so retransmits v to all processes in round r + 1, causing it to be
added to Sir+2 . On the other hand, for any v 6∈ S, v is not transmitted to any
recipient list in round r, which means that no non-faulty process i includes
v in Sir+1 . So S ⊆ Sir+2 ⊆ ∪j Sjr+1 ⊆ S for all i, and the usual induction
0
argument shows that Sir continues to equal S for all non-faulty i and all
r0 ≥ r + 2.
We can have at most bf /kc rounds with ≥ k crashes before we run out,
so the latest possible round in which we have fewer than k crashes is is
r = bf /kc + 1, giving agreement after bf /kc + 2 rounds (since we don’t need
to send any messages in round r + 2).
(With some tinkering, it is not too hard to adapt the Dolev-Strong lower
bound to get a bf /kc + 1 lower bound for this model. The main issue is
now we have to crash k processes fully in round r + 1 before we can remove
one outgoing broadcast from a process in round r, which means we need to
budget tk failures to break a t-round protocol. The details are otherwise
pretty much the same as described in §9.3.)

A.2.2 Asynchronous agreement with limited failures

Algorithm A.2 describes an algorithm for asynchronous agreement with f
crash failures in a fully-connected message-passing network. The idea is to
collect values from n − f other processes in each of m rounds, and then
decide on the smallest value collected.
APPENDIX A. ASSIGNMENTS 347

1 preference ← input
2 for i ← 1 to m do
3 send hi, preferencei to all processes
4 wait to receive hi, vi from n − f processes
5 for each hi, vi received do
6 preference ← min(preference, v)

7decide preference
Algorithm A.2: Candidate algorithm for asynchronous agreement

The value m is a parameter of the algorithm and may depend on n and

f.
As usual, when waiting for messages from round i, any messages delivered
for with other round numbers will be buffered internally and processed when
the algorithm is ready for them.
Note that when a process sends a message to all process, that includes
itself.
Show that, for any n and 0 < f < n/2, there exists a value of m such
that Algorithm A.2 satisfies agreement, termination, and validity; or show
how to construct an execution for any n, 0 < f < n/2, and m that causes
Algorithm A.2 to fail at least one of these requirements.

Solution
We know from the FLP bound ([FLP85], Chapter 11) that Algorithm A.2
can’t work. So the only question is how to find an execution that shows it
doesn’t work.
It’s not too hard to see that Algorithm A.2 satisfies both termination
and validity. So we need to find a problem with agreement.
The easiest way I can see to do this is to pick a patsy process p and give it
input 0, while giving all the other processes input 1. Now run Algorithm A.2
while delaying all outgoing messages hi, vi from p until after the receiver has
finished the protocol. Because each other process is waiting for n − f ≤ n − 1
messages, this will not prevent the other processes from finishing. But all
the other processes have input 1, so we have an invariant that messages in
transit from processes other than p and preferences of processes other than p
will be 1 that holds as long as no messages from p are delivered. This results
in the non-p processes all deciding 1. We can then run p to completion, at
which point it will decide 0.
APPENDIX A. ASSIGNMENTS 348

A.3 Assignment 3: due Wednesday, 2020-10-21, at

5:00pm Eastern US time
A.3.1 Too many Byzantine processes
The phase king algorithm (Algorithm 10.2) described in §10.2.2 solves Byzan-
tine agreement for f < n/4 processes. For larger values of f , it may fail by
violating one or more of the properties of termination, validity, or agreement.
For this algorithm:
1. How big does f need to be to prevent termination?

2. How big does f need to be to prevent validity?

3. How big does f need to be to prevent agreement?

Assume that the processes know the new bound on f , and any thresholds
in the algorithm that use f are adjusted to correspond to this new bound.

Solution
1. Termination: The algorithm always terminates in f + 1 synchronous
rounds, so f doesn’t matter.

2. Validity: To violate validity, we need to convince some non-faulty

process to decide on the wrong value when all non-faulty processes
have the same input.
Suppose all the non-faulty processes have input 0, and we want to
introduce a 1 somewhere. Each process updates its preference in
each round to be either the majority value it sees, if this value has
multiplicity greater than n/2 + f , or the kingMajority broadcast by the
phase king otherwise.
If f < n/2, it’s going to be hard to show a process a bogus majority.
But a Byzantine phase king gives us more options. Suppose that all
the f Byzantine processes send out 1 in all rounds. Then for f ≥ n/4,
the multiplicity of the correct value 0 will be n − f ≤ (3/4)n, while the
required multiplicity to ignore the phase king will be strictly greater
than n/2 + f ≥ (3/4)n. So at f = n/4, all non-faulty processes adopt
the phase king’s bad value 1. In any subsequent round, we can just run
the algorithm with the Byzantine agents pretending to be non-faulty
processes with preference 1, and eventually all processes incorrectly
decide 1.
APPENDIX A. ASSIGNMENTS 349

3. Agreement: Now we need to get two non-faulty processes to decide

different values. Wait to the last round, and use f = n/4 Byzantine
processes to prevent the non-faulty processes from seeing a high enough
multiplicity on any majority value to accept it, and use a Byzantine
phase king to transmit different kingMajority values to different non-
faulty processes. So again, the algorithm fails at f = n/4.

A.3.2 Committee election

Consider the following committee election problem in an asynchronous
message-passing system with f < n/2 crash failures. Each process runs a
committee election protocol, at the end of which it receives a value 1 (elected)
or 0 (not elected). The requirements of the protocol are:

1. Nonempty committee: If no processes fail, at least one process receives

1.

2. No latecomers: In any execution, if some process p finishes the protocol

before another process q starts the protocol, then q receives 0.

Give an algorithm that solves this problem, and show that it satisfies
these requirements.
(For the purpose of defining when a process starts or ends the protocol,
imagine that it uses explicit invoke and respond events. Your protocol should
have the property that all non-faulty processes eventually terminate.)

Solution
The easiest way to do this may be to use ABD (see §16.2). Algorithm A.3
has each process read the simulated register, which we assume is initialized
to 1, then write a 0 before returning the value it read.

1 Let r be an ABD register initialized to 1.

2 procedure elect
3 onCommittee ← r
4 r←0
5 return onCommittee
Algorithm A.3: Committee election using ABD

This satisfies nonempty committee, because the first operation in the

linearization of the register must be a read operation that returns 1. It
APPENDIX A. ASSIGNMENTS 350

satisfies no latecomers, because if p finishes before q starts, then p’s write

finishes before q’s read starts, and linearizability of ABD implies q reads a 0.
This takes 3 round-trips to finish (2 for the ABD read and 1 for the
ABD write). It is not too hard to reduce this to 2 round-trips by replacing
the embedded write in the ABD read operation with a write of 1, but this
requires a more detailed correctness argument.

A.4 Assignment 4: due Wednesday, 2020-11-04, at

5:00pm Eastern US time
A.4.1 Counting without snapshots
Algorithm A.4 gives a wait-free implementation of a generalized counter
using a collect. The inc(v) procedure adjusts the value of the counter by
v: if it was x before inc(v), it should be x + v after. The read procedure
returns the current value of the counter. Assume that the initial value of the
counter is 0, as are the initial values of the registers A[i] that implement it.

1 procedure inc(v)
2 A[i] ← A[i] + v
3 procedure read()
4 s←0
5 for j ← 1 to n do
6 s ← s + A[j]
7 return s
Algorithm A.4: An alleged counter. Code for process i.

This counter implementation is not linearizable in all executions, but it

may be linearizable if we restrict the allowed values v that can be supplied
as arguments to an inc operations. For each of the following sets V , show
that any execution in which all increments are elements of V is linearizable,
or show that there exists an execution with increments in V that is not.

1. V = {0, 1}.

2. V = {−1, 1}.

3. V = {1, 2}.
APPENDIX A. ASSIGNMENTS 351

Solution
1. The {0, 1} case is linearizable. Given an execution S of Algorithm A.4,
we assign to a linearization point to each inc operation at the step
where it writes to A, and assign a linearization point to each read
operation ρ that returns s at the later of the first step that leaves
P
j A[j] = s or the first step of ρ. Since this may assign the same
linearization point to some write operation π and one or more read
operations ρ1 , . . . , ρk , when this occurs, we order the write before the
reads and the reads arbitrarily.
Observe that:

(a) The value of each A[j] individually is non-decreasing over time,

and increases by at most one at each step.
Pn
(b) The same holds for j=1 A[j].

These are easily shown by induction on the steps of the execution, since
each inc operation only changes at most one A[j] and only changes it
by increasing it by 1.
The first condition implies that the value vj of A[j] used by a particular
read operation ρ lies somewhere between the minimum and maximum
values of A[j] during the operation’s interval, which implies the same
P
about the total j A[j]. In particular, if ρ returns s the value of
P
j A[j] is no greater than s, and it reaches s no later than the end of
ρ.
P
Because j A[j] increases by at most one per step, this means that
P P
either j A[j] = s at the first step of ρ, or j A[j] = s at some step
within the execution interval of ρ. In either case, ρ is assigned an
execution point within its interval that follows exactly s non-trivial
increments. This means that the return values of all read operations
are consistent with a sequential generalized counter execution, and
because both read and inc operations are ordered consistently with
the execution ordering in S, we have a linearization of S.

2. For increments in {−1, 1}, there are executions of Algorithm A.4 that
are not linearizable. We will construct a specific bad execution for
n = 3. Let p1 perform inc(1) and p2 perform inc(2), where p1 finishes
its operation before p2 starts. Because the inc(1) must be linearized
before the inc(−1), the values of the counter in any linearization will
be 0, 1, 0 in this order.
APPENDIX A. ASSIGNMENTS 352

Now add a read operation by p3 that is concurrent with both inc

operations. Suppose that in the execution, the follow operations are
performed on the registers A[1] through A[3]:

(a) p3 reads 0 from A[1].

(b) p1 writes 1 to A[1].
(c) p2 writes −1 to A[2].
(d) p3 reads −1 from A[2].
(e) p3 reads 0 from A[3].

Now p3 returns −1. There is no point in the sequential execution at

which this is the correct return value, so there is no linearization of
this execution.

3. For increments in {1, 2}, essentially the same counterexample works.

Here we let p1 do inc(1) and p2 do inc(2), while p3 again does a
concurrent read. The bad execution is:

(a) p3 reads 0 from A[1].

(b) p1 writes 1 to A[1].
(c) p2 writes 2 to A[2].
(d) p3 reads 2 from A[2].
(e) p3 reads 0 from A[3].

Now p3 returns 2, but in any linearization of the two write operations,

the values in the counter are 0, 1, 3.

A.4.2 Rock-paper-scissors
Define a rock-paper-scissors object as having three states 0 (rock), 1
(paper), and 2 (scissors), with a read operation that returns the current
state and a play(v) operation for v ∈ {0, 1, 2} that changes the state from s
to v if v = (s + 1) (mod 3) and has no effect otherwise.
Prove or disprove: There exists a deterministic wait-free linearizable
implementation of a rock-paper-scissors object from atomic registers.
APPENDIX A. ASSIGNMENTS 353

Solution
Proof: We will show how to implement a rock-paper-scissors object using
an unbounded max register, which can be built from atomic registers using
snapshots. The idea is to store a value v such that v mod 3 gives the value
of the rock-paper-scissors object. Pseudocode for both operations is given in
Algorithm A.5.

1 Let m be a shared max register.

2 procedure play(v)
3 s←m
4 if v = ((s + 1) mod 3) then
5 m←s+1

6 procedure read()
7 return (m mod 3)
Algorithm A.5: Implementation of a rock-paper-scissors object

Linearize each play operation that does not write m at the step at which
it reads m.
Linearize each play operation that writes s + 1 to m at the first step
at which m ≥ s + 1. If this produces ties, break first in order of increasing
s + 1 and then arbitrarily. Since each such operation has m ≤ s when the
operation starts and m ≥ s + 1 when it finishes, these linearization points fit
within the intervals of their operations.
Linearize each read() operation at the step where it reads m.
Since each of these linearization points is within the corresponding oper-
ation’s interval, this preserves the observed execution ordering.
Observe that the play operations that write are linearized in order of
increasing values written, and there are no gaps in this sequence because
no process writes s + 1 without first seeing s. (This actually shows there is
no to break ties by value.) So the sequence of values in the max register,
taken mod 3, iterates through the values 0, 1, 2, 0, . . . in sequence, with each
value equal mod 3 to some argument to a play operation. So we can take
these values mod 3 as the actual value of the register for the purposes of
read operations, meaning the read operations all return correct values. The
play operations that don’t write are linearized at a point where they would
have no effect on the state of the rock-paper-scissors object, which is also
consistent with the sequential specification.
It follows that Algorithm A.5 is a linearizable implementation of a rock-
APPENDIX A. ASSIGNMENTS 354

paper-scissors object from max registers. It is also wait-free, since each

operation is implemented using a constant number of max-register operations.
By implementing max registers using snapshots, we get a wait-free linearizable
implementation from atomic registers.

A.5 Assignment 5: due Wednesday, 2020-11-18, at

5:00pm Eastern US time
A.5.1 Randomized consensus with one max register
Prove or disprove: A single max register, with no other objects, is sufficient
to solve randomized wait-free binary consensus for two processes against an
oblivious adversary.

Solution
We’ll disprove it.
Let p0 and p1 be the two processes. The idea is to consider, for each
i ∈ {0, 1} some nonzero-probability solo terminating execution ξi of pi with
input i, then show that ξ0 and ξ1 can be interleaved to form a two-process
execution ξ that is indistinguishable by each pi from ξi .
The oblivious adversary will simply choose to schedule the processes for
ξ. Since the processes flip a finite number of coins in this execution, there
is a nonzero chance that the adversary gets lucky and they flip their coins
exactly the right way.
Fix ξ0 and ξ1 as above. Partition each ξi as αi βi1 βi2 . . . βiki where αi
contains only read operations and each βij starts with a write operation of a
value vij strictly larger than any previous write operation.
Let ξ = α0 α1 βi1 j1 βi2 j2 . . . βik jk where k = k0 + k1 and the blocks βi` j` are
the blocks {β0j } and {β1j } sorted in order of non-decreasing vij . Then each
block βi` j` in ξ starts with a write of a value no smaller than the previous
value in the max register, causing each read operation within the block to
return the value of this write, just as in the solo execution ξi` . Assuming
both processes flip their coins as in the solo executions, they both perform
the same operations and return the same values. These values will either
violate agreement in ξ or validity in at least one of ξ0 or ξ1 .
APPENDIX A. ASSIGNMENTS 355

A.5.2 A plurality object

Consider a shared-memory object with operations vote(v) and winner(),
where winner() returns the value v that appeared in the largest number of
previous vote operations, or ⊥ if there is no such unique v. For example, in
a sequential execution with votes a, b, b, c, c, c, a, a, a, the value returned by a
winner operation following each vote will be a, ⊥, b, b, ⊥, c, c, ⊥, a.
Pick one of these statements, and show that it is true:
1. There is a deterministic wait-free linearizable implementation of this
object for n processes that uses o(n) registers.
2. There is such an implementation that uses O(n) registers, but not o(n)
registers.
3. There is no such implementation using O(n) registers.

Solution
Case (2) holds.
To implement the object, use a snapshot array to hold the total votes
from each process, and have the winner operation take a snapshot, add up
all the votes and return the correct result. This can be done using n registers.
To show that it can’t be done with o(n) registers, use the JTT bound (see
Chapter 20). We need to argue that the object is perturbable. Let ΛΣπ be
an execution that needs to be perturbed, and let m be the maximum number
of vote(v) operations that start in Λ for any value v. Then a sequence γ of
m + 1 votes for some v 0 that does not appear in Λ will leave the object with v 0
as the plurality value, no matter how the remaining operations are linearized.
Since v 0 did not previously appear in Λ, this gives a different return value
for π in ΛγΣπ from ΛΣπ as required. The JTT bound now implies that any
implementation of the object requires at least n − 1 registers.

A.6 Presentation (for students taking CPSC 565):

due Tuesday, 2020-12-01, or Thursday, 2020-
12-03, in class; paper selection due Friday,
2020-11-20.
Students taking CPSC 565, the graduate version of the class, are expected to
give a 10-minute presentation on a recent paper in the theory of distributed
computing.
APPENDIX A. ASSIGNMENTS 356

The choice of paper to present should be made in consultation with the

instructor. To a first approximation, any paper from PODC, DISC, or a
similar conference in the last two or three years (that is not otherwise covered
in class) should work.
Please use the form at https://forms.gle/qafxrqiCPvYbQjQP7 to sub-
mit your proposed paper and preferred presentation time.
To get you started, the program for PODC 2020 can be found at https:
//www.podc.org/podc2020/program/. The web page for DISC 2020 can be
found at http://www.disc-conference.org/wp/disc2020/. Both include
links to papers published in the conference and to videos of conference
presentations.
Because of the limited presentation time, you are not required to get
into all of the technical details of the paper, but your presentation should
include:2

1. Title, authors, and date and venue of publication of the paper.

2. A high-level description of a central result. Unlike a talk for a general

audience, you can assume that your listeners know at least everything
that we’ve talked about so far in the class.

3. A description of where this result fits into the literature (e.g., solves
an open problem previously proposed in [...], improves on the previous
best running time for an algorithm from [...], gives a lower bound or
impossibility result for a problem previously proposed by [...], opens
up a new area of research for studying [...]), and why it is interesting
and/or hard.

4. A description (also possibly high-level) of the main technical mecha-

nism(s) used to get the result.

You do not have to prepare slides for your presentation if you would
prefer to use the whiteboard feature in Zoom, but you should make sure
to practice it in advance to make sure it fits in the allocated time. The
instructor will be happy to offer feedback on draft versions if available far
enough before the actual presentation date.
Relevant dates:
2
Literary theorists will recognize this as a three-act structure (preceded by a title card):
introduce the main character, make their life difficult, then resolve their problems in time
for the final curtain. This is not the only way to organize a talk, but if done right it has
the advantage of keeping the audience awake.
APPENDIX A. ASSIGNMENTS 357

2020-11-20 Paper selection due.

2020-11-27 Last date to send draft slides or arrange for a practice presen-
tation with the instructor if you want guaranteed feedback.

2020-12-01 and 2020-12-03 Presentations, during the usual class time.

Appendix B

Sample assignments from

Spring 2019

B.1 Assignment 1: due Wednesday, 2019-02-13, at

5:00pm
B.1.1 A message-passing bureaucracy
Alice and Bob are communicating with each other by alternately exchanging
messages. But Bob finds Alice’s messages alarming, and whenever he responds
to Alice, he also forwards a copy of Alice’s message to his good friend Charlie
1, a secret policeman. Charlie 1 reports to Charlie 2, but following the rule
that “once is happenstance, twice is coincidence, the third time it’s enemy
action,” [Fle59] Charlie 1 only sends a report to Charlie 2 after receiving
three messages from Bob. Similarly, Charlie 2 only sends a message to his
supervisor Charlie 3 after receiving three messages from Charlie 2, and so
on up until the ultimate boss Charlie n. Pseudocode for each participant is
given in Algorithm B.1.
Assuming we are in a standard asynchronous message-passing system,
that Alice sends her first message at time 0, and that the protocol finishes
as soon as Charlie n receives a message, what is the worst-case time and
message complexity of this protocol as a function of n?

Solution
Time complexity Observe that Alice sends at least k messages by time
2k − 2. This is easily shown by induction on k, because Alice sends at least
1 message by time 0, and if Alice has sent at least k − 1 message by time

358
APPENDIX B. SAMPLE ASSIGNMENTS FROM SPRING 2019 359

1 Alice:
2 initially do
3 send message to Bob
4 upon receiving message from Bob do
5 send message to Bob
6 Bob:
7 upon receiving message from Alice do
8 send message to Alice
9 send message to Charlie 1
10 Charlie i, for i < n:
11 initially do
12 c←0
13 upon receiving message from Bob or Charlie i − 1 do
14 c←c+1
15 if c = 3 then
16 c←0
17 send message to Charlie i + 1

Algorithm B.1: Reporting Alice’s alarming messages

APPENDIX B. SAMPLE ASSIGNMENTS FROM SPRING 2019 360

2k − 4, the last of these is received by Bob no later than time 2k − 3, and

Bob’s response is received by Alice no later than time 2k − 2.
Because each message from Alice prompts a message from Bob at most
one time unit later, this implies that Bob sends at least k messages by time
2k − 1.
Write T0 (k) = 2k − 1 for the maximum time for Bob to send k messages.
Write Ti (k) for the maximum time for Charlie i to send k messages, for
each 0 < i < n. In order for Charlie i to send k messages, it must receive
3k messages from Bob or Charlie i − 1 as appropriate. These messages are
sent no later than Ti−1 (3k), and the last of them is received no later than
Ti−1 (3k) + 1. So we have the recurrence

Ti (k) = Ti−1 (3k) + 1

T0 (k) = 2k − 1

with the exact solution

Ti (k) = (2 · 3i · k − 1) + k.

For i = n − 1 and k = 1, this is 2 · 3n−1 − 1 + n − 1 = 2 · 3n−1 + n = O(3n ).

We can get the exact time to finish by adding one more unit to account for
the delay in delivering the message from Charlie n − 1 to Charlie n. This
gives 2 · 3n−1 + n + 1 time exactly in the worst case, or O(3n ) if we want an
asymptotic bound.

Message complexity Message complexity is easier: there is no bound on

the number of messages that may be sent before Charlie n receives his first
message. This is because in an asynchronous system, Alice and Bob can send
an unbounded (though finite) number of messages to each other even before
Bob’s first message to Charlie 0 is delivered, without violating fairness.

B.1.2 Algorithms on rings

In Chapter 5, we saw several leader election algorithms for rings. But nobody
builds rings. However, it may be that an algorithm for a ring can be adapted
to other network structures.

1. Suppose you have a network in the form of a d-dimensional hypercube

Qd . This means we have n = 2d nodes, where each node is labeled by
a d-bit coordinate vector, and two nodes are adjacent if their vectors
APPENDIX B. SAMPLE ASSIGNMENTS FROM SPRING 2019 361

differ in exactly one coordinate. We also assume that each node knows
its own coordinate vector and those of its neighbors.
Show that any algorithm for an asynchronous ring can be adapted to
an asynchronous d-dimensional hypercube with no increase in its time
or message complexity.

2. What difficulties arise if we try to generalize this to an arbitrary graph

G?

Solution
1. The idea is to embed the ring in the hypercube, so that each node is
given a clockwise and counterclockwise neighbors, and any time the ring
algorithm asks to send a message clockwise or counterclockwise, we send
to the appropriate neighbor in the hypercube. We can then argue that
for any execution of the hypercube algorithm there is a corresponding
execution of the ring algorithm and vice versa; this implies that the
worst-case time and message-complexity in the hypercube is the same
as in the ring.
It remains only to construct an embedding. For d = 0, d = 1, and d = 2,
the ring and hypercube are the same graph, so it’s easy. For larger
d, split the hypercube into two subcubes Qd−1 , consisting of nodes
with coordinate vectors of the form 0x and 1x. Use the previously
constructed embedding for d − 1 to embed a ring on each subcube,
using the same embedding for both. Pick a pair of matching edges
(0x, 0y) and (1x, 1y) and remove them, replacing them with (0x, 1x)
and (0y, 1y). We have now constructed an undirected Hamiltonian
cycle on Qd . Orient the edges to get a directed cycle, and we’re done.

2. There are a several problems that may come up:

(a) Maybe G is not Hamiltonian.

(b) Even if G is Hamiltonian, finding an Hamiltonian cycle in an
arbitrary graph is NP-hard. This could be trouble for a practical
algorithm.
(c) Even if we can find a Hamiltonian cycle for G (maybe because G
is a nice graph of some kind, or maybe by taking advantage of
the unbounded computational power of processes assumed in the
standard message-passing model), the processes don’t necessarily
know what G looks like at the start. So they would need some
APPENDIX B. SAMPLE ASSIGNMENTS FROM SPRING 2019 362

initial start-up cost to map the graph, adding to the time and
message complexity of the ring algorithm.

B.1.3 Shutting down

Suppose we want to be able to stop a running protocol in an asynchronous
message-passing system prematurely. Define a shutdown mechanism as
a modification to an existing protocol in which any process can nondeter-
ministically issue a stop order that eventually causes all processes to stop
sending messages. We would like such a shutdown mechanism to satisfy two
properties:

1. Termination. If some process issues a stop order at time t, no process

sends a message at time t + ∆ or later, for some finite bound ∆ that
may depend on the structure of the network.

2. Non-interference. If no process issues a stop order, the protocol

carries out an execution identical to some execution of the underlying
protocol without a shutdown mechanism.

Show how to implement a shutdown mechanism, and prove tight upper

and lower bounds on ∆ as a function of the structure of the network.

Solution
This is pretty much the same as a Chandy-Lamport snapshot [CL85], as
described in §6.3. The main difference is that instead of recording its state
upon receiving a stop message, a process shuts down the underlying protocol.
Pseudocode is given in Algorithm B.2. We assume that the initial stop order
takes the form of a stop message delivered by a process to itself.

1 initially do
2 stopped ← false
3 upon receiving stop do
4 if ¬stopped then
5 stopped ← true
6 send stop to all neighbors
7 replace all events in underlying protocol with no-ops

Algorithm B.2: Shutdown mechanism based on Chandy-Lamport

APPENDIX B. SAMPLE ASSIGNMENTS FROM SPRING 2019 363

An easy induction argument shows that if p receives a stop message at

time t, then any process q at distance d from p receives a stop message no
later than time t + d. It may be that q sends stop messages in responds to
this stop message, but these are the last messages q ever sends. It follows
that no process sends a message later than time t + D, where D is the
diameter of the graph. This gives an upper bound on ∆.
For the lower bound, we can apply an indistinguishability argument. Let
p and q be processes at distance D from each other, and suppose that the
underlying protocol involves processes sending messages to their neighbors at
every opportunity. Consider two synchronous executions: X, an execution in
which no stop order is ever issued, and Xt , an execution in which p delivers
a stop message to itself at time t.
We can show by induction on d that any process r at distance d from p
carries out the same steps in both X and Xt up until time t + d − 1. The base
case is when d = 0, and we are simply restating that p runs the underlying
protocol before time t. For the induction step, we observe for any time
t0 < t + d − 1, any message sent to r from some neighbor s was sent at time
t0 − 1 < t + d − 2, and since d(p, s) ≥ d − 1, the induction hypothesis gives
that s sends the same messages at t0 − 1 in both X and Xt .
It follows that q sends the same message in X and Xt at time t + D − 1.
If it sends a message, then we have ∆ > D − 1. If it does not send a message,
then the mechanism violates the non-interference condition. So any correct
shutdown mechanism requires exactly ∆ = D time to finish in the worst
case.

B.2 Assignment 2: due Wednesday, 2019-03-06, at

5:00pm
B.2.1 A non-failure detector
Consider the following vaguely monarchist leader election mechanism for an
asynchronous message-passing system with crash failures. Each process has
access to an oracle that starts with the value 0 and may increase over time.
The oracle guarantees:

1. No two processes ever see the same nonzero value.

2. Eventually some non-faulty process is given a fixed value that is larger

than the values for all other processes for the rest of the execution.
APPENDIX B. SAMPLE ASSIGNMENTS FROM SPRING 2019 364

As a function of the number of processes n, what is the largest number of

crash failures f for which it is possible to solve consensus using this oracle?

Solution
We need f < n/2.
To show that f < n/2 is sufficient, observe that we can use the oracle to
construct an eventually strong (♦S) failure detector.
Recall that ♦S has the property that there is some non-faulty process
that is eventually never suspected, and every fault process is eventually
permanently suspected. Have each process broadcast the current value of its
leader oracle whenever it increases; when a process p receives i from some
process q, it stops suspecting q if i is greater than any value p has previously
seen, and starts suspecting all other processes. The guarantee that eventually
some non-faulty q gets a maximum value that never changes ensures that
eventually q is never suspected, and all other processes (including faulty
processes) are suspected. We can now use Algorithm 13.2 to solve consensus.
To show that f < n/2 is necessary, apply a partition argument. In
execution Ξ0 , processes n/2 + 1 through n crash, and processes 1 through
n/2 run with input 0 and with the oracle assigning value 1 to process 1 (and
no others). In execution Ξ1 , processes 1 through n/2 crashes, and processes
n/2 + 1 through n run with input 1 and with the oracle assigning value 2
to process n (and no others). In each of these executions, termination and
validity require that eventually the processes all decide on their respective
input values 0 and 1.
Now construct an execution Ξ2 , in which both groups of processes run
as in Ξ0 and Ξ1 , but no messages are exchanged between the groups until
after both have decided (which must occur after a finite prefix because this
execution is indistinguishable to the processes from Ξ0 or Ξ1 ). We now
violate agreement.

B.2.2 Ordered partial broadcast

Define ordered partial broadcast as a protocol that allows any process
to broadcast a message, with the guarantees that, for messages sent through
the broadcast mechanism:
1. Any message sent by a non-faulty process is received by at least one
process;
2. Any message that is received by at least one process is received by at
least k processes; and
APPENDIX B. SAMPLE ASSIGNMENTS FROM SPRING 2019 365

3. If two processes p and q both receive messages m1 and m2 from the

protocol, then either p and q both receive m1 before m2 , or they both
receive m2 before m1 .
Give an implementation of ordered partial broadcast with k = 3n/4 that
works for sufficiently large n in a fully-connected asynchronous message-
passing system with up to f = n/6 crash failures, or show that no such
implementation is possible.

Solution
No such implementation is possible. The proof is by showing that if some
such implementation could work, we could solve asynchronous consensus with
1 crash failure, contradicting the Fischer-Lynch-Patterson bound [FLP85]
(see Chapter 11).
An implementation of consensus based on totally-ordered partial broad-
cast for k = 3n/4 is given in Algorithm B.3. In fact, k = 3n/4 is overkill
when f = 1; k > n/2 + f is enough.

1 first ← ⊥
2 for i ← 1ton do
3 count[i] ← 0
4 value[i] ← ⊥
5 broadcast hi, inputi
6 upon receiving hj, vi do
7 if first = ⊥ then
8 first ← hj, vi
9 send received(hj, vi) to all processes

10 upon receiving received(hj, vi) do

11 count[j] ← count[j] + 1
12 value[j] ← v
13 if count[j] = k − f then
14 decide value[j]

Algorithm B.3: Consensus from totally-ordered partial broadcast.

Code for process i.

The idea of the algorithm is to use the broadcast mechanism to chose a

decision value, by looking at which value is delivered first. Since not every
process will see the same value delivered first, this requires a second round
APPENDIX B. SAMPLE ASSIGNMENTS FROM SPRING 2019 366

of communication in which processes retransmit their first incoming message.

The following lemma shows that this is enough to get agreement:

Lemma B.2.1. In any execution of Algorithm B.3 with k > n/2 + f . there
is a unique pair hj, vi such that at least k − f non-faulty processes resend
received(hj, vi).

Proof. Because all processes that receive messages m1 and m2 through the
broadcast mechanism receive them in the same order, we can define a partial
order on messages by letting m1 < m2 if any process receives m1 before m2 .
There are only finitely many messages, so there is at least one pair hj, vi
that is minimal in this partial order. This message is received by at least k
processes, of which at least k − f are non-faulty. Each such process receives
hj, vi before any other broadcast messages, so it sets first to hj, vi and resends
received(hj, vi).
To show that hj, vi is unique, observe that k − f > n/2 implies that if
there is some other pair hj 0 , v 0 i that is resent by k − f non-faulty processes,
then there is some process that resends both hj, vi and hj 0 , v 0 i. But each
process resends at most one pair.

Lemma B.2.1 immediately gives agreement, because a process only decides

on a value v after receiving received(hj, vi) from k − f processes, and only
one such pair is sent by so many. Termination follows from the existence of
such a pair: eventually every non-faulty process receives hj, vi from k − f
processes. Validity is immediate from the fact that v is j’s input.
It follows that Algorithm B.3 solves consensus whenever k > n/2 + f ,
which includes the case k = 3n/4 and f = n/6. If an implementation
of ordered partial broadcast with these parameters exists in the standard
message-passing model, this would give a protocol for asynchronous consensus
with f = n/6 ≥ 1 when n ≥ 6. This contradicts FLP, showing that such an
implementation is impossible.

B.2.3 Mutual exclusion using a counter

Algorithm B.4 gives a modified version of Peterson’s two-process mutual
exclusion algorithm (§17.4.1) that replaces the present bits with an atomic
counter count. This object supports read, increment, and decrement opera-
tions, where increment and decrement increase and decrease the value in the
counter by one, respectively. Unlike the present array, count doesn’t depend
on the number of processes n. So in principle this algorithm might work for
arbitrary n.
APPENDIX B. SAMPLE ASSIGNMENTS FROM SPRING 2019 367

shared data:
1 waiting, atomic register, initially arbitrary
2 count, atomic counter, initially 0
3 Code for process i:
4 while true do
// trying
5 increment count
6 waiting ← i
7 while true do
8 if count = 1 then
9 break
10 if waiting = i + 1 (mod n) then
11 break

// critical
12 (do critical section stuff)
// exiting
13 decrement count
// remainder
14 (do remainder stuff)
Algorithm B.4: Peterson’s mutual exclusion algorithm using a
counter
APPENDIX B. SAMPLE ASSIGNMENTS FROM SPRING 2019 368

Show that Algorithm B.4 provides starvation-free mutual exclusion for

two processes, but not for three processes.

Solution
The proof that this works for two processes is essentially the same as in the
original algorithm. The easiest way to see this is to observe that process
pi sees count = 1 in Line 8 under exactly the same circumstances as it sees
present[¬i] = 0 in Line 8 in the original algorithm; and similarly with two
processes waiting is always set to the same value as waiting in the original
algorithm. So we can map any execution of Algorithm B.4 for two processes
to an execution of Algorithm 17.4, and all of the properties of the original
algorithm carry over to the modified version.
To show that the algorithm doesn’t work for three processes, we construct
an explicit bad execution:

1. p0 increments count

2. p1 increments count

3. p2 increments count

4. p0 writes 0 to waiting

5. p1 writes 1 to waiting

6. p2 writes 2 to waiting

7. p0 reads 3 from count

8. p1 reads 3 from count

9. p2 reads 3 from count

10. p1 reads 2 from waiting and enters the critical section.

11. p1 leaves the critical section and decrements count.

At this point we have count = 2 and waiting = 2, with both p0 and p2 at

the start of the loop to check these variables. Suppose that p1 doesn’t come
back. Because neither p0 nor p2 changes count or waiting, both variables
remain at 2 forever. But then neither p0 nor p2 enters the critical section,
because count is never 1 and waiting is never equal to 0 + 1 or 2 + 1 (mod 3).
APPENDIX B. SAMPLE ASSIGNMENTS FROM SPRING 2019 369

B.3 Assignment 3: due Wednesday, 2019-04-17, at

5:00pm
B.3.1 Zero, one, many
Consider a counter supporting inc and read operations that is capped at 2.
This means that after the first two calls to inc, any further calls to inc have
no effect: a read operation will return 0 if it follows no calls to inc, 1 if it
follows exactly one call to inc, and 2 if it follows two or more calls to inc.
There is a straightforward implementation of this object using snapshot.
This requires O(n) space and O(n) steps per operation in the worst case.
Is it possible to do better? That is, can one give a deterministic, wait-free,
linearizable implementation of a 2-bounded counter from atomic registers
that uses o(n) space and o(n) steps per operation in the worst case?

Solution
One possible implementation is given in Algorithm B.5. This requires O(1)
space and O(1) steps per call to inc or read.

1 procedure inc
2 if c[1] = 1 then
// somebody already did inc
3 c[2] ← 1
4 else
5 c[1] ← 1
// maybe somebody else is doing inc
6 if splitter returns right or down then
7 c[2] ← 1

8 procedure read
9 if c[2] = 1 then
10 return 2
11 else if c[1] = 1 do
12 return 1
13 else
14 return 0

Algorithm B.5: A 2-bounded counter

APPENDIX B. SAMPLE ASSIGNMENTS FROM SPRING 2019 370

The implementation uses two registers c[1] and c[2] to represent the value
of the counter. Two additional registers implement a splitter object as in
Algorithm 17.5.1
Claim: For any two calls to inc, at least one sets c[2] to 1. Proof: Suppose
otherwise. Then both calls are by different processes p and q (or else the
second call would see c[1] = 1) and both execute the splitter. Since a splitter
returns stop to at most one process, one of the two processes gets right or
down, and sets c[2].
It is also straightforward to show that a single inc running alone will set
c[1] but not c[2], since in this case the splitter will return stop.
Now we need to argue linearizability. We will do so by assigning lineariza-
tion points to each operation.
If some inc does not set c[2], assign it the step at which it sets c[1].
Assign each other inc the step at which it first sets c[2].
If every inc sets c[2], assign the first inc to set c[1] the step at which it
does so, and assign all others the first point during its execution interval at
which c[2] is nonzero.
For a read operation that returns 2, assign the step at which it reads c[2].
For a read operation that returns 1, assign the first point in the execution
interval after it reads c[2] at which c[1] = 1. For a read operation that
returns 0, assign the step at which it reads c[0].
This will assign the same linearization point to some operations; in this
case, put incs before reads and otherwise break ties arbitrarily.
These choices create a linearization which consists of (a) a sequence of
read operations that return 0, all of which are assigned linearization points
before the first step at which c[1] = 1; (b) the first inc operation that
sets c[1]; (c) a sequence of read operations that return 1, all of which are
linearized after c[1] = 1 but before c[2] = 1; (c) some inc that is either
the first to set c[2] or spans the step that sets c[2]; and (d) additional inc
operations together with read operations that all return 2. Since each read
returns the minimum of 2 and the number of incs that precede it, this is a
correct linearization.

B.3.2 A very slow counter

Consider a slow counter object with operations inc and read, where the
value v of a counter starts at 0 and increases by 1 as the result of each call
1
It may be possible shave off a register by breaking the splitter abstraction and using
the race or door register in place of c[1], but I haven’t worked out the linearizability for
this case.
APPENDIX B. SAMPLE ASSIGNMENTS FROM SPRING 2019 371

to inc, but read returns log∗ v instead of v.

Suppose we want a deterministic, wait-free, linearizable implementation
of a slow counter as defined above from atomic registers. Give tight bounds
on the worst-case step complexity of operations on such an implementation.

Solution
The worst-case step complexity of an operation is Θ(n).
For the upper bound, implement a counter on top of snapshots (or just
collect), and have read compute log∗ of whatever value is read.
For the lower bound, observe that a slow counter has the perturbability
property needed for the JTT proof. Given an execution of the form Λk Σk π
as described in Chapter 20, we can always insert some sequence of inc
operations between Λk and Σk that will change the return value of π. The
number of incs needed will be the number needed to raise log∗ v, plus an
extra n to overcome the possibility of pending incs in Σk being linearized
before or after π. Since this object is perturbable, and the atomic registers
we are implementing it from are historyless, JTT applies and gives an Ω(n)
lower bound on the cost of read in the worst case.

B.3.3 Double-entry bookkeeping

Consider an object that implements an unbounded collection of accounts
A1 , A2 , . . . , each of which holds an integer value, and that provides three
operations:

• The operation read(i) returns the current value of Ai .

• The operation transfer(i, j, n) moves n units from Ai to Aj ; if A0i and

A0j are the new values, then A0i = Ai − n and A0j = Aj + n.

• The operation close(i, j) sets Ai to zero and adds the previous value
to Aj . It is equivalent to atomically executing transfer(i, j, read(i)).

1. What is the consensus number of this object?

2. What is the consensus number of a restricted version of this object

that provides only the read and transfer operations?
APPENDIX B. SAMPLE ASSIGNMENTS FROM SPRING 2019 372

Solution
1. The consensus number of the object is infinite. Initialize A0 to 1 and
the remaining Ai to 0. We can solve id consensus by having process
i (where i > 0 execute close(0, i) and then applying read to scan all
the Aj values for itself and other processes. Whichever process gets
the 1 wins.

2. The consensus number without close is 1. Proof: Observe that

transfer operations commute.

B.4 Presentation (for students taking CPSC 565):

due Monday, 2019-04-22, or Wednesday, 2019-
04-24, in class
Students taking CPSC 565, the graduate version of the class, are expected to
give a 10-minute presentation on a recent paper in the theory of distributed
computing.
The choice of paper to present should be made in consultation with the
instructor. To a first approximation, any paper from PODC, DISC, or a
similar conference in the last two or three years (that is not otherwise covered
in class) should work.
To get you started, the 2018 PODC proceedings can be found at https://
dl.acm.org/citation.cfm?id=3212734&picked=prox, and the 2018 DISC
proceedings can be found at http://drops.dagstuhl.de/opus/portals/
lipics/index.php?semnr=16090.
Because of the limited presentation time, you are not required to get
into all of the technical details of the paper, but your presentation should
include:2

1. Title, authors, and date and venue of publication of the paper.

2. A high-level description of a central result. Unlike a talk for a general

audience, you can assume that your listeners know at least everything
that we’ve talked about so far in the class.
2
Literary theorists will recognize this as a three-act structure (preceded by a title card):
introduce the main character, make their life difficult, then resolve their problems in time
for the final curtain. This is not the only way to organize a talk, but if done right it has
the advantage of keeping the audience awake.
APPENDIX B. SAMPLE ASSIGNMENTS FROM SPRING 2019 373

3. A description of where this result fits into the literature (e.g., solves
an open problem previously proposed in [...], improves on the previous
best running time for an algorithm from [...], gives a lower bound or
impossibility result for a problem previously proposed by [...], opens
up a new area of research for studying [...]), and why it is interesting
and/or hard.

4. A description (also possibly high-level) of the main technical mecha-

nism(s) used to get the result.
You do not have to prepare slides for your presentation if you would
prefer to use the blackboard, but you should make sure to practice it in
advance to make sure it fits in the allocated time. The instructor will be
happy to offer feedback on draft versions if available far enough before the
actual presentation date.
Relevant dates:
2019-04-10 Paper selection due.

2019-04-17 Last date to send draft slides or arrange for a practice presen-
tation with the instructor if you want guaranteed feedback.

2019-04-22 and 2019-04-24 Presentations, during the usual class time.

B.5 CS465/CS565 Final Exam, May 7th, 2019

Write your answers in the blue book(s). Justify your answers. Work alone.
Do not use any notes or books.
There are four problems on this exam, each worth 20 points, for a total
of 80 points. You have approximately three hours to complete this exam.

B.5.1 A roster (20 points)

A roster object has operations announce and read, where read returns a
list of the identities of all processes that have previously called announce at
least once.
Suppose you want a wait-free, linearizable implementation of this object
from multiwriter atomic registers. As a function of the number of processes
n, give the best upper and lower bound you can on the number of registers
you will need.
You may assume that the set of process identities is fixed for each n and
that each process knows its own identity.
APPENDIX B. SAMPLE ASSIGNMENTS FROM SPRING 2019 374

Solution
You will need exactly n registers (Θ(n) is also an acceptable answer).
For the upper bound, have each process write its id to its own register,
and use a double-collect snapshot to read all of them. This uses exactly n
registers. The double-collect snapshot is wait-free because after each process
has called announce once, the contents of the registers never change, so read
finishes after O(n) collects or O(n2 ) register reads. It’s linearizable because
double-collect snapshot returns the exact contents of the registers at some
point during its execution.3
For the lower bound, use a covering argument.4
Have the processes p1 , . . . , pn run announce in order, stopping each
process when it covers a new register. This will give sequence of partial
executions Ξi where at the end of Ξi , there is a set of i registers r1 . . . ri that
are covered by p1 . . . pi , and no other operations are in progress.
To show this works, we need to argue that each pi+1 does in fact cover a
register ri+1 6∈ {r1 , . . . , ri }. If not, then we can extend Ξi by running pi+1 ’s
announce operation to completion, then delivering all the covering writes
by p1 . . . pi . Now any subsequent read will fail to return pi+1 , violating the
specification. (If we have a spare process, we can have it do the bad read;
otherwise we can run p1 to completion and let it do it.)
At the end of Ξn , we have covered n distinct registers, proving the lower
bound.

B.5.2 Self-stabilizing consensus (20 points)

Consider a model with n processes p0 , . . . , pn−1 organized in a ring, where
pi can directly observe both its own state and that of p(i−1) mod n . Suppose
that the state xi of each process pi is always a natural number.
3
If we only do a single collect, the implementation is not linearizable. An example of a
bad execution is one where a reader reads r1 , then p1 writes to r1 (starting and finishing
its announce operation), then p2 writes to r2 (starting an finishing its announce operation),
and finally the reader reads r2 . In this execution the reader will return {p2 } only, which is
inconsistent with the observed ordering that puts announce(p1 ) before announce(p2 ).
4
It’s tempting to use JTT [JTT00] here, but the roster object is not perturbable. Once
all ids of p1 through pn−1 have been registered, subsequent announce operations by p1
through pn−1 have no effect.
The subtlety here is that in the JTT argument, we probably won’t choose γ when
perturbing Λk Σk to include more than one new announce, but replacing γ by γ 0 to hit the
first possible uncovered register in π might involve an arbitrary sequence of operations by
p1 through pn−1 , including a sequence where all of them call announce at least once. Once
this happens, we get a Λk+1 Σk+1 execution that can no longer be perturbed.
APPENDIX B. SAMPLE ASSIGNMENTS FROM SPRING 2019 375

At each step, an adversary chooses a process pi to run. This process then

updates its own state based on its previous state xi and the state x(i−1) mod n
of its counterclockwise neighbor p(i−1) mod n . The adversary is required to
run every process infinitely often but is not otherwise constrained.
Is there a protocol for the processes that guarantees that, starting from
an arbitrary initial configuration, they eventually reach a configuration where
(a) all processes have the same state x ∈ N; and (b) no process ever changes
its state as the result of taking additional steps?
Give such a protocol and prove that it works, or show that no such
protocol is possible.

Solution
It turns out that this problem is a good example of what happens if you
don’t remember to include some sort of validity condition. As pointed in
several student solutions, having each process pick a fixed constant xi the
first time it updates works.
Here is a protocol that also works, and satisfies the validity condition that
the common output was some process’s input (which was not required in the
problem statement). When pi takes a step, it sets xi to max(xi , x(i−1) mod n ).
To show that this works, we argue by induction that the maximum value
eventually propagates to all processes. Let x = xi be the initial maximum
value. The induction hypothesis is that for each j ∈ {0, . . . , n − 1}, eventually
all processes in the range i through i + j (mod n) hold value x forever.
Suppose that the hypothesis holds for j; to show that it holds for j + 1,
start in a configuration where xi through xi+j are all x. No transition can
change any of these values, because taking the max of x and any other value
yields x. Because each process is scheduled infinitely often, eventually pi+j+1
takes a step; when this happens, xi+j+1 is set to max(x, xi+j+1 ) = x.
Since the hypothesis holds for all j ∈ {0, . . . , n − 1}, it holds for j = n−1;
but this just says that eventually all n processes hold x forever.

B.5.3 All-or-nothing intermittent faults (20 points)

Recall that in the standard synchronous message-passing model with crash
failures, a faulty process runs correctly up until the round in which it crashes,
during which it sends out some subset of the correct messages, and after
which it sends out no messages at all.
Suppose instead we have intermittent faults, where any process may fail
to send outgoing messages in a particular round, but these are all-or-nothing
APPENDIX B. SAMPLE ASSIGNMENTS FROM SPRING 2019 376

faults in the sense that a process either sends all of its messages in a given
round or no messages in that round. To avoid shutting down a protocol
completely, we require that in every round, there is at least one process that
sends all of its messages. We also allow a process to send a message to itself.
If we wish to solve agreement (that is, get agreement, termination, and
validity) in this model, what is the minimum number of rounds we need in
the worst case?

Solution
We need one round. Every process transmits its input to all processes,
including itself. From the all-or-nothing property, all processes receive the
same set of messages. From the assumption that some process is not faulty
in this round, this set is nonempty. So the processes can reach agreement by
applying any consistent rule to choose an input from the set.

B.5.4 A tamper-proof register (20 points)

Consider a tamper-proof register, which is a modified version of a standard
multiwriter atomic register for which the read operation returns ⊥ if no write
operation has occurred, v if exactly one write(v) operation has occurred,
and fail if two or more write operations have occurred.
What is the consensus number of this object?

Solution
The consensus number is 1.
Proof: We can implement it from atomic snapshot, which can be imple-
mented from atomic registers, which have consensus number 1.
For my first write(v) operation, write v to my component of the snapshot;
for subsequent write(v) operations, write fail. For a read operation, take
a snapshot and return (a) ⊥ if all components are empty; (b) v if exactly
one component is non-empty and has value v; and (c) fail if more than one
component is non-empty or any component contains fail.
Appendix C

Sample assignments from

Spring 2016

C.1 Assignment 1: due Wednesday, 2016-02-17, at

5:00pm
Bureaucratic part
Send me email! My address is [email protected].
In your message, include:

1. Your name.

2. Your status: whether you are an undergraduate, grad student, auditor,

etc.

3. Whether you are taking the course as CPSC 465 or CPSC 565.

4. Anything else you’d like to say.

(You will not be graded on the bureaucratic part, but you should do it
anyway.)

C.1.1 Sharing the wealth

A kindergarten consists of n children in a ring, numbered 0 through n − 1,
with all arithmetic on positions taken mod n.
In the initial configuration, child 0 possesses n cookies. The children take
steps asynchronously, and whenever child i takes a step in a configuration

377
APPENDIX C. SAMPLE ASSIGNMENTS FROM SPRING 2016 378

where they have a cookie but child i + 1 does not, child i gives one cookie to
child i + 1. If child i + 1 already has a cookie, or child i has none, nothing
happens. We assume that a fairness condition guarantees that even though
some children are fast, and some are slow, each of them takes a step infinitely
often.

1. Show that after some finite number of steps, every child has exactly
one cookie.

2. Suppose that we define a measure of time in the usual way by assigning

each step the largest possible time consistent with the assumption that
that no child ever waits more than one time unit to take a step. Show
the best asymptotic upper bound you can, as a function of n, on the
time until every child has one cookie.

3. Show the best asymptotic lower bound you can, as a function of n, on

the worst-case time until every child has one cookie.

Solution
1. First observe that in any configuration reachable from the initial con-
figuration, child 0 has k cookies, n − k of the remaining children have
one cookie each, and the rest have zero cookies. Proof: Suppose we
are in a configuration with this property, and consider some possible
step that changes the configuration. Let i be the child that takes the
step. If i = 0, then child i goes from k to k − 1 cookies, and child 1
goes from 0 to 1 cookies, increasing the number of children with one
cookie to n − k + 1. If i > 0, then child i goes from 1 to 0 cookies and
child i + 1 from 0 to 1 cookies, with k unchanged. In either case, the
invariant is preserved.
Now let us show that k must eventually drop as long as some cookie-less
child remains. Let i be the smallest index such that the i-th child has
no cookie. Then after finitely many steps, child i − 1 takes a step and
gives child i a cookie. If i − 1 = 0, k drops. If i − 1 > 0, then the
leftmost 0 moves one place to the left. It can do so only finitely many
times until i = 1 and k drops the next time child 0 takes a step. It
follows that after finitely many steps, k = 1, and by the invariant all
n − 1 remaining children also have one cookie each.

2. Number the cookies 0 through n − 1. When child 0 takes a step, have it

give the largest-numbered cookie it still possesses to child 1. For each
APPENDIX C. SAMPLE ASSIGNMENTS FROM SPRING 2016 379

cookie i, let xti be the position of the i-th cookie after t asynchronous
rounds, where an asynchronous round is the shortest interval in which
each child takes at least one step.
Observe that no child j > 0 ever gets more than one cookie, since no
step adds a cookie to a child that already has one. It follows that cookie
0 never moves, because if child 0 has one cookie, so does everybody
else (including child 1). We can thus ignore the fact that the children
are in a cycle and treat them as being in a line 0 . . . n − 1.
We will show by induction on t that, for all i and t, xti ≥ yit =
max(0, min(i, zit )) where zit = t + 2(i − n + 1).
Proof: The base case is when t = 0. Here xti = 0 for all i. We also have
zit = 2(i − n + 1) ≤ 0 so yit = max(0, min(i, zit )) = max(0, zit ) = 0. So
the induction hypothesis holds with xti = yit = 0.
Now suppose that the induction hypothesis holds for t. For each i,
there are several cases to consider:

(a) xti = xti+1 = 0. In this case cookie i will not move, because it’s not
at the top of child 0’s stack. But from the induction hypothesis
we have that xti+1 = 0 implies zi+1 t = t + 2(i + 1 − n + 1) ≤ 0,
which gives zi = zi+1 − 2 ≤ −2. So zit+1 ≤ zi+1
t t t + 1 ≤ −1 and
yi = 0, and the induction hypothesis holds for xt+1
t+1
i .
(b) xti = i. Then even if cookie i doesn’t move (and it doesn’t), we
have xt+1
i ≥ xti ≥ min(i, zit ).
(c) xti < i and xti+1 = xti + 1. Again, even if cookie i doesn’t move, we
still have xt+1
i ≥ xti = xti+1 −1 ≥ yi+1
t −1 ≥ t+2(i+1−n+1)−1 =

t + 2(i − n + 1) + 1 > yi .t

(d) xti < i and xti+1 > xti + 1. Nothing is blocking cookie i, so it moves:
xit+1 = xti + 1 ≥ t + 2(i − n + 1) + 1 = (t + 1) + 2(i − n + 1) = yit+1 .

It follows that our induction hypothesis holds for all t. In particular,

at t = 2n − 2 we have zit = 2n − 2 + 2(i − n + 1) = 2i − 1 ≥ i for all
i > 0. So at time 2n − 2, xti ≥ yit = i for all i and every child has one
cookie. This gives an asymptotic upper bound of O(n).

3. There is an easy lower bound of n − 1 time. Suppose we run the

processes in round-robin order, i.e., the i-th step is taken by process
i mod n. Then one time unit goes by for every n steps, during which
each process takes exactly one step. Since process 0 reduces its count
APPENDIX C. SAMPLE ASSIGNMENTS FROM SPRING 2016 380

by at most 1 per step, it takes at least n − 1 time to get it to 1. This

gives an asymptotic lower bound of Ω(n), which is tight.
I believe it should be possible to show an exact lower bound of 2n − 2
time by considering a schedule that runs in reverse round-robin order
n − 1, n − 2, . . . , 0, n − 1, n − 2, . . . , but this is more work and gets the
same bound up to constants.

C.1.2 Eccentricity
Given a graph G = (V, E), the eccentricity (v) of a vertex v is the
maximum distance maxv0 d(v, v 0 ) from v to any vertex in the graph.
Suppose that you have an anonymous1 asynchronous message-passing
system with no failures whose network forms a tree.

1. Give an algorithm that allows each node in the network to compute its
eccentricity.

2. Safety: Prove using an invariant that any value computed by a node

using your algorithm is in fact equal to its eccentricity. (You should
probably have an explicit invariant for this part.)

3. Liveness: Show that every node eventually computes its eccentricity in

your algorithm, and that the worst-case message complexity and time
complexity are both within a constant factor of optimal for sufficiently
large networks.

Solution
1. Pseudocode is given in Algorithm C.1. For each edge vu, the algorithm
sends a message d from v to u, where d is the maximum length of
any simple path starting with uv. This can be computed as soon as v
knows the maximum distances from all of its other neighbors u0 6= u.

2. We now show correctness of the values computed by Algorithm C.1.

Let dv [u] be the value of d[u] at v. Let `v [u] be the maximum length
of any simple path starting with the edge vu. To show that the
1
Clarification added 2016-02-13: Anonymous means that processes don’t have global
ids, but they can still tell their neighbors apart. If you want to think of this formally,
imagine that each process has a local identifier for each of its neighbors: a process with
three neighbors might number them 1, 2, 3 and when it receives or sends a message one of
these identifiers is attached. But the local identifiers are arbitrary, and what I call you has
no relation to what you call me or where either of us is positioned in the network.
APPENDIX C. SAMPLE ASSIGNMENTS FROM SPRING 2016 381

local data: d[u] for each neighbor u, initially ⊥

notified[u] for each neighbor u, initially false

1 initially do
2 notify ()
3 upon receiving d from u do
4 d[u] ← d
5 notify ()
6 procedure notify ()
7 foreach neighbor u do
8 if ¬notified[u] and d[u0 ] 6= ⊥ for all u0 6= u then
9 Send 1 + maxu0 6=u d[u0 ] to u
10 notified[u] ← true

11 if notified[u] = true for all neighbors u then

12  ← maxu d[u]

Algorithm C.1: Computing eccentricity in a tree

algorithm computes the correct values, we will prove the invariant that
dv [u] ∈ {⊥, `v [u]} always, and for any message d in transit from u to v,
d = `v [u].
In the initial configuration, dv [u] = ⊥ for all v and u, and there are no
messages in transit. So the invariant holds.
Now let us show that calling notify at some process v preserves the
invariant. Because notify() does not change dv , we need only show
that the messages it sends contain the correct distances.
Suppose notify() causes v to send a message d to u. Then d = 1 +
maxu0 6=u dv [u0 ] = 1+maxu0 6=u `v [u0 ], because dv [u0 ] 6= ⊥ for all neighbors
u0 6= u by the condition on the if statement and thus dv [u0 ] = `v [u0 ] for
all u0 6= u by the invariant.
So the invariant will continue to hold in this case provided `u [v] =
1 + maxu0 6=u `v [u0 ]. The longest simple path starting with uv either
consists of uv alone, or is of the form uvw . . . for some neighbor w of v
with w 6= u. In the former case, v has no other neighbors u0 , in which
case d = 1 + maxu0 6=u `v [u0 ] = 1 + 0 = 1, the correct answer. In the
latter case, d = 1 + maxu0 6=u `v [u0 ] = 1 + `v [w], again the length of the
longest path starting with uv.
APPENDIX C. SAMPLE ASSIGNMENTS FROM SPRING 2016 382

This shows that notify preserves the invariant. We must also show
that assigning dv [u] ← d upon receiving d from u does so. But in this
case we know from the invariant that d = `v [u], so assigning this value
to dv [u] leaves dv [u] ∈ {⊥, `v [u]} as required.

3. First let’s observe that at most one message is sent in each direction
across each edge, for a total of 2|E| = 2(n − 1) messages. This is
optimal, because if in some execution we do not send a message across
some edge uv, then we can replace the subtree rooted at u with an
arbitrarily deep path, and obtain an execution indistinguishable to v
in which its eccentricity is different from whatever it computed.
For time complexity (and completion!) we’ll argue by induction on
`v [u] that we send a message across uv by time `v [u] − 1.
If `v [u] = 1, then u is a leaf; as soon as notify is called in its initial
computation event (which we take as occurring at time 0), u notices it
has no neighbors other than v and sends a message to v.
If `v [u] > 1, then since `v [u] = 1 + maxv0 6=v `u [v 0 ], we have `u [v]0 ≤
`v [u]−1 for all neighbors v 0 6= v of u, which by the induction hypothesis
means that each such neighbor v 0 sends a message to u no later than
time `v [u] − 2. These messages all arrive at u no later than time
`v [u] − 1; when the last one is delivered, u sends a message to v.
It follows that the last time a message is sent is no later than time
maxuv (`v [u] − 1), and so the last delivery event occurs no later than
time maxuv `v [u]. This is just the diameter D of the tree, giving a
worst-case time complexity of exactly D.
To show that this is optimal, consider an execution of some hypothetical
algorithm that terminates by time D − 1 in the worst case. Let u and
v be nodes such that d(u, v) = D. Then there is an execution of this
algorithm in no chain of messages passes from u to v, meaning that
no event of u is causally related to any event of v. So we can replace
u with a pair uw of adjacent nodes with d(w, v) = d(u, v) + 1, which
changes (v) but leaves an execution that is indistinguishable to v
from the original. It follows that v returns an incorrect value in some
executions, and this hypothetical algorithm is not correct. So time
complexity D is the best possible in the worst case.
APPENDIX C. SAMPLE ASSIGNMENTS FROM SPRING 2016 383

C.1.3 Leader election on an augmented ring

Suppose that we have an asynchronous ring where each process has a distinct
identity, but the processes do not know the size n of the ring. Suppose also
that each process i can send messages not only to its immediate neighbors,
but also to the processes at positions at positions i − 3 and i + 3 (mod n) in
the ring.
Show that Θ(n log n) messages are both necessary and sufficient in the
worst case to elect a unique leader in this system.

Solution
For sufficiency, ignore the extra edges and use Hirschberg-Sinclair [HS80]
(see §5.2.2).
For necessity, we’ll show that an algorithm that solves leader election in
this system using at most T (n) messages can be modified to solve leader
election in a standard ring without the extra edges using at most 3T (n)
messages. The idea is that whenever a process i attempts to send to i + 3,
we replace the message with a sequence of three messages relayed from i
to i + 1, i + 2, and then i + 3, and similarly for messages sent in the other
direction. Otherwise the original algorithm is unmodified. Because both
systems are asynchronous, any admissible execution in the simulated system
has a corresponding admissible execution in the simulating system (replace
each delivery event by three delivery events in a row for the relay messages)
and vice versa (remove the initial two relay delivery events for each message
and replace the third delivery event with a direct delivery event). So in
particular if there exists an execution in the simulating system that requires
Ω(n log n) messages, then there is a corresponding execution in the simulated
system that requires at least Ω(n log n/3) = Ω(n log n) messages as well.

C.2 Assignment 2: due Wednesday, 2016-03-09, at

5:00pm
C.2.1 A rotor array
Suppose that you are given an object that acts as described in Algorithm C.2.
A write operation on this object writes to location A[r] and increments r
mod n. A read operation by process i (where i ∈ {0 . . . n − 1}) returns A[i].
Initially, r = 0 and A[i] = ⊥ for all i.
What is the consensus number of this object?
APPENDIX C. SAMPLE ASSIGNMENTS FROM SPRING 2016 384

1 procedure write(A, v)
2 atomically do
3 A[r] ← v; r ← (r + 1) mod n

4 procedure read(A)
5 return A[i]
Algorithm C.2: Rotor array: code for process i

Solution
First let’s show that it is at least 2, by exhibiting an algorithm that uses
a single rotor array plus two atomic registers to solve 2-process wait-free
consensus.

1 procedure consensus(v)
2 input[i] ← v
3 write(A, i)
4 i0 ← read(A)
5 if i0 = i then
// Process 0 wrote first
6 return input[0]
7 else
// Process 1 wrote first
8 return input[1]

Algorithm C.3: Two-process consensus using a rotor array

The algorithm is given as Algorithm C.3. Each process i first writes its
input value to a single-writer register input[i]. The process then writes its id
to the rotor array. There are two cases:
1. If process 0 writes first, then process 0 reads 0 and process 1 reads
1. Thus both processes see i0 = i and return input[0], which gives
agreement, and validity because input[0] is then equal to 0’s input.
2. If process 1 writes first, then process 0 reads 1 and process 1 reads
either 0 (if 0 wrote quickly enough) or ⊥ (if it didn’t). In either case,
both processes see i0 6= i and return input[1].
Now let us show that a rotor array can’t be used to solve wait-free
consensus with three processes. We will do the usual bivalence argument,
APPENDIX C. SAMPLE ASSIGNMENTS FROM SPRING 2016 385

and concentrate on some bivalent configuration C and pair of operations π0

and π1 such that Cπi is i-valent for each i.
If π0 and π1 are operations on different objects or operations on an atomic
register, then they either commute or the usual analysis for atomic registers
gives a contradiction. So the interesting case is when π0 and π1 are both
operations on a single rotor array object A.
If either operation is a read, then only the process that carries out the
read knows whether it occurred. The same argument as for atomic registers
applies in this case. So the only remaining case is when both operations are
writes.
Consider the configurations Cπ0 π1 (which is 0-valent) and Cπ1 π0 (which
is 1-valent). These differ in that there are two locations j and (j + 1) mod n
(which we will just write as j + 1) that contain values v0 and v1 in the first
configuration and v1 and v0 in the second. Suppose that we stop processes j
and j + 1, and let some other process run alone until it decides. Because this
third process can’t observe either locations j or j + 1, it can’t distinguish
between Cπ0 π1 and Cπ1 π0 , and thus decides the same value starting from
either configuration. But this contradicts the assumption that Cπi is i-valent.
It follows that there is no escape from bivalence with three processes, and
the rotor array plus atomic registers cannot be used to solve three-process
wait-free consensus.
The consensus number of this object is 2.

C.2.2 Set registers

Suppose we want to implement a set register S in a message-passing system,
where a set register provides operations insert(S, v), which inserts a new
element v in S, and read(S), which returns the set of all elements previously
inserted into S. So, for example, after executing insert(S, 3), insert(S, 1),
and insert(S, 1); read(S) would return {1, 3}.

1. Give an algorithm for implementing a linearizable set register where

all operations terminate in finite time, in a deterministic asynchronous
message-passing system with f < n/2 crash failures and no failure
detectors, or show that no such algorithm is possible.

2. Suppose that we change the read(S) operation to return a list of all

the elements of S in the order they were first inserted (e.g., [3, 1] in the
example above). Call the resulting object an ordered set register.
Give an algorithm for implementing a linearizable ordered set register
APPENDIX C. SAMPLE ASSIGNMENTS FROM SPRING 2016 386

under the same conditions as above, or show that no such algorithm is

possible.

Solution
1. It’s probably possible to do this with some variant of ABD, but getting
linearizability when there are multiple concurrent insert operations
will be tricky.
Instead, we’ll observe that it is straightforward to implement a set
register using a shared-memory snapshot: each process writes to A[i]
the set of all values it has ever inserted, and a read consists of taking
a snapshot and then taking the union of the values. Because we can
implement snapshots using atomic registers, and we can implement
atomic registers in a message-passing system with f < n/2 crash failures
using ABD, we can implement this construction in a message-passing
system with f < n/2 failures.

2. This we can’t do. The problem is that an ordered set register can solve
agreement: each process inserts its input, and the first input wins. But
FLP says we can’t solve agreement in an asynchronous message-passing
system with one crash failure.

C.2.3 Bounded failure detectors

Suppose you have a deterministic asynchronous message-passing system
equipped with a failure detector that is eventually weakly accurate and k-
bounded strongly complete, meaning that at least min(k, f ) faulty processes
are eventually permanently suspected by all processes, where f is the number
of faulty processes.
For what values of k, f , and n can this system solve agreement?

Solution
We can solve agreement using the k-bounded failure detector for n ≥ 2
processes if and only if f ≤ k and f < n/2.
Proof:
If k ≥ f , then every faulty process is eventually permanently suspected,
and the k-bounded failure detector is equivalent to the ♦S failure detector.
The Chandra-Toueg protocol [CT96] then solves consensus for us provided
f < n/2.
APPENDIX C. SAMPLE ASSIGNMENTS FROM SPRING 2016 387

If f ≥ n/2, the same partitioning argument used to show impossibility

with ♦P applies to the k-bounded detector as well (as indeed it applies to
any failure detector that is only eventually accurate).
If k < f , then if we have an algorithm that solves agreement for n
processes, then we can turn it into an algorithm that solves agreement for
n − k processes with f − k failures, using no failure detector at all. The
idea is that the n − k processes can pretend that there are an extra k faulty
processes that send no messages and that are permanently suspected. But
this algorithm runs in a standard asynchronous system with f − k failures,
and FLP says we can’t solve agreement in such a system with n ≥ 2 and
f ≥ 1. So this rules out solving agreement in the original system if k < f
and k ≤ n − 2.
There is one remaining case, where k = n − 1 and f = n. Here we can
actually solve consensus when n = 1 (because we can always solve consensus
when n = 1). For larger n, we have f ≥ n/2. So there is only one exception
to the general rule that we need f ≤ k and f < n/2.

C.3 Assignment 3: due Wednesday, 2016-04-20, at

5:00pm
C.3.1 Fetch-and-max

1 procedure fetchAndMax(r, 0 : x)
2 if switch = 0 then
3 return 0 : fetchAndMax(left, x)
4 else
5 return 1 : fetchAndMax(right, 0)

6 procedure fetchAndMax(r, 1 : x)
7 v ← fetchAndMax(right, x)
8 if TAS(switch) = 0 then
9 return 0 : fetchAndMax(left, 0)
10 else
11 return 1 : v

Algorithm C.4: Max register modified to use a test-and-set bit

Algorithm C.4 replaces the switch bit in the max register implementation
from Algorithm 21.2 with a test-and-set, and adds some extra machinery to
APPENDIX C. SAMPLE ASSIGNMENTS FROM SPRING 2016 388

return the old value of the register before the write.

Define a fetch-and-max register as a RMW object that supports a single
operation fetchAndMax(x) that, as an atomic action, (a) replaces the old
value v in the register with the maximum of x and v; and (b) returns the
old value v.
Suppose that left and right are both linearizable wait-free k-bit fetch-and-
max registers. Show that Algorithm C.4 implements a linearizable wait-free
(k + 1)-bit fetch-and-max register, or give an example of an execution that
violates linearizability.

Solution
Here is a bad execution (there are others). Let k = 1, and let π1 do
fetchAndMax(01) and π2 do fetchAndMax(10). Run these operations concur-
rently as follows:

1. π1 reads switch and sees 0.

2. π2 does fetchAndMax(right, 0).

3. π2 does TAS(switch) and sees 0.

4. π2 does fetchAndMax(left, 0) and sees 0.

5. π1 does fetchAndMax(left, 1) and sees 0.

Now both π1 and π2 return 00. But in the sequential execution π1 π2 , π2

returns 01; and in the sequential execution π2 π1 , π1 returns 10. Since π1
and π2 return the values they return in the concurrent execution in neither
sequential execution, the concurrent execution is not linearizable.

C.3.2 Median
Define a median register as an object r with two operations addSample(r, v),
where v is any integer, and computeMedian(r). The addSample operation
adds a sample to the multiset M of integers stored in the register, which
is initially empty. The computeMedian operations returns a median of this
multiset, defined as a value x with the property that (a) x is in the multiset;
(b) at least |M |/2 values v in the multiset are less than or equal to x; (c) at
least |M |/2 values v in the multiset are greater than or equal to x.
For example, if we add the samples 1, 1, 3, 5, 5, 6, in any order, then a
subsequent computeMedian can return either 3 or 5.
APPENDIX C. SAMPLE ASSIGNMENTS FROM SPRING 2016 389

Suppose that you wish to implement a linearizable wait-free median

register using standard atomic registers and resettable test-and-set bits. Give
tight (up to constants) asymptotic upper and lower bounds on the number
of such objects you would need. You may assume that the atomic registers
may hold arbitrarily large values.

Solution
For the upper bound, we can do it with O(n) registers using any linear-space
snapshot algorithm (for example, Afek et al. [AAD+ 93]). Each process stores
in its own segment of the snapshot object the multiset of all samples added
by that process; addSample just adds a new sample to the process’s segment.
For computeMedian, take a snapshot, then take the union of all the multisets,
then compute the median of this union. Linearizability and wait-freedom
of both operations are immediate from the corresponding properties of the
snapshot object.
For the lower bound, use JTT [JTT00]. Observe that both atomic
registers and resettable test-and-sets are historyless: for both types, the new
state after an operation doesn’t depend on the old state. So JTT applies if
we can show that the median register is perturbable.
Suppose that we have a schedule Λk Σk π in which Λk consists of an
arbitrary number of median-register operations of which at most k are
incomplete, Σk consists of k pending base object operations (writes, test-and-
sets, or test-and-set resets) covering k distinct base objects, and π is a read
operation by a process not represented in Λk Σk . We need to find a sequence
of operations γ that can be inserted between Λk and Σk that changes the
outcome of π.
Let S be the multiset of all values appearing as arguments to addSample
operations that start in Λk or Σk . Let x = max S (or 0 if S is empty), and let
γ consist of |S| + 1 addSample(r, x + 1) operations. Write T for the multiset
of |S| + 1 copies of x + 1. Then in any linearization of Λk γΣk π, the multiset
U of samples contained in r when π executes includes at least all of T and at
most all of S; this means that a majority of values in U are equal to x + 1,
and so the median is x + 1. But x + 1 does not appear in S, so π can’t
return it in Λk Σk π. It follows that a median register is in fact perturbable,
and JTT applies, which means that we need at least Ω(n) base objects to
implement a median register.
APPENDIX C. SAMPLE ASSIGNMENTS FROM SPRING 2016 390

C.3.3 Randomized two-process test-and-set with small reg-

isters
Algorithm C.5 gives an implementation of a randomized one-shot test-and-set
for two processes, each of which may call the procedure at most once, with
its process id (0 or 1) as an argument.
The algorithm uses two registers, a0 and a1 , that are both initialized
to 0 and hold values in the range 0 . . . m − 1, where m is a positive integer.
Unfortunately, whoever wrote it forgot to specify the value of m.

1 procedure TAS(i)
2 myPosition ← 0
3 while true do
4 otherPosition ← read(a¬i )
5 x ← myPosition − otherPosition
6 if x ≡ 2 (mod m) then
7 return 0
8 else if x ≡ −1 (mod m) do
9 return 1
10 else if fair coin comes up heads do
11 myPosition ← (myPosition + 1) mod m
12 write(ai , myPosition)

Algorithm C.5: Randomized one-shot test-and-set for two processes

For what values of m does Algorithm C.5 correctly implement a one-shot,

probabilistic wait-free, linearizable test-and-set, assuming:

1. An oblivious adversary?

2. An adaptive adversary?

Solution
For the oblivious adversary, we can quickly rule out m < 5, by showing that
there is an execution in each case where both processes return 0:

• When m = 1 or m = 2, both processes immediately return 0, because

the initial difference 0 is congruent to 2 mod m.

• When m = 3, there is an execution in which p0 writes 1 to a0 , p1 reads

this 1 and computes x = −1 ≡ 2 (mod 3) and returns 0, then p0 reads
APPENDIX C. SAMPLE ASSIGNMENTS FROM SPRING 2016 391

0 from a0 , computes x = 1, advances a0 to 2, then re-reads 0 from a0 ,

computes x = 2, and returns 0.

• When m = 4, run p0 until it writes 2 to a0 . It then computes x = 2

and returns 0. If we now wake up p1 , it computes x = −2 ≡ 2 (mod 4)
and also returns 0.

When m ≥ 5 and the adversary is oblivious, the implementation works.

We need to show both linearizability and termination. We’ll start with
linearizability.
Observe that in a sequential execution, first process to perform TAS
returns 0 and the second 1. So we need to show (a) that the processes
between them return both values, and (b) that if one process finishes before
the other starts, the first process returns 0.
It is immediate from Algorithm C.5 that in any reachable configuration,
myPositioni ∈ {ai , ai + 1}, because process i can only increment myPosition
at most once before writing its value to ai .
Below we will assume without loss of generality that p0 is the first process
to perform its last read before returning.
Suppose that p0 returns 0. This means that p0 observed a1 ≡ a0 − 2
(mod m). So at the time p0 last read a1 , myPosition1 was congruent to either
a0 − 1 or a0 − 2. This means that on its next read of a0 , p1 will compute x
congruent to either −1 or −2. Because m is at least 5, in neither case will
it mistake this difference for 2. If it computed x ≡ −1, it returns 1; if it
computed x ≡ −2, it does not return immediately, but eventually it will flip
its coin heads, increment myPosition1 , and return 1. In either case we have
that exactly one process returns each value.
Alternatively, suppose that p0 returns 1. Then p0 reads a1 ≡ a0 + 1,
and at the time of this read, myPosition1 is either congruent to a0 + 1 or
a0 + 2. In the latter case, p1 returns 0 after its next read; in the former,
p1 eventually increments myPosition1 and then returns 0. In either case we
again have that exactly one process returns each value.
Now suppose that p0 runs to completion before p1 starts. Initially, p0
sees a0 ≡ a1 , but eventually p0 increments a0 enough times that a0 − a1 ≡ 2;
p0 returns 0.
To show termination (with probability 1), consider any configuration in
which neither process has returned. During the next 2k steps, at least one
process takes k steps. Suppose that during this interval, this fast process
increments myPosition at every opportunity, while the other process does
not increment myPosition at all (this event occurs with nonzero probability
APPENDIX C. SAMPLE ASSIGNMENTS FROM SPRING 2016 392

for any fixed k, because the coin-flips are uncorrelated with the oblivious
adversary’s choice of which process is fast). Then for k sufficiently large, the
fast process eventually sees a0 − a1 congruent to either 2 or −1 and returns.
Since this event occurs with independent nonzero probability in each interval
of length 2k, eventually it occurs.2
Once one process has terminated, the other increments myPosition in-
finitely often, so it too eventually sees a gap of 2 or −1.
For the adaptive adversary, the adversary can prevent the algorithm from
terminating. Starting from a state in which both processes are about to
read and a0 = a1 = k, run p0 until it is about to write (k + 1) mod m to a0
(unlike the oblivious adversary, the adaptive adversary can see when this will
happen). Then run p1 until it is about to write (k + 1) mod m to a1 . Let
both writes go through. We are now in a state in which both processes are
about to read, and a0 = a1 = (k + 1) mod m. So we can repeat this strategy
forever.

C.4 Presentation (for students taking CPSC 565):

due Wednesday, 2016-04-27
Students taking CPSC 565, the graduate version of the class, are expected to
give a 15-minute presentation on a recent paper in the theory of distributed
computing.
The choice of paper to present should be made in consultation with the
instructor. To a first approximation, any paper from PODC, DISC, or a
similar conference in the last two or three years (that is not otherwise covered
in class) should work.
Because of the limited presentation time, you are not required to get
into all of the technical details of the paper, but your presentation should
include3

1. Title, authors, and date and venue of publication of the paper.

2. A high-level description of the main result. Unlike a talk for a general

2
The fancy way to prove this is to invoke the second Borel-Cantelli lemma of probability
theory. Or we can just argue that the probability that we don’t terminate in the first `
intervals is at most (1 − )` , which goes to zero in the limit.
3
Literary theorists will recognize this as a three-act structure (preceded by a title card):
introduce the main character, make their life difficult, then resolve their problems in time
for the final curtain. This is not the only way to organize a talk, but if done right it has
the advantage of keeping the audience awake.
APPENDIX C. SAMPLE ASSIGNMENTS FROM SPRING 2016 393

audience, you can assume that your listeners know at least everything
that we’ve talked about so far in the class.

3. A description of where this result fits into the literature (e.g., solves
an open problem previously proposed in [...], improves on the previous
best running time for an algorithm from [...], gives a lower bound or
impossibility result for a problem previously proposed by [...], opens
up a new area of research for studying [...]), and why it is interesting
and/or hard.

4. A description (also possibly high-level) of the main technical mecha-

nism(s) used to get the main result.

You do not have to prepare slides for your presentation if you would
prefer to use the blackboard, but you should make sure to practice it in
advance to make sure it fits in the allocated time. The instructor will be
happy to offer feedback on draft versions if available far enough before the
actual presentation date.
Relevant dates:

2016-04-13 Paper selection due.

2016-04-22 Last date to send draft slides or arrange for a practice presen-
tation with the instructor if you want guaranteed feedback.

2016-04-27 Presentations, during the usual class time.

C.5 CS465/CS565 Final Exam, May 10th, 2016

Write your answers in the blue book(s). Justify your answers. Work alone.
Do not use any notes or books.
There are four problems on this exam, each worth 20 points, for a total
of 80 points. You have approximately three hours to complete this exam.

C.5.1 A slow register (20 points)

Define a second-to-last register as having a read operation that always
returns the second-to-last value written to it. For example, after write (1),
write (2), write (3), a subsequent read operation will return 2. If fewer
that two write operations have occurred, a read will return ⊥.
What is the consensus number of this object?
APPENDIX C. SAMPLE ASSIGNMENTS FROM SPRING 2016 394

Solution
The consensus number of this object is 2.
For two processes, have each process i write its input to a standard
atomic register r[i], and then write its id to a shared second-to-last-value
register s. We will have whichever process writes to s first win. After writing,
process i can detect which process wrote first by reading s once, because
it either sees ⊥ (meaning the other process has not written yet) or it sees
the identity of the process that wrote first. In either case it can return the
winning process’s input.
For three processes, the usual argument gets us to a configuration C
where all three processes are about to execute operations x, y, and z on
the same object, where each operation moves from a bivalent to a univalent
state. Because we know that this object can’t be a standard atomic register,
it must be a second-to-last register. We can also argue that all of x, y, and
z are writes, because if one of them is not, the processes that don’t perform
it can’t tell if it happened or not.
Suppose that Cx is 0-valent and Cy is 1-valent. Then Cxyz is 0-valent
and Cyz is 1-valent. But these configurations are indistinguishable to any
process but x. It follows that the second-to-last register can’t solve consensus
for three processes.

C.5.2 Two leaders (20 points)

Assume that you are working in an asynchronous message-passing system
organized as a connected graph, where all processes run the same code except
that each process starts with an id and the knowledge of the ids of all of its
neighbors. Suppose that all of these ids are unique, except that the smallest
id (whatever it is) might appear on either one or two processes.
Is it possible in all cases to detect which of these situations hold? Either
give an algorithm that allows all processes to eventually correctly return
whether there are one or two minimum-id processes in an arbitrary connected
graph, or show that no such algorithm is possible.

Solution
Here is an algorithm.
If there are two processes p and q with the same id that are adjacent to
each other, they can detect this in the initial configuration, and transmit
this fact to all the other processes by flooding.
APPENDIX C. SAMPLE ASSIGNMENTS FROM SPRING 2016 395

If these processes p and q are not adjacent, we will need some other
mechanism to detect them. Define the extended id of a process as its own id
followed by a list of the ids of its neighbors in some fixed order. Order the
extended ids lexicographically, so that a process with a smaller id also has a
smaller extended id.
Suppose now that p and q are not adjacent and have the same extended
id. Then they share the same neighbors, and each of these neighbors will see
that p and q have duplicate ids. So we can do an initial round of messages
where each process transmits its extended id to its neighbors, and if p and q
observe that their id is a duplicate, they can again notify all the processes to
return that there are two leaders by flooding.
The remaining case is that p and q have distinct extended ids, or that
only one minimum-process id exists. In either case we can run any standard
broadcast-based leader-election algorithm, using the extended ids, which will
leave us with a tree rooted at whichever process has the minimum extended
id. This process can then perform convergecast to detect if there is another
process with the same id, and perform broadcast to inform all processes of
this fact.

C.5.3 A splitter using one-bit registers (20 points)

Algorithm C.6 implements a splitter-like object using one-bit registers. It
assumes that each process has a unique id id consisting of k = dlg ne bits
idk−1 idk−2 . . . id0 . We would like this object to have the properties that (a)
if exactly one process executes the algorithm, then it wins; and (b) in any
execution, at most one process wins.
Show that the algorithm has these properties, or give an example of an
execution where it does not.

Solution
The implementation is correct.
If one process runs alone, it sets A[i][idi ] for each i, sees 0 in door, then
sees 0 in each location A[i][¬idi ] and wins. So we have property (a).
Now suppose that some process with id p wins in an execution that may
involve other processes. Then p writes A[i][pi ] for all i before observing 0 in
door, which means that it sets all these bits before any process writes 1 to
door. If some other process q also wins, then there is at least one position i
where pi = ¬qi , and q reads A[i][pi ] after writing 1 to door. But then q sees
1 in this location and loses, a contradiction.
APPENDIX C. SAMPLE ASSIGNMENTS FROM SPRING 2016 396

shared data:
1 one-bit atomic registers A[i][j] for i = 0 . . . dlg ne − 1 and j ∈ {0, 1},
all initially 0
2 one-bit atomic register door, initially 0
3 procedure splitter(id)
4 for i ← 0 to k − 1 do
5 A[i][idi ] ← 1
6 if door = 1 then
7 return lose
8 door ← 1
9 for i ← 0 to k − 1 do
10 if A[i][¬idi ] = 1 then
11 return lose

12 return win
Algorithm C.6: Splitter using one-bit registers

C.5.4 Symmetric self-stabilizing consensus (20 points)

Suppose we have a synchronous system consisting of processes organized in a
connected graph. The state of each process is a single bit, and each process
can directly observe the number of neighbors that it has and how many of
them have 0 bits and how many have 1 bits. At each round, a process counts
the number of neighbors k0 with zeros, the number k1 with ones, and its
own bit b, and chooses a new bit for the next round f (b, k0 , k1 ) according to
some rule f that is the same for all processes. The goal of the processes is
to reach consensus, where all processes have the same bit forever, starting
from an arbitrary initial configuration. An example of a rule that has this
property is for f to output 1 if b = 1 or k1 > 0.
However, this rule is not symmetric with respect to bit values: if we
replace all ones by zeros and vice versa, we get different behavior.
Prove or disprove: There exists a rule f that is symmetric, by which we
mean that f (b, k0 , k1 ) = ¬f (¬b, k1 , k0 ) always, such that applying this rule
starting from an arbitrary configuration in an arbitrary graph eventually
converges to all processes having the same bit forever.
APPENDIX C. SAMPLE ASSIGNMENTS FROM SPRING 2016 397

Solution
Disproof by counterexample: Fix some f , and consider a graph with two
processes p0 and p1 connected by an edge. Let p0 start with 0 and p1 start
with 1. Then p0 ’s next state is f (0, 0, 1) = ¬f (1, 1, 0) 6= f (1, 1, 0), which is
p1 ’s next state. So either p0 still has 0 and p1 still has 1, in which case we
never make progress; or they swap their bits, in which case we can apply the
same analysis with p0 and p1 reversed to show that they continue to swap
back and forth forever. In either case the system does not converge.
Appendix D

Sample assignments from

Spring 2014

D.1 Assignment 1: due Wednesday, 2014-01-29, at

5:00pm
Bureaucratic part
Send me email! My address is [email protected].
In your message, include:

1. Your name.

2. Your status: whether you are an undergraduate, grad student, auditor,

etc.

3. Anything else you’d like to say.

(You will not be graded on the bureaucratic part, but you should do it
anyway.)

D.1.1 Counting evil processes

A connected bidirectional asynchronous network of n processes with identities
has diameter D and may contain zero or more evil processes. Fortunately,
the evil processes, if they exist, are not Byzantine, fully conform to RFC
3514 [Bel03], and will correctly execute any code we provide for them.
Suppose that all processes wake up at time 0 and start whatever protocol
we have given them. Suppose that each process initially knows whether it is

398
APPENDIX D. SAMPLE ASSIGNMENTS FROM SPRING 2014 399

evil, and knows the identities of all of its neighbors. However, the processes
do not know the number of processes n or the diameter of the network D.
Give a protocol that allows every process to correctly return the number
of evil processes no later than time D. Your protocol should only return a
value once for each process (no converging to the correct answer after an
initial wrong guess).

Solution
There are a lot of ways to do this. Since the problem doesn’t ask about
message complexity, we’ll do it in a way that optimizes for algorithmic
simplicity.
At time 0, each process initiates a separate copy of the flooding algorithm
(Algorithm 3.1). The message hp, N (p), ei it distributes consists of its own
identity, the identities of all of its neighbors, and whether or not it is evil.
In addition to the data for the flooding protocol, each process tracks a
set I of all processes it has seen that initiated a protocol and a set N of all
processes that have been mentioned as neighbors. The initial values of these
sets for process p are {p} and N (p), the neighbors of p.
Upon receiving a message hq, N (q), ei, a process adds q to I and N (q) to
N . As soon as I = N , the process returns a count of all processes for which
e = true.
Termination by D: Follows from the same analysis as flooding. Any
process at distance d from p has p ∈ I by time d, so I is complete by time D.
S
Correct answer: Observe that N = i∈I N (i) always. Suppose that there
is some process q that is not in I. Since the graph is connected, there is a
path from p to q. Let r be the last node in this path in I, and let s be the
following node. Then s ∈ N \ I and N 6= I. By contraposition, if I = N
then I contains all nodes in the network, and so the count returned at this
time is correct.

D.1.2 Avoiding expensive processes

Suppose that you have a bidirectional but not necessarily complete asyn-
chronous message-passing network represented by a graph G = (V, E) where
each node in V represents a process and each edge in E connects two pro-
cesses that can send messages to each other. Suppose further that each
process is assigned a weight 1 or 2. Starting at some initiator process, we’d
like to construct a shortest-path tree, where each process points to one of
APPENDIX D. SAMPLE ASSIGNMENTS FROM SPRING 2014 400

its neighbors as its parent, and following the parent pointers always gives a
path of minimum total weight to the initiator.1
Give a protocol that solves this problem with reasonable time, message,
and bit complexity, and show that it works.

Solution
There’s an ambiguity in the definition of total weight: does it include the
weight of the initiator and/or the initial node in the path? But since these
values are the same for all paths to the initiator from a given process, they
don’t affect which is lightest.
If we don’t care about bit complexity, there is a trivial solution: Use an
existing BFS algorithm followed by convergecast to gather the entire structure
of the network at the initiator, run your favorite single-source shortest-path
algorithm there, then broadcast the results. This has time complexity O(D)
and message complexity O(DE) if we use the BFS algorithm from §4.3. But
the last couple of messages in the convergecast are going to be pretty big.
A solution by reduction: Suppose that we construct a new graph G0
where each weight-2 node u in G is replaced by a clique of nodes u1 , u2 , . . . uk ,
with each node in the clique attached to a different neighbor of u. We then
run any breadth-first search protocol of our choosing on G0 , where each
weight-2 node simulates all members of the corresponding clique. Because
any path that passes through a clique picks up an extra edge, each path in
the breadth-first search tree has a length exactly equal to the sum of the
weights of the nodes other than its endpoints.
A complication is that if I am simulating k nodes, between them they
may have more than one parent pointer. So we define u.parent to be ui .parent
where ui is a node at minimum distance from the initiator in G0 . We also
re-route any incoming pointers to uj 6= ui to point to ui instead. Because ui
was chosen to have minimum distance, this never increases the length of any
path, and the resulting modified tree is a still a shortest-path tree.
Adding nodes blows up |E 0 |, but we don’t need to actually send messages
between different nodes ui represented by the same process. So if we use the
§4.3 algorithm again, we only send up to D messages per real edge, giving
O(D) time and O(DE) messages.
If we don’t like reductions, we could also tweak one of our existing
algorithms. Gallager’s layered BFS (§4.2) is easily modified by changing the
1
Clarification added 2014-01-26: The actual number of hops is not relevant for the
construction of the shortest-path tree. By shortest path, we mean path of minimum total
weight.
APPENDIX D. SAMPLE ASSIGNMENTS FROM SPRING 2014 401

depth bound for each round to a total-weight bound. The synchronizer-based

BFS can also be modified to work, but the details are messy.

D.2 Assignment 2: due Wednesday, 2014-02-12, at

5:00pm
D.2.1 Synchronous agreement with weak failures
Suppose that we modify the problem of synchronous agreement with crash
failures from Chapter 9 so that instead of crashing a process forever, the
adversary may jam some or all of its outgoing messages for a single round.
The adversary has limited batteries on its jamming equipment, and can
only cause f such one-round faults. There is no restriction on when these
one-round jamming faults occur: the adversary might jam f processes for
one round, one process for f rounds, or anything in between, so long as the
sum over all rounds of the number of processes jammed in each round is at
most f . For the purposes of agreement and validity, assume that a process
is non-faulty if it is never jammed.2
As a function of f and n, how many rounds does it take to reach agreement
in the worst case in this model, under the usual assumptions that processes
are deterministic and the algorithm must satisfy agreement, termination,
and validity? Give the best upper and lower bounds that you can.

Solution
√
The par solution for this is an Ω( f ) lower bound and O(f ) upper bound. I
don’t know if it is easy to do better than this.
For the lower bound, observe that the adversary can simulate an ordinary
crash failure by jamming a process in every round starting in the round it
crashes in. This means that in an r-round protocol, we can simulate k crash
failures with kr jamming faults. From the Dolev-Strong lower bound [DS83]
(see also Chapter 9), we know that there is no r-round protocol with k = r
crash failures faults, so there is√no r-round protocol with r2 jamming faults.
This gives a lower bound of b f c + 1 on the number of rounds needed to
solve synchronous agreement with f jamming faults.3
2
Clarifications added 2014-02-10: We assume that processes don’t know that they are
being jammed or which messages are lost (unless the recipient manages to tell them that a
message was not delivered). As in the original model, we assume a complete network and
that all processes have known identities.
3
Since Dolev-Strong only needs to crash one process per round, we don’t really need
APPENDIX D. SAMPLE ASSIGNMENTS FROM SPRING 2014 402

Figure D.1: Connected Byzantine nodes take over half a cut

For the upper bound, have every process broadcast its input every round.
After f + 1 rounds, there is at least one round in which no process is jammed,
so every process learns all the inputs and can take, say, the majority value.

D.2.2 Byzantine agreement with contiguous faults

Suppose that we restrict the adversary in Byzantine agreement to corrupting
a connected subgraph of the network; the idea is that the faulty nodes need
to coordinate, but can’t relay messages through the non-faulty nodes to do
so.
Assume the usual model for Byzantine agreement with a network in
the form of an m × m torus. This means that each node has a position
(x, y) in {0, . . . , m − 1}×{0, . . . , m − 1}, and its neighbors are the four nodes
(x + 1 mod m, y), (x − 1 mod m, y), (x, y + 1 mod m), and (x, y − 1 mod m).
For sufficiently large m,4 what is the largest number of faults f ; that this
system can tolerate and still solve Byzantine agreement?

Solution
The relevant bound here is the requirement that the network have enough
connectivity that the adversary can’t take over half of a vertex cut (see
§10.1.3). This is complicated slightly by the requirement that the faulty
nodes be contiguous.
The smallest vertex cut in a sufficiently large torus consists of the four
neighbors of a single node; however, these nodes are not connected. But we
can add a third node to connect two of them (see Figure D.1).
By adapting the usual lower bound we can use this construction to show
that f = 3 faults are enough to prevent agreement when m ≥ 3. The question
the full r jamming faults for processes that crash late. This could be used to improve the
constant for this argument.
4
Problem modified 2014-02-03. In the original version, it asked to compute f for all m,
but there are some nasty special cases when m is small.
APPENDIX D. SAMPLE ASSIGNMENTS FROM SPRING 2014 403

then is whether f = 2 faults is enough.

By a case analysis, we can show that any two nodes in a sufficiently large
torus are either adjacent themselves or can be connected by three paths,
where no two paths have adjacent vertices. Assume without loss of generality
that one of the nodes is at position (0, 0). Then any other node is covered
by one of the following cases:
1. Nodes adjacent to (0, 0). These can communicate directly.
2. Nodes at (0, i) or (i, 0). These cases are symmetric, so we’ll describe the
solution for (0, i). Run one path directly north: (0, 1), (0, 2), . . . , (0, i −
1). Similarly, run a second path south: (0, −1), (0, −2), . . . (0, i + 1).
For the third path, take two steps east and then run north and back
west: (1, 0), (2, 0), (2, 1), (2, 2), . . . , (2, i), (1, i). These paths are all non-
adjacent as long as m ≥ 4.
3. Nodes at (±1, i) or (i, ±1), where i is not −1, 0, or 1. Suppose the node
is at (1, i). Run one path east then north through (1, 0), (1, 1), . . . , (1, i−
1). The other two paths run south and west, with a sideways jog in the
middle as needed. This works for m sufficiently large to make room
for the sideways jogs.
4. Nodes at (±1, ±1) or (i, j) where neither of i or j is −1, 0, or 1. Now
we can run one path north then east, one east then north, one south
then west, and one west then south, creating four paths in a figure-eight
pattern centered on (0, 0).

D.3 Assignment 3: due Wednesday, 2014-02-26, at

5:00pm
D.3.1 Among the elect
The adversary has decided to be polite and notify each non-faulty processes
when he gives up crashing it. Specifically, we have the usual asynchronous
message-passing system with up to f faulty processes, but every non-faulty
process is eventually told that it is non-faulty. (Faulty processes are told
nothing.)
For what values of f can you solve consensus in this model?

Solution
We can tolerate f < n/2, but no more.
APPENDIX D. SAMPLE ASSIGNMENTS FROM SPRING 2014 404

If f < n/2, the following algorithm works: Run Paxos, where each
process i waits to learn that it is non-faulty, then acts as a proposer for
proposal number i. The highest-numbered non-faulty process then carries
out a proposal round that succeeds because no higher proposal is ever issued,
and both the proposer (which is non-faulty) and a majority of accepters
participate.
If f ≥ n/2, partition the processes into two groups of size bn/2c, with
any leftover process crashing immediately. Make all of the processes in both
groups non-faulty, and tell each of them this at the start of the protocol.
Now do the usual partitioning argument: Run group 0 with inputs 0 with
no messages delivered from group 1 until all processes decide 0 (we can do
this because the processes can’t distinguish this execution from one in which
the group 1 processes are in fact faulty). Run group 1 similarly until all
processes decide 1. We have then violated agreement, assuming we didn’t
previously violate termination of validity.

D.3.2 Failure detectors on the cheap

Suppose we do not have the budget to equip all of our machines with failure
detectors. Instead, we order an eventually strong failure detector for k
machines, and the remaining n − k machines get fake failure detectors that
never suspect anybody. The choice of which machines get the real failure
detectors and which get the fake ones is under the control of the adversary.
This means that every faulty process is eventually permanently suspected
by every non-faulty process with a real failure detector, and there is at
least one non-faulty process that is eventually permanently not suspected by
anybody. Let’s call the resulting failure detector ♦Sk .
Let f be the number of actual failures. Under what conditions on f , k,
and n can you still solve consensus in the usual deterministic asynchronous
message-passing model using ♦Sk ?

Solution
First observe that ♦S can simulate ♦Sk for any k by having n − k processes
ignore the output of their failure detectors. So we need f < n/2 by the usual
lower bound on ♦S.
If f ≥ k, we are also in trouble. The f > k case is easy: If there exists
a consensus protocol for f > k, then we can transform it into a consensus
protocol for n − k processes and f − k failures, with no failure detectors at all,
by pretending that there are an extra k processes with real failure detectors
APPENDIX D. SAMPLE ASSIGNMENTS FROM SPRING 2014 405

that crash immediately. The FLP impossibility result rules this out.
If f = k, we have to be a little more careful. By immediately crashing
f − 1 processes with real failure detectors, we can reduce to the f = k = 1
case. Now the adversary runs the FLP strategy. If no processes crash, then
all n − k + 1 surviving process report no failures; if it becomes necessary to
crash a process, this becomes the one remaining process with the real failure
detector. In either case the adversary successfully prevents consensus.
So let f < k. Then we have weak completeness, because every faulty
process is eventually permanently suspected by at least k − f > 0 processes.
We also have weak accuracy, because it is still the case that some process
is eventually permanently never suspected by anybody. By boosting weak
completeness to strong completeness as described in §13.2.3, we can turn
out failure detector into ♦S, meaning we can solve consensus precisely when
f < min(k, n/2).

D.4 Assignment 4: due Wednesday, 2014-03-26, at

5:00pm
D.4.1 A global synchronizer with a global clock
Consider an asynchronous message-passing system with n processes in a
bidirectional ring with no failures. Suppose that the processes are equipped
with a global clock, which causes a local event to occur simultaneously at
each process every c time units, where as usual 1 is the maximum message
delay. We would like to use this global clock to build a global synchronizer.
Provided c is at least 1, a trivial approach is to have every process advance
to the next round whenever the clock pulse hits. This gives one synchronous
round every c time units.
Suppose that c is greater than 1 but still o(n). Is it possible to build a
global synchronizer in this model that runs more than a constant ratio faster
than this trivial global synchronizer in the worst case?

Solution
No. We can adapt the lower bound on the session problem from §7.4.2 to
apply in this model.
Consider an execution of an algorithm for the session problem in which
each message is delivered exactly one time unit after it is sent. Divide
it as in the previous proof into a prefix β containing special actions and
a suffix δ containing no special actions. Divide β further into segments
APPENDIX D. SAMPLE ASSIGNMENTS FROM SPRING 2014 406

β1 , β2 , β3 , . . . , βk , where each segment ends with a clock pulse. Following

the standard argument, because each segment has duration less than the
diameter of the network, there is no causal connection between any special
actions done by processes at opposite ends of the network that are in the
same segment βi . So we can causally shuffle each βi to get a new segment βi0
where all special actions of process p0 (say) occur before all special actions
of process pn/2 . This gives at most one session per segment, or at most one
session for every c time units.
Since a globally synchronous system can do one session per round, this
means that our global synchronizer can only produce one session per c time
units as well.

D.4.2 A message-passing counter

A counter is a shared object that support operations inc and read, where
read returns the number of previous inc operations.
Algorithm D.1 purports to implement a counter in an asynchronous
message-passing system subject to f < n/2 crash failures. In the algorithm,
each process i maintains a vector ci of contributions to the counter from all
the processes, as well as a nonce ri used to distinguish responses to different
read operations from each other. All of these values are initially zero.
Show that the implemented counter is linearizable, or give an example of
an execution where it isn’t.

Solution
This algorithm is basically implementing an array of ABD registers [ABND95],
but it omits the second phase on a read where any information the reader
learns is propagated to a majority. So we expect it to fail the same way ABD
would without this second round, by having two read operations return
values that are out of order with respect to their observable ordering.
Here is one execution that produces this bad outcome:

1. Process p1 starts an inc by updating c1 [1] to 1.

2. Process p2 carries out a read operation in which it receives responses

from p1 and p2 , and returns 1.

3. After p2 finishes, process p3 carries out a read operation in which it

receives responses from p2 and p3 , and returns 0.
APPENDIX D. SAMPLE ASSIGNMENTS FROM SPRING 2014 407

1 procedure inc
2 ci [i] ← ci [i] + 1
3 Send ci [i] to all processes.
4 Wait to receive ack(ci [i]) from a majority of processes.
5 upon receiving c from j do
6 ci [j] ← max(ci [j], c)
7 Send ack(c) to j.
8 procedure read
9 ri ← ri + 1
10 Send read(ri ) to all processes.
11 Wait to receive respond(ri , cj ) from a majority of processes j.
P
12 return k maxj cj [k]
13 upon receiving read(r) from j do
14 Send respond(r, ci ) to j
Algorithm D.1: Counter algorithm for Problem D.4.2.

If we want to be particularly perverse, we can exploit the fact that p2

doesn’t record what it learns in its first read to have p2 do the second read
that returns 0 instead of p3 . This shows that Algorithm D.1 isn’t even
sequentially consistent.
The patch, if we want to fix this, is to include the missing second phase
from ABD in the read operation: after receiving values from a majority, I
set ci [k] to maxj cj [k] and send my updated values to a majority. That the
resulting counter is linearizable is left as an exercise.

D.5 Assignment 5: due Wednesday, 2014-04-09, at

5:00pm
D.5.1 A concurrency detector
Consider the following optimistic mutex-like object, which we will call a
concurrency detector. A concurrency detector supports two operations
for each process i, enteri and exiti . These operations come in pairs: a
process enters a critical section by executing enteri , and leaves by executing
exiti . The behavior of the object is undefined if a process calls enteri twice
without an intervening exiti , or calls exiti without first calling enteri .
Unlike mutex, a concurrency detector does not enforce that only one
APPENDIX D. SAMPLE ASSIGNMENTS FROM SPRING 2014 408

process is in the critical section at a time; instead, exiti returns 1 if the

interval between it and the previous enteri overlaps with some interval
between a enterj and corresponding exitj for some j 6= i, and returns 0 if
there is no overlap.
Is there a deterministic linearizable wait-free implementation of a concur-
rency detector from atomic registers? If there is, give an implementation. If
there is not, give an impossibility proof.

Solution
It is not possible to implement this object using atomic registers.
Suppose that there were such an implementation. Algorithm D.2 im-
plements two-process consensus using a two atomic registers and a single
concurrency detector, initialized to the state following enter1 .

1 procedure consensus1 (v)

2 r1 ← v
3 if exit1 () = 1 then
4 return r2
5 else
6 return v

7 procedure consensus2 (v)

8 r2 ← v
9 enter2 ()
10 if exit2 () = 1 then
11 return v
12 else
13 return r1

Algorithm D.2: Two-process consensus using the object from Prob-

lem D.5.1

Termination is immediate from the absence of loops in the code.

To show validity and termination, observe that one of two cases holds:

1. Process 1 executes exit1 before process 2 executes enter2 . In this

case there is no overlap between the interval formed by the implicit
enter1 and exit1 and the interval formed by enter2 and exit2 . So
the exit1 and exit2 operations both return 0, causing process 1 to
return its own value and process 2 to return the contents of r1 . These
APPENDIX D. SAMPLE ASSIGNMENTS FROM SPRING 2014 409

will equal process 1’s value, because process 2’s read follows its call to
enter2 , which follows exit1 and thus process 1’s write to r1 .
2. Process 1 executes exit1 after process 2 executes enter2 . Now both
exit operations return 1, and so process 2 returns its own value while
process 1 returns the contents of r2 , which it reads after process 2
writes its value there.
In either case, both processes return the value of the first process to access
the concurrency detector, satisfying both agreement and validity. This would
give a consensus protocol for two processes implemented from atomic registers,
contradicting the impossibility result of Loui and Abu-Amara [LAA87].

D.5.2 Two-writer sticky bits

A two-writer sticky bit is a sticky bit that can be read by any process,
but that can only be written to by two specific processes.
Suppose that you have an unlimited collection of two-writer sticky bits
for each pair of processes, plus as many ordinary atomic registers as you
need. What is the maximum number of processes for which you can solve
wait-free binary consensus?

Solution
If n = 2, then a two-writer sticky bit is equivalent to a sticky bit, so we can
solve consensus.
If n ≥ 3, suppose that we maneuver our processes as usual to a bivalent
configuration C with no bivalent successors. Then there are three pending
operations x, y, and z, that among them produce both 0-valent and 1-valent
configurations. Without loss of generality, suppose that Cx and Cy are both
0-valent and Cz is 1-valent. We now consider what operations these might
be.

1. If x and z apply to different objects, then Cxz = Czx must be both

0-valent and 1-valent, a contradiction. Similarly if y and z apply to
different objects. This shows that all three operations apply to the
same object O.
2. If O is a register, then the usual case analysis of Loui and Abu-
Amara [LAA87] gives us a contradiction.
3. If O is a two-writer sticky bit, then we can split cases further based on
z:
APPENDIX D. SAMPLE ASSIGNMENTS FROM SPRING 2014 410

(a) If z is a read, then either:

i. At least one of x and y is a read. But then Cxz = Czx or
Cyz = Czy, and we are in trouble.
ii. Both x and y are writes. But then Czx (1-valent) is indistin-
guishable from Cx (0-valent) by the two processes that didn’t
perform z: more trouble.
(b) If z is a write, then at least one of x or y is a read; suppose it’s x.
Then Cxz is indistinguishable from Cz by the two processes that
didn’t perform x.

Since we reach a contradiction in all cases, it must be that when n ≥ 3,

every bivalent configuration has a bivalent successor, which shows that we
can’t solve consensus in this case. The maximum value of n for which we
can solve consensus is 2.

D.6 Assignment 6: due Wednesday, 2014-04-23, at

5:00pm
D.6.1 A rotate register
Suppose that you are asked to implement a concurrent m-bit register that
supports in addition to the usual read and write operations a RotateLeft
operation that rotates all the bits to the left; this is equivalent to doing a
left shift (multiplying the value in the register by two) followed by replacing
the lowest-order bit with the previous highest-order bit.
For example, if the register contains 1101, and we do RotateLeft, it now
contains 1011.
Show that if m is sufficiently large as a function of the number of processes
n, Θ(n) steps per operation in the worst case are necessary and sufficient to
implement a linearizable wait-free m-bit shift register from atomic registers.

Solution
The necessary part is easier, although we can’t use JTT (Chapter 20) di-
rectly because having write operations means that our rotate register is not
perturbable. Instead, we argue that if we initialize the register to 1, we
get a mod-m counter, where increment is implemented by RotateLeft and
read is implemented by taking the log of the actual value of the counter.
Letting m ≥ 2n gives the desired Ω(n) lower bound, since a mod-2n counter
is perturbable.
APPENDIX D. SAMPLE ASSIGNMENTS FROM SPRING 2014 411

For sufficiency, we’ll show how to implement the rotate register using
snapshots. This is pretty much a standard application of known tech-
niques [AH90b, AM93], but it’s not a bad exercise to write it out.
Pseudocode for one possible solution is given in Algorithm D.3.
The register is implemented using a single snapshot array A. Each entry
in the snapshot array holds four values: a timestamp and process id indicating
which write the process’s most recent operations apply to, the initial write
value corresponding to this timestamp, and the number of rotate operations
this process has applied to this value. A write operation generates a new
timestamp, sets the written value to its input, and resets the rotate count
to 0. A rotate operation updates the timestamp and associated write value
to the most recent that the process sees, and adjusts the rotate count as
appropriate. A read operation combines all the rotate counts associated with
the most recent write to obtain the value of the simulated register.

1 procedure write(A, v)
2 s ← snapshot(A)
3 A[id] ← hmaxi s[i].timestamp + 1, id, v, 0i
4 procedure RotateLeft(A)
5 s ← snapshot(A)
6 Let i maximize hs[i].timestamp, s[i].processi
7 if s[i].timestamp = A[id].timestamp and
s[i].process = A[id].process then
// Increment my rotation count
8 A[id].rotations ← A[id].rotations + 1
9 else
// Reset and increment my rotation count
10 A[id] ← hs[i].timestamp, s[i].process, s[i].value, 1i

11 procedure read(A)
12 s ← snapshot(A)
13 Let i maximize hs[i].timestamp, s[i].processi
14 Let
P
r = j,s[j].timestamp=s[i].timestamp∧s[j].process=s[i].process s[j].rotations
15 return s[i].value rotated r times.
Algorithm D.3: Implementation of a rotate register

Since each operation requires one snapshot and at most one update, the
cost is O(n) using the linear-time snapshot algorithm of Inoue et al. [IMCT94].
APPENDIX D. SAMPLE ASSIGNMENTS FROM SPRING 2014 412

Linearizability is easily verified by observing that ordering all operations by

the maximum timestamp/process tuple that they compute and then by the
total number of rotations that they observe produces an ordering consistent
with the concurrent execution for which all return values of reads are correct.

D.6.2 A randomized two-process test-and-set

Algorithm D.4 gives pseudocode for a protocol for two processes p0 and p1 .
It uses two shared unbounded single-writer atomic registers r0 and r1 , both
initially 0. Each process also has a local variable s.

1 procedure TASi ()
2 while true do
3 with probability 1/2 do
4 ri ← ri + 1
5 else
6 ri ← ri
7 s ← r¬i
8 if s > ri then
9 return 1
10 else if s < ri − 1 do
11 return 0

Algorithm D.4: Randomized two-process test-and-set for D.6.2

1. Show that any return values of the protocol are consistent with a
linearizable, single-use test-and-set.

2. Will this protocol always terminate with probability 1 assuming an

oblivious adversary?

3. Will this protocol always terminate with probability 1 assuming an

adaptive adversary?

Solution
1. To show that this implements a linearizable test-and-set, we need to
show that exactly one process returns 0 and the other 1, and that if one
process finishes before the other starts, the first process to go returns
1.
APPENDIX D. SAMPLE ASSIGNMENTS FROM SPRING 2014 413

Suppose that pi finishes before p¬i starts. Then pi reads only 0 from
r¬i , and cannot observe ri < r¬i : pi returns 0 in this case.
We now show that the two processes cannot return the same value.
Suppose that both processes terminate. Let i be such that pi reads r¬i
for the last time before p¬i reads ri for the last time. If pi returns 0,
then it observes ri ≥ r¬i + 2 at the time of its read; p¬i can increment
r¬i at most once before reading ri again, and so observed r¬i < ri and
returns 1.
Alternatively, if pi returns 1, it observed ri < r¬i . Since it performs
no more increments on ri , pi also observes ri < r¬i in all subsequent
reads, and so cannot also return 1.
2. Let’s run the protocol with an oblivious adversary, and track the value
of r0t − r1t over time, where rit is the value of ri after t writes (to either
register). Each write to r0 increases this value by 1/2 on average, with
a change of 0 or 1 equally likely, and each write to r1 decreases it by
1/2 on average.
To make things look symmetric, let ∆t be the change caused by the
t-th write and write ∆t as ct + X t where ct = ±1/2 is a constant
determined by whether p0 or p1 does the t-th write and X t = ±1/2 is
a random variable with expectation 0. Observe that the X t variables
are independent of each other and the constants ct (which depend only
on the schedule).
For the protocol to run forever, at every time t it must hold that
r0t − r1t ≤ 3; otherwise, even after one or both processes does its
0 0
next write, we will have r0t − r1t and the next process to read will
terminate. But
t
X
r0t − r1t = ∆s
s=1
Xt
= (cs + Xs )
s=1
Xt t
X
= cs + Xs .
s=1 s=1

The left-hand sum is a constant, while the right-hand sum has a

binomial distribution. For any fixed constant, the probability that a
binomial distribution lands within ±2 of the constant goes to zero in
APPENDIX D. SAMPLE ASSIGNMENTS FROM SPRING 2014 414

the limit as t → ∞, so with probability 1 there is some t for which this

event does not occur.

3. For an adaptive adversary, the following strategy prevents agreement:

(a) Run p0 until it is about to increment r0 .

(b) Run p1 until it is about to increment r1 .
(c) Allow both increments to proceed and repeat.

The effect is that both processes always observe r0 = r1 whenever

they do a read, and so never finish. This works because the adaptive
adversary can see the coin-flips done by the processes before they act
on them; it would not work with an oblivious adversary or in a model
that supported probabilistic writes.

D.7 CS465/CS565 Final Exam, May 2nd, 2014

Write your answers in the blue book(s). Justify your answers. Work alone.
Do not use any notes or books.
There are four problems on this exam, each worth 20 points, for a total
of 80 points. You have approximately three hours to complete this exam.

D.7.1 Maxima (20 points)

Some deterministic processes organized in an anonymous, synchronous ring
are each given an integer input (which may or may not be distinct from other
processes’ inputs), but otherwise run the same code and do not know the
size of the ring. We would like the processes to each compute the maximum
input. As usual, each process may only return an output once, and must do
so after a finite number of rounds, although it may continue to participate
in the protocol (say, by relaying messages) even after it returns an output.
Prove or disprove: It is possible to solve this problem in this model.

Solution
It’s not possible.
Consider an execution with n = 3 processes, each with input 0. If the
protocol is correct, then after some finite number of rounds t, each process
returns 0. By symmetry, the processes all have the same states and send the
same messages throughout this execution.
APPENDIX D. SAMPLE ASSIGNMENTS FROM SPRING 2014 415

Now consider a ring of size 2(t + 1) where every process has input 0,
except for one process p that has input 1. Let q be the process at maximum
distance from p. By induction on r, we can show that after r rounds of
communication, every process that is more than r + 1 hops away from p has
the same state as all of the processes in the 3-process execution above. So in
particular, after t rounds, process q (at distance t + 1) is in the same state
as it would be in the 3-process execution, and thus it returns 0. But—as it
learns to its horror, one round too late—the correct maximum is 1.

D.7.2 Historyless objects (20 points)

Recall that a shared-memory object is historyless if any operation on the
object either (a) always leaves the object in the same state as before the
operation, or (b) always leaves the object in a new state that doesn’t depend
on the state before the operation.
What is the maximum possible consensus number for a historyless object?
That is, for what value n is it possible to solve wait-free consensus for n
processes using some particular historyless object but not possible to solve
wait-free consensus for n + 1 processes using any historyless object?

Solution
Test-and-sets are (a) historyless, and (b) have consensus number 2, so n is
at least 2.
To show that no historyless object can solve wait-free 3-process consensus,
consider an execution that starts in a bivalent configuration and runs to a
configuration C with two pending operations x and y such that Cx is 0-valent
and Cy is 1-valent. By the usual arguments x and y must both be operations
on the same object. If either of x and y is a read operation, then (0-valent)
Cxy and (1-valent) Cyx are indistinguishable to a third process pz if run
alone, because the object is left in the same state in both configurations;
whichever way pz decides, it will give a contradiction in an execution starting
with one of these configurations. If neither of x and y is a read, then x
overwrites y, and Cx is indistinguishable from Cyxto pz if pz runs alone;
again we get a contradiction.

D.7.3 Hams (20 points)

Hamazon, LLC, claims to be the world’s biggest delivery service for canned
hams, with guaranteed delivery of a canned ham to your home anywhere
on Earth via suborbital trajectory from secret launch facilities at the North
APPENDIX D. SAMPLE ASSIGNMENTS FROM SPRING 2014 416

and South Poles. Unfortunately, these launch facilities may be subject to

crash failures due to inclement weather, trademark infringement actions, or
military retaliation for misdirected hams.
For this problem, you are to evaluate Hamazon’s business model from
the perspective of distributed algorithms. Consider a system consisting of
a client process and two server processes (corresponding to the North and
South Pole facilities) that communicate by means of asynchronous message
passing. In addition to the usual message-passing actions, each server also
has an irrevocable launch action that launches a ham at the client. As with
messages, hams are delivered asynchronously: it is impossible for the client
to tell if a ham has been launched until it arrives.
A ham protocol is correct provided (a) a client that orders no ham receives
no ham; and (b) a client that orders a ham receives exactly one ham. Show
that there can be no correct deterministic protocol for this problem if one of
the servers can crash.

Solution
Consider an execution in which the client orders ham. Run the northern
server together with the client until the server is about to issue a launch
action (if it never does so, the client receives no ham when the southern
server is faulty).
Now run the client together with the southern server. There are two
cases:

1. If the southern server ever issues launch, execute both this and the
northern server’s launch actions: the client gets two hams.

2. If the southern server never issues launch, never run the northern
server again: the client gets no hams.

In either case, the one-ham rule is violated, and the protocol is not
correct.5
5
It’s tempting to try to solve this problem by reduction from a known impossibility
result, like Two Generals or FLP. For these specific problems, direct reductions don’t
appear to work. Two Generals assumes message loss, but in this model, messages are not
lost. FLP needs any process to be able to fail, but in this model, the client never fails.
Indeed, we can solve consensus in the Hamazon model by just having the client transmit
its input to both servers.
APPENDIX D. SAMPLE ASSIGNMENTS FROM SPRING 2014 417

D.7.4 Mutexes (20 points)

A swap register s has an operation swap(s, v) that returns the argument to
the previous call to swap, or ⊥ if it is the first such operation applied to the
register. It’s easy to build a mutex from a swap register by treating it as a
test-and-set: to grab the mutex, I swap in 1, and if I get back ⊥ I win (and
otherwise try again); and to release the mutex, I put back ⊥.
Unfortunately, this implementation is not starvation-free: some other
process acquiring the mutex repeatedly might always snatch the ⊥ away just
before I try to swap it out. Algorithm D.5 uses a swap object s along with
an atomic register r to try to fix this.

1 procedure mutex()
2 predecessor ← swap(s, myId)
3 while r 6= predecessor do
4 try again
// Start of critical section
5 ...
// End of critical section
6 r ← myId
Algorithm D.5: Mutex using a swap object and register

Prove that Algorithm D.5 gives a starvation-free mutex, or give an

example of an execution where it fails. You should assume that s and r are
both initialized to ⊥.

Solution
Because processes use the same id if they try to access the mutex twice, the
algorithm doesn’t work.
Here’s an example of a bad execution:

1. Process 1 swaps 1 into s and gets ⊥, reads ⊥ from r, performs its

critical section, and writes 1 to r.

2. Process 2 swaps 2 into s and gets 1, reads 1 from r, and enters the
critical section.

3. Process 1 swaps 1 into s and gets 2, and spins waiting to see 2 in r.

APPENDIX D. SAMPLE ASSIGNMENTS FROM SPRING 2014 418

4. Process 3 swaps 3 into s and gets 1. Because r is still 1, process 3 reads

this 1 and enters the critical section. We now have two processes in
the critical section, violating mutual exclusion.

I believe this works if each process adopts a new id every time it calls
mutex, but the proof is a little tricky.6

6
The simplest proof I can come up with is to apply an invariant that says that (a)
the processes that have executed swap(s, myId) but have not yet left the while loop have
predecessor values that form a linked list, with the last pointer either equal to ⊥ (if no
process has yet entered the critical section) or the last process to enter the critical section;
(b) r is ⊥ if no process has yet left the critical section, or the last process to leave the
critical section otherwise; and (c) if there is a process that is in the critical section, its
predecessor field points to the last process to leave the critical section. Checking the effects
of each operation shows that this invariant is preserved through the execution, and (a)
combined with (c) show that we can’t have two processes in the critical section at the same
time. Additional work is still needed to show starvation-freedom. It’s a good thing this
algorithm doesn’t work as written.
Appendix E

Sample assignments from

Fall 2011

E.1 Assignment 1: due Wednesday, 2011-09-28, at

17:00
Bureaucratic part
Send me email! My address is [email protected].
In your message, include:

1. Your name.

2. Your status: whether you are an undergraduate, grad student, auditor,

etc.

3. Anything else you’d like to say.

(You will not be graded on the bureaucratic part, but you should do it
anyway.)

E.1.1 Anonymous algorithms on a torus

An n × m torus is a two-dimensional version of a ring, where a node at
position (i, j) has a neighbor to the north at (i, j − 1), the east at (i + 1, j),
the south at (i, j + 1), and the west at (i − 1, j). These values wrap around
modulo n for the first coordinate and modulo m for the second; so (0, 0) has
neighbors (0, m − 1), (1, 0), (0, 1), and (n − 1, 0).

419
APPENDIX E. SAMPLE ASSIGNMENTS FROM FALL 2011 420

Suppose that we have a synchronous message-passing network in the

form of an n × m torus, consisting of anonymous, identical processes that do
not know n, m, or their own coordinates, but do have a sense of direction
(meaning they can tell which of their neighbors is north, east, etc.).
Prove or disprove: Under these conditions, there is a deterministic1
algorithm that computes whether n > m.

Solution
Disproof: Consider two executions, one in an n × m torus and one in an
m × n torus where n > m and both n and m are at least 2.2 Using the same
argument as in Lemma 5.1.1, show by induction on the round number that,
for each round r, all processes in both executions have the same state. It
follows that if the processes correctly detect n > m in the n × m execution,
then they incorrectly report m > n in the m × n execution.

E.1.2 Clustering
Suppose that k of the nodes in an asynchronous message-passing network
are designated as cluster heads, and we want to have each node learn the
identity of the nearest head. Given the most efficient algorithm you can for
this problem, and compute its worst-case time and message complexities.
You may assume that processes have unique identifiers and that all
processes know how many neighbors they have.3

Solution
The simplest approach would be to run either of the efficient distributed
breadth-first search algorithms from Chapter 4 simultaneously starting at all
cluster heads, and have each process learn the distance to all cluster heads
at once and pick the nearest one. This gives O(D2 ) time and O(k(E + V D))
messages if we use layering and O(D) time and O(kDE) messages using
local synchronization.
We can get rid of the dependence on k in the local-synchronization
algorithm by running it almost unmodified, with the only difference being
the attachment of a cluster head id to the exactly messages. The simplest
way to show that the resulting algorithm works is to imagine coalescing
1
Clarification added 2011-09-28.
2
This last assumption is not strictly necessary, but it avoids having to worry about
what it means when a process sends a message to itself.
3
Clarification added 2011-09-26.
APPENDIX E. SAMPLE ASSIGNMENTS FROM FALL 2011 421

all cluster heads into a single initiator; the clustering algorithm effectively
simulates the original algorithm running in this modified graph, and the
same proof goes through. The running time is still O(D) and the message
complexity O(DE).

E.1.3 Negotiation
Two merchants A and B are colluding to fix the price of some valuable
commodity, by sending messages to each other for r rounds in a synchronous
message-passing system. To avoid the attention of antitrust regulators, the
merchants are transmitting their messages via carrier pigeons, which are
unreliable and may become lost. Each merchant has an initial price pA or
pB , which are integer values satisfying 0 ≤ p ≤ m for some known value
m, and their goal is to choose new prices p0A and p0B , where |p0A − p0B | ≤ 1.
If pA = pB and no messages are lost, they want the stronger goal that
p0A = p0B = pA = pB .
Prove the best lower bound you can on r, as a function of m, for all
protocols that achieve these goals.

Solution
This is a thinly-disguised version of the Two Generals Problem from Chap-
ter 8, with the agreement condition p0A = p0B replaced by an approximate
agreement condition |p0A − p0B | ≤ 1. We can use a proof based on the
indistinguishability argument in §8.2 to show that r ≥ m/2.
Fix r, and suppose that in a failure-free execution both processes send
messages in all rounds (we can easily modify an algorithm that does not
have this property to have it, without increasing r). We will start with a
sequence of executions with pA = pB = 0. Let X0 be the execution in which
no messages are lost, X1 the execution in which A’s last message is lost,
X2 the execution in which both A and B’s last messages are lost, and so
on, with Xk for 0 ≤ k ≤ 2r losing k messages split evenly between the two
processes, breaking ties in favor of losing messages from A.
When i is even, Xi is indistinguishable from Xi+1 by A; it follows that
p0A is the same in both executions. Because we no longer have agreement,
it may be that p0B (Xi ) and p0B (Xi+1 ) are not the same as p0A in either
execution; but since both are within 1 of p0A , the difference between them is
at most 2. Next, because Xi+1 to Xi+2 are indistinguishable to B, we have
p0B (Xi+1 ) = p0B (Xi+2 ), which we can combine with the previous claim to get
|p0B (Xi ) − p0B (Xi+2 )|. A simple induction then gives p0B (X2r ) ≤ 2r, where
APPENDIX E. SAMPLE ASSIGNMENTS FROM FALL 2011 422

X2r is an execution in which all messages are lost.

Now construct executions X2r+1 and X2r+2 by changing pA and pB to
m one at a time. Using essentially the same argument as before, we get
|p0B (X2r ) − p0B (X2r+2 )| ≤ 2 and thus p0B (X2r+2 ) ≤ 2r + 2.
Repeat the initial 2r steps backward to get to an execution X4r+2 with
pA = pB = m and no messages lost. Applying the same reasoning as above
shows m = p0B (X4r+2 ) ≤ 4r + 2 or r ≥ m−2 4 = Ω(m).
Though it is not needed for the solution, it is not too hard to unwind
the lower bound argument to extract an algorithm that matches the lower
bound up to a small constant factor. For simplicity, let’s assume m is even.
The protocol is to send my input in the first message and then use m/2−1
subsequent acknowledgments, stopping immediately if I ever fail to receive
a message in some round; the total number of rounds r is exactly m/2. If
I receive s messages in the first s rounds, I decide on min(pA , pB ) if that
value lies in [m/2 − s, m/2 + s] and the nearest endpoint otherwise. (Note
that if s = 0, I don’t need to compute min(pA , pB ), and if s > 0, I can do so
because I know both inputs.)
This satisfies the approximate agreement condition because if I see only s
messages, you see at most s + 1, because I stop sending once I miss a message.
So either we both decide min(pA , pB ) or we choose endpoints m/2 ± sA
and m/2 ± sB that are within 1 of each other. It also satisfies the validity
condition p0A = p0B = pA = pB when both inputs are equal and no messages
are lost (and even the stronger requirement that p0A = p0B when no messages
are lost), because in this case [m/2 − s, m/2 + s] is exactly [0, m] and both
processes decide min(pA , pB ).
There is still a factor-of-2 gap between the upper and lower bounds. My
guess would be that the correct bound is very close to m/2 on both sides,
and that my lower bound proof is not quite clever enough.

E.2 Assignment 2: due Wednesday, 2011-11-02, at

17:00
E.2.1 Consensus with delivery notifications
The FLP bound (Chapter 11) shows that we can’t solve consensus in an
asynchronous system with one crash failure. Part of the reason for this is
that only the recipient can detect when a message is delivered, so the other
processes can’t distinguish between a configuration in which a message has
or has not been delivered to a faulty process.
APPENDIX E. SAMPLE ASSIGNMENTS FROM FALL 2011 423

Suppose that we augment the system so that senders are notified imme-
diately when their messages are delivered. We can model this by making the
delivery of a single message an event that updates the state of both sender
and recipient, both of which may send additional messages in response. Let
us suppose that this includes attempted deliveries to faulty processes, so that
any non-faulty process that sends a message m is eventually notified that m
has been delivered (although it might not have any effect on the recipient if
the recipient has already crashed).

1. Show that this system can solve consensus with one faulty process
when n = 2.
2. Show that this system cannot solve consensus with two faulty processes
when n = 3.

Solution
1. To solve consensus, each process sends its input to the other. Whichever
input is delivered first becomes the output value for both processes.
2. To show impossibility with n = 3 and two faults, run the usual FLP
proof until we get to a configuration C with events e0 and e such that
Ce is 0-valent and Ce0 e is 1-valent (or vice versa). Observe that e
and e0 may involve two processes each (sender and receiver), for up
to four processes total, but only a process that is involved in both e
and e0 can tell which happened first. There can be at most two such
processes. Kill both, and get that Ce0 e is indistinguishable from Cee0
for the remaining process, giving the usual contradiction.

E.2.2 A circular failure detector

Suppose we equip processes 0 . . . n − 1 in an asynchronous message-passing
system with n processes subject to crash failures with a failure detector that
is strongly accurate (no non-faulty process is ever suspected) and causes
process i + 1 (mod n) to eventually permanently suspect process i if process
i crashes. Note that this failure detector is not even weakly complete (if
both i and i + 1 crash, no non-faulty process suspects i). Note also that the
ring structure of the failure detector doesn’t affect the actual network: even
though only process i + 1 (mod n) may suspect process i, any process can
send messages to any other process.
Prove the best upper and lower bounds you can on the largest number of
failures f that allows solving consensus in this system.
APPENDIX E. SAMPLE ASSIGNMENTS FROM FALL 2011 424

Solution
There is an easy reduction to FLP that shows f ≤ n/2 is necessary (when n
√
is even), and a harder reduction that shows f < 2 n − 1 is necessary. The
easy reduction is based on crashing every other process; now no surviving
process can suspect any other survivor, and we are back in an asynchronous
message-passing system with no failure detector and 1 remaining failure (if
f is at least n/2 + 1).
√
The harder reduction is to crash every ( n)-th process. This partitions
√ √
the ring into n segments of length n − 1 each, where there is no failure
detector in any segment that suspects any process in another segment. If an
algorithm exists that solves consensus in this situation, then it does so even
if (a) all processes in each segment have the same input, (b) if any process in
√
one segment crashes, all n − 1 process in the segment crash, and (c) if any
process in a segment takes a step, all take a step, in some fixed order. Under
this additional conditions, each segment can be simulated by a single process
√
in an asynchronous system with no failure detectors, and the extra n − 1
√
failures in 2 n − 1 correspond to one failure in the simulation. But we can’t
solve consensus in the simulating system (by FLP), so we can’t solve it in
the original system either.
On the other side, let’s first boost completeness of the failure detector,
by having any process that suspects another transmit this submission by
reliable broadcast. So now if any non-faulty process i suspects i + 1, all the
non-faulty processes will suspect i + 1. Now with up to t failures, whenever
I learn that process i is faulty (through a broadcast message passing on the
suspicion of the underlying failure detector, I will suspect processes i + 1
through i + t − f as well, where f is the number of failures I have heard
about directly. I don’t need to suspect process i + t − f + 1 (unless there is
some intermediate process that has also failed), because the only way that
this process will not be suspected eventually is if every process in the range
i to i + t − f is faulty, which can’t happen given the bound t.
Now if t is small enough that I can’t cover the entire ring with these
segments, then there is some non-faulty processes that is far enough away
from the nearest preceding faulty process that it is never suspected: this gives
us an eventually strong failure detector, and we can solve consensus using the
standard Chandra-Toueg ♦S algorithm from §13.4 or [CT96]. The inequality
I am looking for is f (t − f ) < n, where the √left-hand side is maximized by
setting 2
f = t/2, which gives t /4 < n or t < 2n. This leaves a gap of about
√
2 between the upper and lower bounds; I don’t know which one can be
improved.
APPENDIX E. SAMPLE ASSIGNMENTS FROM FALL 2011 425

√
I am indebted to Hao Pan for suggesting the Θ( n) upper and lower
bounds, which corrected an error in my original draft solution to this problem.

E.2.3 An odd problem

Suppose that each of n processes in a message-passing system with a complete
network is attached to a sensor. Each sensor has two states, active and
inactive; initially, all sensors are off. When the sensor changes state, the
corresponding process is notified immediately, and can update its state
and send messages to other processes in response to this event. It is also
guaranteed that if a sensor changes state, it does not change state again for
at least two time units. We would like to detect when an odd number of
sensors are active, by having at least one process update its state to set off
an alarm at a time when this condition holds.
A correct protocol for this problem should satisfy two conditions:
No false positives If a process sets of an alarm, then an odd number of
sensors are active.

Termination If at some time an odd number of sensors are active, and from
that point on no sensor changes its state, then some process eventually
sets off an alarm.
For what values of n is it possible to construct such a protocol?

Solution
It is feasible to solve the problem for n < 3.
For n = 1, the unique process sets off its alarm as soon as its sensor
becomes active.
For n = 2, have each process send a message to the other containing
its sensor state whenever the sensor state changes. Let s1 and s2 be the
state of the two process’s sensors, with 0 representing inactive and 1 active,
and let pi set off its alarm if it receives a message s such that s ⊕ si = 1.
This satisfies termination, because if we reach a configuration with an odd
number of active sensors, the last sensor to change causes a message to be
sent to the other process that will cause it to set off its alarm. It satisfies
no-false-positives, because if pi sets off its alarm, then s¬i = s because at
most one time unit has elapsed since p¬i sent s; it follows that s¬i ⊕ si = 1
and an odd number of sensors are active.
No such protocol is possible for n ≥ 3. Make p1 ’s sensor active. Run the
protocol until some process pi is about to enter an alarm state (this occurs
APPENDIX E. SAMPLE ASSIGNMENTS FROM FALL 2011 426

eventually because otherwise we violate termination). Let pj be one of p2

or p3 with j 6= i, activate pj ’s sensor (we can do this without violating the
once-per-time-unit restriction because it has never previously been activated)
and then let pi set off its alarm. We have now violated no-false-positives.

E.3 Assignment 3: due Friday, 2011-12-02, at 17:00

E.3.1 A restricted queue
Suppose you have an atomic queue Q that supports operations enq and deq,
restricted so that:

• enq(Q) always pushes the identity of the current process onto the tail
of the queue.

• deq(Q) tests if the queue is nonempty and its head is equal to the
identity of the current process. If so, it pops the head and returns
true. If not, it does nothing and returns false.

The rationale for these restrictions is that this is the minimal version of
a queue needed to implement a starvation-free mutex using Algorithm 17.2.
What is the consensus number of this object?

Solution
The restricted queue has consensus number 1.
Suppose we have 2 processes, and consider all pairs of operations on Q
that might get us out of a bivalent configuration C. Let x be an operation
carried out by p that leads to a b-valent state, and y an operation by q that
leads to a (¬b)-valent state. There are three cases:

• Two deq operations. If Q is empty, the operations commute. If the

head of the Q is p, then y is a no-op and p can’t distinguish between
Cx and Cyx. Similarly for q if the head is q.

• One enq and one deq operation. Suppose x is an enq and y a deq. If
Q is empty or the head is not q, then y is a no-op: p can’t distinguish
Cx from Cyx. If the head is q, then x and y commute. The same holds
in reverse if x is a deq and y an enq.

• Two enq operations. This is a little tricky, because Cxy and Cyx are
different states. However, if Q is nonempty in C, whichever process
APPENDIX E. SAMPLE ASSIGNMENTS FROM FALL 2011 427

isn’t at the head of Q can’t distinguish them, because any deq operation
returns false and never reaches the newly-enqueued values. This leaves
the case where Q is empty in C. Run p until it is poised to do
x0 = deq(Q) (if this never happens, p can’t distinguish Cxy from Cyx);
then run q until it is poised to do y 0 = deq(Q) as well (same argument
as for p). Now allow both deq operations to proceed in whichever order
causes them both to succeed. Since the processes can’t tell which deq
happened first, they can’t tell which enq happened first either. Slightly
more formally, if we let α be the sequence of operations leading up to
the two deq operations, we’ve just shown Cxyαx0 y 0 is indistinguishable
from Cyxαy 0 x0 to both processes.

In all cases, we find that we can’t escape bivalence. It follows that Q can’t
solve 2-process consensus.

E.3.2 Writable fetch-and-increment

Suppose you are given an unlimited supply of atomic registers and fetch-and-
increment objects, where the fetch-and-increment objects are all initialized to
0 and supply only a fetch-and-increment operation that increments the object
and returns the old value. Show how to use these objects to construct a wait-
free, linearizable implementation of an augmented fetch-and-increment that
also supports a write operation that sets the value of the fetch-and-increment
and returns nothing.

Solution
We’ll use a snapshot object a to control access to an infinite array f
of fetch-and-increments, where each time somebody writes to the imple-
mented object, we switch to a new fetch-and-increment. Each cell in a
holds (timestamp, base), where base is the starting value of the simulated
fetch-and-increment. We’ll also use an extra fetch-and-increment T to hand
out timestamps.
Code is in Algorithm E.1.
Since this is all straight-line code, it’s trivially wait-free.
Proof of linearizability is by grouping all operations by timestamp, us-
ing s[i].timestamp for FetchAndIncrement operations and t for write op-
erations, then putting write before FetchAndIncrement, then ordering
FetchAndIncrement by return value. Each group will consist of a write(v)
for some v followed by zero or more FetchAndIncrement operations, which
will return increasing values starting at v since they are just returning values
APPENDIX E. SAMPLE ASSIGNMENTS FROM FALL 2011 428

1 procedure FetchAndIncrement()
2 s ← snapshot(a)
3 i ← arg maxi (s[i].timestamp)
4 return f [s[i].timestamp] + s[i].base

5 procedure write(v)
6 t ← FetchAndIncrement(T )
7 a[myId] ← (t, v)
Algorithm E.1: Resettable fetch-and-increment

from the underlying FetchAndIncrement object; the implementation thus

meets the specification.
To show consistency with the actual execution order, observe that time-
stamps only increase over time and that the use of snapshot means that
any process that observes or writes a timestamp t does so at a time later
than any process that observes or writes any t0 < t; this shows the group
order is consistent. Within each group, the write writes a[myId] before
any FetchAndIncrement reads it, so again we have consistency between the
write and any FetchAndIncrement operations. The FetchAndIncrement
operations are linearized in the order in which they access the underlying
f [. . . ] object, so we win here too.

E.3.3 A box object

Suppose you want to implement an object representing a w × h box whose
width (w) and height (h) can be increased if needed. Initially, the box is
1 × 1, and the coordinates can be increased by 1 each using IncWidth and
IncHeight operations. There is also a GetArea operation that returns the
area w · h of the box.
Give an obstruction-free deterministic implementation of this object from
atomic registers that optimizes the worst-case individual step complexity of
GetArea, and show that your implementation is optimal by this measure up
to constant factors.

Solution
Let b be the box object. Represent b by a snapshot object a, where a[i] holds
a pair (∆wi , ∆hi ) representing the number of times process i has executed
IncWidth and IncHeight; these operations simply increment the appropriate
APPENDIX E. SAMPLE ASSIGNMENTS FROM FALL 2011 429

value and update the snapshot object. Let GetArea take a snapshot and
P P
return ( i ∆wi ) ( i ∆hi ); the cost of the snapshot is O(n).
To see that this is optimal, observe that we can use IncWidth and GetArea
to represent inc and read for a standard counter. The Jayanti-Tan-Toueg
bound applies to counters, giving a worst-case cost of Ω(n) for GetArea.

E.4 CS465/CS565 Final Exam, December 12th, 2011

Write your answers in the blue book(s). Justify your answers. Work alone.
Do not use any notes or books.
There are four problems on this exam, each worth 20 points, for a total
of 80 points. You have approximately three hours to complete this exam.

General clarifications added during exam Assume all processes have

unique ids and know n. Assume that the network is complete in the message-
passing model.

E.4.1 Lockable registers (20 points)

Most memory-management units provide the ability to control access to
specific memory pages, allowing a page to be marked (for example) read-only.
Suppose that we model this by a lockable register that has the usual
register operations read(r) and write(r, v) plus an additional operation
lock(r). The behavior of the register is just like a normal atomic register
until somebody calls lock(r); after this, any call to write(r) has no effect.
What is the consensus number of this object?

Solution
The consensus number is ∞; a single lockable register solves consensus for
any number of processes. Code is in Algorithm E.2.

1 write(r, input)
2 lock(r)
3 return read(r)
Algorithm E.2: Consensus using a lockable register

Termination and validity are trivial. Agreement follows from the fact
that whatever value is in r when lock(r) is first called will never change,
and thus will be read and returned by all processes.
APPENDIX E. SAMPLE ASSIGNMENTS FROM FALL 2011 430

E.4.2 Byzantine timestamps (20 points)

Suppose you have an asynchronous message passing system with exactly one
Byzantine process.
You would like the non-faulty processes to be able to acquire an increasing
sequence of timestamps. A process should be able to execute the timestamp
protocol as often as it likes, and it should be guaranteed that when a process
is non-faulty, it eventually obtains a timestamp that is larger than any
timestamp returned in any execution of the protocol by a non-faulty process
that finishes before the current process’s execution started.
Note that there is no bound on the size of a timestamp, so having the
Byzantine process run up the timestamp values is not a problem, as long as
it can’t cause the timestamps to go down.
For what values of n is it possible to solve this problem?

Solution
It is possible to solve the problem for all n except n = 3. For n = 1, there are
no non-faulty processes, so the specification is satisfied trivially. For n = 2,
there is only one non-faulty process: it can just keep its own counter and
return an increasing sequence of timestamps without talking to the other
process at all.
For n = 3, it is not possible. Consider an execution in which messages
between non-faulty processes p and q are delayed indefinitely. If the Byzantine
process r acts to each of p and q as it would if the other had crashed, this
execution is indistinguishable to p and q from an execution in which r is
correct and the other is faulty. Since there is no communication between
p and q, it is easy to construct and execution in which the specification is
violated.
For n ≥ 4, the protocol given in Algorithm E.3 works.
The idea is similar to the Attiya, Bar-Noy, Dolev distributed shared
memory algorithm [ABND95]. A process that needs a timestamp polls n − 1
other processes for the maximum values they’ve seen and adds 1 to it; before
returning, it sends the new timestamp to all other processes and waits to
receive n − 1 acknowledgments. The Byzantine process may choose not to
answer, but this is not enough to block completion of the protocol.
To show the timestamps are increasing, observe that after the completion
of any call by i to getTimestamp, at least n − 2 non-faulty processes j have
a value vj ≥ vi . Any call to getTimestamp that starts later sees at least
n − 3 > 0 of these values, and so computes a max that is at least as big as
APPENDIX E. SAMPLE ASSIGNMENTS FROM FALL 2011 431

1 procedure getTimestamp()
2 ci ← ci + 1
3 send probe(ci ) to all processes
4 wait to receive response(ci , vj ) from n − 1 processes
5 vi ← (maxj vj ) + 1
6 send newTimestamp(ci , vi ) to all processes
7 wait to receive ack(ci ) from n − 1 processes
8 return vi

9 upon receiving probe(cj ) from j do

10 send response(cj , vi ) to j

11 upon receiving newTimestamp(cj , vj ) from j do

12 vi ← max(vi , vj )
13 send ack(cj ) to j
Algorithm E.3: Timestamps with n ≥ 3 and one Byzantine process

vi and then adds 1 to it, giving a larger value.

E.4.3 Failure detectors and k-set agreement (20 points)

Recall that in the k-set agreement problem we want each of n processes to
choose a decision value, with the property that the set of decision values has at
most k distinct elements. It is known that k-set agreement cannot be solved
deterministically in an asynchronous message-passing or shared-memory
system with k or more crash failures.
Suppose that you are working in an asynchronous message-passing system
with an eventually strong (♦S) failure detector. Is it possible to solve k-set
agreement deterministically with f crash failures, when k ≤ f < n/2?

Solution
Yes. With f < n/2 and ♦S, we can solve consensus using Chandra-
Toueg [CT96]. Since this gives a unique decision value, it solves k-set
agreement for any k ≥ 1.
APPENDIX E. SAMPLE ASSIGNMENTS FROM FALL 2011 432

E.4.4 A set data structure (20 points)

Consider a data structure that represents a set S, with an operation add(S, x)
that adds x to S by setting S ← S ∪ {x}), and an operation size(S) that
returns the number of distinct4 elements |S| of S. There are no restrictions
on the types or sizes of elements that can be added to the set.
Show that any deterministic wait-free implementation of this object from
atomic registers has individual step complexity Ω(n) for some operation in
the worst case.

Solution
Algorithm E.4 implements a counter from a set object, where the counter
read consists of a single call to size(S). The idea is that each increment is
implemented by inserting a new element into S, so |S| is always equal to the
number of increments.

1 procedure inc(S)
2 nonce ← nonce + 1
3 add(S, hmyId, noncei).

4 procedure read(S)
5 return size(S)
Algorithm E.4: Counter from set object

Since the Jayanti-Tan-Toueg lower bound [JTT00] gives a lower bound

of Ω(n) on the worst-case cost of a counter read, there exists an execution in
which size(S) takes Ω(n) steps.
(We could also apply JTT directly by showing that the set object is
perturbable; this follows because adding an element not added by anybody
else is always visible to the reader.)

4
Clarification added during exam.
Appendix F

Additional sample final

exams

This appendix contains final exams from previous times the course was
offered, and is intended to give a rough guide to the typical format and
content of a final exam. Note that the topics covered in past years were not
necessarily the same as those covered this year.

F.1 CS425/CS525 Final Exam, December 15th, 2005

Write your answers in the blue book(s). Justify your answers. Work alone.
Do not use any notes or books.
There are three problems on this exam, each worth 20 points, for a total
of 60 points. You have approximately three hours to complete this exam.

F.1.1 Consensus by attrition (20 points)

Suppose you are given a bounded fetch-and-subtract register that holds
a non-negative integer value and supports an operation fetch-and-subtract(k)
for each k > 0 that (a) sets the value of the register to the previous value
minus k, or zero if this result would be negative, and (b) returns the previous
value of the register.
Determine the consensus number of bounded fetch-and-subtract under
the assumptions that you can use arbitrarily many such objects, that you can
supplement them with arbitrarily many multiwriter/multireader read/write
registers, that you can initialize all registers of both types to initial values of

433
APPENDIX F. ADDITIONAL SAMPLE FINAL EXAMS 434

your choosing, and that the design of the consensus protocol can depend on
the number of processes N .

Solution
The consensus number is 2.
To implement 2-process wait-free consensus, use a single fetch-and-
subtract register initialized to 1 plus two auxiliary read/write registers
to hold the input values of the processes. Each process writes its input to its
own register, then performs a fetch-and-subtract(1) on the fetch-and-subtract
register. Whichever process gets 1 from the fetch-and-subtract returns its
own input; the other process (which gets 0) returns the winning process’s
input (which it can read from the winning process’s read/write register.)
To show that the consensus number is at most 2, observe that any two
fetch-and-subtract operations commute: starting from state x, after fetch-and-
subtract(k1 ) and fetch-and-subtract(k2 ) the value in the fetch-and-subtract
register is max(0, x − k1 − k2 ) regardless of the order of the operations.

F.1.2 Long-distance agreement (20 points)

Consider an asynchronous message-passing model consisting of N processes
p1 . . . pN arranged in a line, so that each process i can send messages only to
processes i − 1 and i + 1 (if they exist). Assume that there are no failures,
that local computation takes zero time, and that every message is delivered
at most 1 time unit after it is sent no matter how many messages are sent
on the same edge.
Now suppose that we wish to solve agreement in this model, where
the agreement protocol is triggered by a local input event at one or more
processes and it terminates when every process executes a local decide event.
As with all agreement problems, we want Agreement (all processes decide
the same value), Termination (all processes eventually decide), and Validity
(the common decision value previously appeared in some input). We also
want no false starts: the first action of any process should either be an input
action or the receipt of a message.
Define the time cost of a protocol for this problem as the worst-case
time between the first input event and the last decide event. Give the best
upper and lower bounds you can on this time as function of N . Your upper
and lower bounds should be exact: using no asymptotic notation or hidden
constant factors. Ideally, they should also be equal.
APPENDIX F. ADDITIONAL SAMPLE FINAL EXAMS 435

Solution
Upper bound
Because there are no failures, we can appoint a leader and have it decide.
The natural choice is some process near the middle, say pb(N +1)/2c . Upon
receiving an input, either directly through an input event or indirectly from
another process, the process sends the input value along the line toward the
leader. The leader takes the first input it receives and broadcasts it back out
in both directions as the decision value. The worst case is when the protocol
is initiated at pN ; then we pay 2(N − b(N + 1)/2c) time to send all messages
out and back, which is N time units when N is even and N − 1 time units
when N is odd.

Lower bound
Proving an almost-matching lower bound of N − 1 time units is trivial: if
p1 is the only initiator and it starts at time t0 , then by an easy induction
argument,in the worst case pi doesn’t learn of any input until time t0 + (i − 1),
and in particular pN doesn’t find out until after N − 1 time units. If pN
nonetheless decides early, its decision value will violate validity in some
executions.
But we can actually prove something stronger than this: that N time
units are indeed required when N is odd. Consider two slow executions Ξ0
and Ξ1 , where (a) all messages are delivered after exactly one time unit in
each execution; (b) in Ξ0 only p1 receives an input and the input is 0; and (c)
in Ξ1 only pN receives an input and the input is 1. For each of the executions,
construct a causal ordering on events in the usual fashion: a send is ordered
before a receive, two events of the same process are ordered by time, and
other events are partially ordered by the transitive closure of this relation.
Now consider for Ξ0 the set of all events that precede the decide(0) event
of p1 and for Ξ1 the set of all events that precede the decide(1) event of
pN . Consider further the sets of processes S0 and S1 at which these events
occur; if these two sets of processes do not overlap, then we can construct
an execution in which both sets of events occur, violating Agreement.
Because S0 and S1 overlap, we must have |S0 | + |S1 | ≥ N + 1, and so
at least one of the two sets has size at least d(N + 1)/2e, which is N/2 + 1
when N is even. Suppose that it is S0 . Then in order for any event to occur
at pN/2+1 at all some sequence of messages must travel from the initial input
to p1 to process pN/2+1 (taking N/2 time units), and the causal ordering
implies that an additional sequence of messages travels back from pN/2+1 to
APPENDIX F. ADDITIONAL SAMPLE FINAL EXAMS 436

p1 before p1 decides (taking and additional N/2 time units). The total time
is thus N .

F.1.3 Mutex appendages (20 points)

An append register supports standard read operations plus an append
operation that appends its argument to the list of values already in the
register. An append-and-fetch register is similar to an append register,
except that it returns the value in the register after performing the append
operation. Suppose that you have an failure-free asynchronous system with
anonymous deterministic processes (i.e., deterministic processes that all run
exactly the same code). Prove or disprove each of the following statements:

1. It is possible to solve mutual exclusion using only append registers.

2. It is possible to solve mutual exclusion using only append-and-fetch

registers.

In either case, the solution should work for arbitrarily many processes—solving
mutual exclusion when N = 1 is not interesting. You are also not required
in either case to guarantee lockout-freedom.

Clarification given during exam

1. If it helps, you may assume that the processes know N . (It probably
doesn’t help.)

Solution
1. Disproof: With append registers only, it is not possible to solve mutual
exclusion. To prove this, construct a failure-free execution in which
the processes never break symmetry. In the initial configuration, all
processes have the same state and thus execute either the same read
operation or the same append operation; in either case we let all N
operations occur in some arbitrary order. If the operations are all
reads, all processes read the same value and move to the same new
state. If the operations are all appends, then no values are returned and
again all processes enter the same new state. (It’s also the case that
the processes can’t tell from the register’s state which of the identical
append operations went first, but we don’t actually need to use this
fact.)
APPENDIX F. ADDITIONAL SAMPLE FINAL EXAMS 437

Since we get a fair failure-free execution where all processes move

through the same sequence of states, if any process decides it’s in its
critical section, all do. We thus can’t solve mutual exclusion in this
model.

2. Since the processes are anonymous, any solution that depends on them
having identifiers isn’t going to work. But there is a simple solution
that requires only appending single bits to the register.
Each process trying to enter a critical section repeatedly executes an
append-and-fetch operation with argument 0; if the append-and-fetch
operation returns either a list consisting only of a single 0 or a list
whose second-to-last element is 1, the process enters its critical section.
To leave the critical section, the process does append-and-fetch(1).

F.2 CS425/CS525 Final Exam, May 8th, 2008

Write your answers in the blue book(s). Justify your answers. Work alone.
Do not use any notes or books.
There are four problems on this exam, each worth 20 points, for a total
of 80 points. You have approximately three hours to complete this exam.

F.2.1 Message passing without failures (20 points)

Suppose you have an asynchronous message-passing system with a complete
communication graph, unique node identities, and no failures. Show that any
deterministic atomic shared-memory object can be simulated in this model,
or give an example of a shared-memory object that can’t be simulated.

Solution
Pick some leader node to implement the object. To execute an operation,
send the operation to the leader node, then have the leader carry out the
operation (sequentially) on its copy of the object and send the results back.

F.2.2 A ring buffer (20 points)

Suppose you are given a ring buffer object that consists of k ≥ 1 memory
locations a[0] . . . a[k − 1] with an atomic shift-and-fetch operation that takes
an argument v and (a) shifts v into the buffer, so that a[i] ← a[i + 1] for
APPENDIX F. ADDITIONAL SAMPLE FINAL EXAMS 438

each i less than k − 1 and a[k − 1] ← v; and (b) returns a snapshot of the
new contents of the array (after the shift).
What is the consensus number of this object as a function of k?

Solution
We can clearly solve consensus for at least k processes: each process calls
shift-and-fetch on its input, and returns the first non-null value in the buffer.
So now we want to show that we can’t solve consensus for k + 1 processes.
Apply the usual FLP-style argument to get to a bivalent configuration C
where each of the k + 1 processes has a pending operation that leads to a
univalent configuration. Let e0 and e1 be particular operations leading to
0-valent and 1-valent configurations, respectively, and let e2 . . . ek be the
remaining k − 1 pending operations.
We need to argue first that no two distinct operations ei and ej are
operations of different objects. Suppose that Cei is 0-valent and Cej is
1-valent; then if ei and ej are on different objects, Cei ej (still 0-valent) is
indistinguishable by all processes from Cej ei (still 1-valent), a contradiction.
Alternatively, if ei and ej are both b-valent, there exists some (1−b)-valent ek
such that ei and ej both operate on the same object as ek , by the preceding
argument. So all of e0 . . . ek are operations on the same object.
By the usual argument we know that this object can’t be a register. Let’s
show it can’t be a ring buffer either. Consider the configurations Ce0 e1 . . . ek
and Ce1 . . . ek . These are indistinguishable to the process carrying out ek
(because its sees only the inputs to e1 through ek in its snapshot). So they
must have the same valence, a contradiction.
It follows that the consensus number of a k-element ring buffer is exactly
k.

F.2.3 Leader election on a torus (20 points)

An n × n torus is a graph consisting of n2 nodes, where each node (i, j),
0 ≤ i, j ≤ n − 1, is connected to nodes (i − 1, j), (i + 1, j), (i, j − 1), and
(i, j + 1), where all computation is done mod n.
Suppose you have an asynchronous message-passing system with a com-
munication graph in the form of an n × n torus. Suppose further that each
node has a unique identifier (some large natural number) but doesn’t know
the value of n. Give an algorithm for leader election in this model with the
best message complexity you can come up with.
APPENDIX F. ADDITIONAL SAMPLE FINAL EXAMS 439

Solution
First observe that each row and column of the torus is a bidirectional ring, so
we can run e.g. Hirschbirg and Sinclair’s O(n log n)-message protocol within
each of these rings to find the smallest identifier in the ring. We’ll use this
to construct the following algorithm:
1. Run Hirschbirg-Sinclair in each row to get a local leader for each row;
this takes n × O(n log n) = O(n2 log n) messages. Use an additional n
messages per row to distribute the identifier for the row leader to all
nodes and initiate the next stage of the protocol.
2. Run Hirschbirg-Sinclair in each column with each node adopting the row
leader identifier as its own. This costs another O(n2 log n) messages;
at the end, every node knows the minimum identifier of all nodes in
the torus.
The total message complexity is O(n2 log n). (I suspect this is optimal,
but I don’t have a proof.)

F.2.4 An overlay network (20 points)

A collection of n nodes—in an asynchronous message-passing system with a
connected, bidirectional communications graph with O(1) links per node—wish
to engage in some strictly legitimate file-sharing. Each node starts with some
input pair (k, v), where k is a key and v is a value, and the search problem
is to find the value v corresponding to a particular key k.
1. Suppose that we can’t do any preparation ahead of time. Give an
algorithm for searching with the smallest asymptotic worst-case message
complexity you can find as a function of n. You may assume that there
are no limits on time complexity, message size, or storage space at each
node.
2. Suppose now that some designated leader node can initiate a protocol
ahead of time to pre-process the data in the nodes before any query is
initiated. Give a pre-processing algorithm (that does not depend on
which key is eventually searched for) and associated search algorithm
such that the search algorithm minimizes the asymptotic worst-case
message complexity. Here you may assume that there are no limits on
time complexity, message size, or storage space for either algorithm, and
that you don’t care about the message complexity of the pre-processing
algorithm.
APPENDIX F. ADDITIONAL SAMPLE FINAL EXAMS 440

3. Give the best lower bound you can on the total message complexity of
the pre-processing and search algorithms in the case above.

Solution
1. Run depth-first search to find the matching key and return the corre-
sponding value back up the tree. Message complexity is O(|E|) = O(n)
(since each node has only O(1) links).

2. Basic idea: give each node a copy of all key-value pairs, then searches
take zero messages. To give each node a copy of all key-value pairs we
could do convergecast followed by broadcast (O(n) message complexity)
or just flood each pair O(n2 ). Either is fine since we don’t care about
the message complexity of the pre-processing stage.

3. Suppose the total message complexity of both the pre-processing stage

and the search protocol is less than n − 1. Then there is some node
other than the initiator of the search that sends no messages at any
time during the protocol. If this is the node with the matching key-
value pair, we don’t find it. It follows that any solution to the search
problem. requires a total of Ω(n) messages in the pre-processing and
search protocols.

F.3 CS425/CS525 Final Exam, May 10th, 2010

Write your answers in the blue book(s). Justify your answers. Work alone.
Do not use any notes or books.
There are four problems on this exam, each worth 20 points, for a total
of 80 points. You have approximately three hours to complete this exam.

F.3.1 Anti-consensus (20 points)

A wait-free anti-consensus protocol satisfies the conditions:

Wait-free termination Every process decides in a bounded number of its

own steps.

Non-triviality There is at least one process that decides different values

in different executions.

Disagreement If at least two processes decide, then some processes decide

on different values.
APPENDIX F. ADDITIONAL SAMPLE FINAL EXAMS 441

Show that there is no deterministic wait-free anti-consensus protocol

using only atomic registers for two processes and two possible output values,
but there is one for three processes and three possible output values.

Clarification: You should assume processes have distinct identities.

Solution
No protocol for two: turn an anti-consensus protocol with outputs in {0, 1}
into a consensus protocol by having one of the processes always negate its
output.
A protocol for three: Use a splitter.

F.3.2 Odd or even (20 points)

Suppose you have a protocol for a synchronous message-passing ring that is
anonymous (all processes run the same code) and uniform (this code is the
same for rings of different sizes). Suppose also that the processes are given
inputs marking some, but not all, of them as leaders. Give an algorithm
for determining if the size of the ring is odd or even, or show that no such
algorithm is possible.

Clarification: Assume a bidirectional, oriented ring and a deterministic

algorithm.

Solution
Here is an impossibility proof. Suppose there is such an algorithm, and let
it correctly decide “odd” on a ring of size 2k + 1 for some k and some set
of leader inputs. Now construct a ring of size 4k + 2 by pasting two such
rings together (assigning the same values to the leader bits in each copy)
and run the algorithm on this ring. By the usual symmetry argument, every
corresponding process sends the same messages and makes the same decisions
in both rings, implying that the processes incorrectly decide the ring of size
4k + 2 is odd.

F.3.3 Atomic snapshot arrays using message-passing (20 points)

Consider the following variant of Attiya-Bar-Noy-Dolev for obtaining snap-
shots of an array instead of individual register values, in an asynchronous
message-passing system with t < n/4 crash failures. The data structure we
APPENDIX F. ADDITIONAL SAMPLE FINAL EXAMS 442

are simulating is an array a consisting of an atomic register a[i] for each

process i, with the ability to perform atomic snapshots.
Values are written by sending a set of hi, v, ti i values to all processes,
where i specifies the segment a[i] of the array to write, v gives a value for
this segment, and ti is an increasing timestamp used to indicate more recent
values. We use a set of values because (as in ABD) some values may be
obtained indirectly.
To update segment a[i] with value v, process i generates a new timestamp
ti , sends {hi, v, ti i} to all processes, and waits for acknowledgments from at
least 3n/4 processes.
Upon receiving a message containing one or more hi, v, ti i triples, a process
updates its copy of a[i] for any i with a higher timestamp than previously
seen, and responds with an acknowledgment (we’ll assume use of nonces so
that it’s unambiguous which message is being acknowledged).
To perform a snapshot, a process sends snapshot to all processes, and
waits to receive responses from at least 3n/4 processes, which will consist of
the most recent values of each a[i] known by each of these processes together
with their timestamps (it’s a set of triples as above). The snapshot process
then takes the most recent versions of a[i] for each of these responses and
updates its own copy, then sends its entire snapshot vector to all processes
and waits to receive at least 3n/4 acknowledgments. When it has received
these acknowledgments, it returns its own copy of a[i] for all i.
Prove or disprove: The above procedure implements an atomic snapshot
array in an asynchronous message-passing system with t < n/4 crash failures.

Solution
Disproof: Let s1 and s2 be processes carrying out snapshots and let w1 and
w2 be processes carrying out writes. Suppose that each wi initiates a write
of 1 to a[wi ], but all of its messages to other processes are delayed after it
updates its own copy awi [wi ]. Now let each si receive responses from 3n/4 − 1
processes not otherwise mentioned plus wi . Then s1 will return a vector
with a[w1 ] = 1 and a[w2 ] = 0 while s2 will return a vector with a[w1 ] = 0
and a[w2 ] = 1, which is inconsistent. The fact that these vectors are also
disseminated throughout at least 3n/4 other processes is a red herring.

F.3.4 Priority queues (20 points)

Let Q be a priority queue whose states are multisets of natural numbers and
that has operations enq(v) and deq(), where enq(p) adds a new value v to
APPENDIX F. ADDITIONAL SAMPLE FINAL EXAMS 443

the queue, and deq() removes and returns the smallest value in the queue,
or returns null if the queue is empty. (If there is more than one copy of the
smallest value, only one copy is removed.)
What is the consensus number of this object?

Solution
The consensus number is 2. The proof is similar to that for a queue.
To show we can do consensus for n = 2, start with a priority queue with
a single value in it, and have each process attempt to dequeue this value. If
a process gets the value, it decides on its own input; if it gets null, it decides
on the other process’s input.
To show we can’t do consensus for n = 3, observe first that starting from
any states C of the queue, given any two operations x and y that are both
enqueues or both dequeues, the states Cxy and Cyx are identical. This
means that a third process can’t tell which operation went first, meaning
that a pair of enqueues or a pair of dequeues can’t get us out of a bivalent
configuration in the FLP argument. We can also exclude any split involving
two operations on different queues (or other objects) But we still need to
consider the case of a dequeue operation d and an enqueue operation e on
the same queue Q. This splits into several subcases, depending on the state
C of the queue in some bivalent configuration:
1. C = {}. Then Ced = Cd = {}, and a third process can’t tell which of
d or e went first.
2. C is nonempty and e = enq(v), where v is greater than or equal to the
smallest value in C. Then Cde and Ced are identical, and no third
process can tell which of d or e went first.
3. C is nonempty and e = enq(v), where v is less than any value in C.
Consider the configurations Ced and Cde. Here the process pd that
performs d can tell which operation went first, because it either obtains
v or some other value v 0 6= v. Kill this process. No other process in Ced
or Cde can distinguish the two states without dequeuing whichever of
v or v 0 was not dequeued by pd . So consider two parallel executions
Cedσ and Cdeσ where σ consists of an arbitrary sequence of operations
ending with a deq on Q by some process p (if no process ever attempts
to dequeue from Q, then we have already won, since the survivors can’t
distinguish Ced from Cde). Now the state of all objects is the same
after Cedσ and Cdeσ, and only pd and p have different states in these
two configurations. So any third process is out of luck.
Appendix G

I/O automata

G.1 Low-level view: I/O automata

An I/O automaton [LT87] is an automaton where transitions are labeled
by actions, which come in three classes: input actions, triggered by the
outside world; output actions triggered by the automaton and visible
to the outside world; and internal actions, triggered by the automaton
but not visible to the outside world. These classes correspond to inputs,
outputs, and internal computation steps of the automaton; the latter are
provided mostly to give merged input/output actions a place to go when
automata are composed together. A transition relation trans(A) relates
states(A) × acts(A) × states(A); if (s, a, s0 ) is in trans(A), it means that
A can move from state s to state s0 by executing action a.
There is also an equivalence relation task(A) on the output and internal
actions, which is used for enforcing fairness conditions—the basic idea is that
in a fair execution some action in each equivalence class must be executed
eventually (a more accurate definition will be given below).
The I/O automaton model carries with it a lot of specialized jargon.
We’ll try to avoid it as much as possible. One thing that will be difficult to
avoid in reading [Lyn96] is the notion of a signature, which is just the tuple
sig(A) = (in(A), out(A), int(A)) describing the actions of an automaton
A.

G.1.1 Enabled actions

An action a is enabled in some state s if trans(A) contains at least one
transition (s, a, s0 ). Input actions are always enabled—this is a requirement of
the model. Output and internal actions—the “locally controlled” actions—are

444
APPENDIX G. I/O AUTOMATA 445

not subject to this restriction. A state s is quiescent if only input actions

are enabled in s.

G.1.2 Executions, fairness, and traces

An execution of A is a sequence s0 a0 s1 a1 . . . where each triple (si , ai si+1 )
is in trans(A). Executions may be finite or infinite; if finite, they must end
in a state.
A trace of A is a subsequence of some execution consisting precisely
of the external (i.e., input and output) actions, with states and internal
actions omitted. If we don’t want to get into the guts of a particular
I/O automaton—and we usually don’t, unless we can’t help it because we
have to think explicitly about states for some reason—we can describe its
externally-visible behavior by just giving its set of traces.

G.1.3 Composition of automata

Composing a set of I/O automata yields a new super-automaton whose state
set is the Cartesian product of the state sets of its components and whose
action set is the union of the action sets of its components. A transition
with a given action a updates the states of all components that have a as an
action and has no effect on the states of other components. The classification
of actions into the three classes is used to enforce some simple compatibility
rules on the component automata; in particular:

1. An internal action of a component is never an action of another com-

ponent—internal actions are completely invisible.

2. No output action of a component can be an output action of another

component.

3. No action is shared by infinitely many components.1 In practice

this means that no action can be an input action of infinitely many
components, since the preceding rules mean that any action is an
output or internal action of at most one component.

All output actions of the components are also output actions of the
composition. An input action of a component is an input of the composition
only if some other component doesn’t supply it as an output; in this case
1
Note that infinite (but countable) compositions are permitted.
APPENDIX G. I/O AUTOMATA 446

it becomes an output action of the composition. Internal actions remain

internal (and largely useless, except for bookkeeping purposes).
The task equivalence relation is the union of the task relations for the
components: this turns out to give a genuine equivalence relation on output
and internal actions precisely because the first two compatibility rules hold.
Given an execution or trace X of a composite automaton that includes
A, we can construct the corresponding execution or trace X|A of A which
just includes the states of A and the actions visible to A (events that don’t
change the state of A drop out). The definition of composition is chosen so
that X|A is in fact an execution/trace of A whenever X is.

G.1.4 Hiding actions

Composing A and B continues to expose the outputs of A even if they line
up with inputs of B. While this may sometimes be desirable, often we want
to shove such internal communication under the rug. The model lets us do
this by redefining the signature of an automaton to make some or all of the
output actions into internal actions.

G.1.5 Fairness
I/O automata come with a built-in definition of fair executions, where an
execution of A is fair if, for each equivalence class C of actions in task(A),

1. the execution is finite and no action in C is enabled in the final state,

or

2. the execution is infinite and there are infinitely many occurrences of

actions in C, or

3. the execution is infinite and there are infinitely many states in which
no action in C is enabled.

If we think of C as corresponding to some thread or process, this says

that C gets infinitely many chances to do something in an infinite execution,
but may not actually do them if it gives ups and stops waiting (the third
case). The finite case essentially says that a finite execution isn’t fair unless
nobody is waiting at the end. The motivation for this particular definition
is that it guarantees (a) that any finite execution can be extended to a fair
execution and (b) that the restriction X|A of a fair execution or trace X is
also fair.
APPENDIX G. I/O AUTOMATA 447

Fairness is useful e.g. for guaranteeing message delivery in a message-

passing system: make each message-delivery action its own task class and each
message will eventually be delivered; similarly make each message-sending
action its own task class and a process will eventually send every message
it intends to send. Tweaking the task classes can allow for possibilities of
starvation, e.g. if all message-delivery actions are equivalent then a spammer
can shut down the system in a “fair” execution where only his (infinitely
many) messages are delivered.

G.1.6 Specifying an automaton

The typical approach is to write down preconditions and effects for each
action (for input actions, the preconditions are empty). An example would
be the spambot in Algorithm G.1.

1 input action setMessage(m)

2 effects
3 state ← m

4 output action spam(m)

5 precondition
6 spam = m
7 effects
8 none (keep spamming)

Algorithm G.1: Spambot as an I/O automaton

(Plus an initial state, e.g. state = ⊥, where ⊥ is not a possible message,

and a task partition, of which we will speak more below when we talk about
liveness properties.)

G.2 High-level view: traces

When studying the behavior of a system, traces are what we really care
about, and we want to avoid talking about states as much as possible. So
what we’ll aim to do is to get rid of the states early by computing the set
of traces (or fair traces) of each automaton in our system, then compose
traces to get traces for the system as a whole. Our typical goal will be to
show that the resulting set of traces has some desirable properties, usually
of the form (1) nothing bad happens (a safety property); (2) something
APPENDIX G. I/O AUTOMATA 448

good eventually happens (a liveness property); or (3) the horribly complex

composite automaton representing this concrete system acts just like that
nice clean automaton representing a specification (a simulation).
Very formally, a trace property specifies both the signature of the
automaton and a set of traces, such that all traces (or perhaps fair traces) of
the automata appear in the set. We’ll usually forget about the first part.
Tricky detail: It’s OK if not all traces in P are generated by A (we want
trace(A) ⊆ P , but not necessarily trace(A) = P ). But trace(A) will be
pretty big (it includes, for example, all finite sequences of input actions)
so hopefully the fact that A has to do something with inputs will tell us
something useful.

G.2.1 Example
A property we might demand of the spambot above (or some other abstraction
of a message channel) is that it only delivers messages that have previously
been given to it. As a trace property this says that in any trace t, if
tk = spam(m), then tj = setMessage(m) for some j < k. (As a set, this
is just the set of all sequences of external spambot-actions that have this
property.) Call this property P .
To prove that the spambot automaton given above satisfies P , we might
argue that for any execution s0 a0 s1 a1 . . . , that si = m in the last setMessage
action preceding si , or ⊥ if there is no such action. This is easily proved
by induction on i. It then follows that since spam(m) can only transmit the
current state, that if spam(m) follows si = m that it follows some earlier
setMessage(m) as claimed.
However, there are traces that satisfy P that don’t correspond to execu-
tions of the spambot; for example, consider the trace setMessage(0)setMessage(1)spam(0).
This satisfies P (0 was previously given to the automaton spam(0)), but
the automaton won’t generate it because the 0 was overwritten by the later
setMessage(1) action. Whether this is indicates a problem with our automa-
ton not being nondeterministic enough or our trace property being too weak
is a question about what we really want the automaton to do.

G.2.2 Types of trace properties

G.2.2.1 Safety properties
P is a safety property if

1. P is nonempty.
APPENDIX G. I/O AUTOMATA 449

2. P is prefix-closed, i.e. if xy is in P then x is in P .

3. P is limit-closed, i.e. if x1 , x1 x2 , x1 x2 x3 , . . . are all in P , then so is

the infinite sequence obtained by taking their limit.

Because of the last restrictions, it’s enough to prove that P holds for all
finite traces of A to show that it holds for all traces (and thus for all fair
traces), since any trace is a limit of finite traces. Conversely, if there is some
trace or fair trace for which P fails, the second restriction says that P fails
on any finite prefix of P , so again looking at only finite prefixes is enough.
The spambot property mentioned above is a safety property.
Safety properties are typically proved using invariants, properties that
are shown by induction to hold in all reachable states.

G.2.2.2 Liveness properties

P is a liveness property of A if any finite sequence of actions in acts(A)
has an extension in P . Note that liveness properties will in general include
many sequences of actions that aren’t traces of A, since they are extensions
of finite sequences that A can’t do (e.g. starting the execution with an action
not enabled in the initial state). If you want to restrict yourself only to
proper executions of A, use a safety property. (It’s worth noting that the
same property P can’t do both: any P that is both a liveness and a safety
property includes all sequences of actions because of the closure rules.)
Liveness properties are those that are always eventually satisfiable; as-
serting one says that the property is eventually satisfied. The typical way
to prove a liveness property is with a progress function, a function f
on states that (a) drops by at least 1 every time something that happens
infinitely often happens (like an action from an always-enabled task class)
and (b) guarantees P once it reaches 0.
An example would be the following property we might demand of our
spambot: any trace with at least one setMessage(. . . ) action contains infinitely
many spam(. . . ) actions. Whether the spambot automaton will satisfy this
property (in fair traces) depends on its task partition. If all spam(. . . ) actions
are in the same equivalence class, then any execution with at least one setMes-
sage will have some spam (. . . ) action enabled at all times thereafter, so a fair
trace containing a setMessage can’t be finite (since spam is enabled in the last
state) and if infinite contains infinitely many spam messages (since spam mes-
sages of some sort are enabled in all but an initial finite prefix). On the other
hand, if spam(m1 ) and spam(m2 ) are not equivalent in task(A), then the
spambot doesn’t satisfy the liveness property: in an execution that alternates
APPENDIX G. I/O AUTOMATA 450

setMessage(m1 )setMessage(m2 )setMessage(m1 )setMessage(m2 ) . . . there are

infinitely many states in which spam(m1 ) is not enabled, so fairness doesn’t
require doing it even once, and similarly for spam(m2 ).

G.2.2.3 Other properties

Any other property P can be expressed as the intersection of a safety property
(the closure of P ) and a liveness property (the union of P and the set of all
finite sequences that aren’t prefixes of traces in P ). The intuition is that
the safety property prunes out the excess junk we threw into the liveness
property to make it a liveness property, since any sequence that isn’t a prefix
of a trace in P won’t go into the safety property. This leaves only the traces
in P .
Example: Let P = {0n 1∞ } be the set of traces where we eventually give
up on our pointless 0-action and start doing only 1-actions forever. Then P
is the intersection of the safety property S = {0n 1m } ∪ P (the extra junk is
from prefix-closure) and the liveness property L = {0n 11m 0x|xin{0, 1}∗ } ∪ P .
Property S says that once we do a 1 we never do a 0, but allows finite
executions of the form 0n where we never do a 1. Property L says that we
eventually do a 1-action, but that we can’t stop unless we later do at least
one 0-action.

G.2.3 Compositional arguments

The product of trace properties P1 , P2 . . . is the trace property P where
T is in P if and only if T |sig(Pi ) is in Pi for each i. If the {Ai } satisfy
corresponding propertties {Pi } individually, then their composition satisfies
the product property. (For safety properties, often we prove something
weaker about the Ai , which is that each Ai individually is not the first to
violate P —i.e., it can’t leave P by executing an internal or output action.
In an execution where inputs by themselves can’t violate P , P then holds.)
Product properties let us prove trace properties by smashing together
properties of the component automata, possibly with some restrictions on
the signatures to get rid of unwanted actions. The product operation itself
is in a sense a combination of a Cartesian product (pick traces ti and smash
them together) filtered by a consistency rule (the smashed trace must be
consistent); it acts much like intersection (and indeed can be made identical
to intersection if we treat a trace property with a given signature as a way
of describing the set of all T such that T |sig(Pi ) is in Pi ).
APPENDIX G. I/O AUTOMATA 451

G.2.3.1 Example
Consider two spambots A1 and A2 where we identify the spam(m) operation
of A1 with the setMessage(m) operation of A2 ; we’ll call this combined
action spam1 (m) to distinguish it from the output actions of A2 . We’d like
to argue that the composite automaton A1 + A2 satisfies the safety property
(call it Pm ) that any occurrence of spam(m) is preceded by an occurrence
of setMessage(m), where the signature of Pm includes setMessage(m) and
spam(m) for some specific m but no other operations. (This is an example
of where trace property signatures can be useful without being limited to
actions of any specific component automaton.)
To do so, we’ll prove a stronger property Pm 0 , which is P
m modified
to include the spam1 (m) action in its signature. Observe that Pm 0 is the

product of the safety properties for A1 and A2 restricted to sig(Pm 0 ), since

the later says that any trace that includes spam(m) has a previous spam1 (m)
and the former says that any trace that includes spam1 (m) has a previous
setMessage(m). Since these properties hold for the individual A1 and A2 ,
their product, and thus the restriction Pm0 , holds for A + A , and so P (as
1 2 m
a further restriction) holds for A1 + A2 as well.
Now let’s prove the liveness property for A1 + A2 , that at least one
occurrence of setMessage yields infinitely many spam actions. Here we
let L1 = {at least one setMessage action ⇒ infinitely many spam1 actions}
and L2 = {at least one spam1 action ⇒ infinitely many spam actions}. The
product of these properties is all sequences with (a) no setMessage actions or
(b) infinitely many spam actions, which is what we want. This product holds
if the individual properties L1 and L2 hold for A1 + A2 , which will be the
case if we set task(A1 ) and task(A2 ) correctly.

G.2.4 Simulation arguments

Show that traces(A) is a subset of traces(B) (possibly after hiding some ac-
tions of A) by showing a simulation relation f : states(A) → states(B)
between states of A and states of B. Requirements on f are

1. If s is in start(A), then f (s) includes some element of start(B).

2. If (s, a, s0 ) is in trans(A) and s is reachable, then for any reachable u

in f (s), there is a sequence of actions x that takes u to some v in f (s0 )
with trace(x) = trace(a).

Using these we construct an execution of B matching (in trace) an

APPENDIX G. I/O AUTOMATA 452

execution of A by starting in f (s0 ) and applying the second part of the

definition to each action in the A execution (including the hidden ones!)

G.2.4.1 Example
A single spambot A can simulate the conjoined spambots A1 +A2 . Proof: Let
f (s) = (s, s). Then f (⊥) = (⊥, ⊥) is a start state of A1 + A2 . Now consider
a transition (s, a, s0 ) of A; the action a is either (a) setMessage(m), giving
s0 = m; here we let x = setMessage(m)spam1 (m) with trace(x) = trace(a)
since spam1 (m) is internal and f (s0 ) = (m, m) the result of applying x; or (b)
a = spam(m), which does not change s or f (s); the matching x is spam(m),
which also does not change f (s) and has the same trace.
A different proof could take advantage of f being a relation by defining
f (s) = {(s, s0 )|s0 ∈ states(A2 )}. Now we don’t care about the state of
A2 , and treat a setMessage(m) action of A as the sequence setMessage(m)
in A1 + A2 (which updates the first component of the state correctly) and
treat a spam(m) action as spam1 (m)spam(m) (which updates the second
component—which we don’t care about—and has the correct trace.) In some
cases an approach of this sort is necessary because we don’t know which
simulated state we are heading for until we get an action from A.
Note that the converse doesn’t work: A1 +A2 don’t simulate A, since there
are traces of A1 +A2 (e.g. setMessage(0)spam1 (0)setMessage(1)spam(0)) that
don’t restrict to traces of A. See [Lyn96, §8.5.5] for a more complicated
example of how one FIFO queue can simulate two FIFO queues and vice
versa (a situation called bisimulation).
Since we are looking at traces rather than fair traces, this kind of simula-
tion doesn’t help much with liveness properties, but sometimes the connection
between states plus a liveness proof for B can be used to get a liveness proof
for A (essentially we have to argue that A can’t do infinitely many action
without triggering a B-action in an appropriate task class). Again see [Lyn96,
§8.5.5].
Bibliography

[AA11] Dan Alistarh and James Aspnes. Sub-logarithmic test-and-

set against a weak adversary. In Distributed Computing: 25th
International Symposium, DISC 2011, volume 6950 of Lecture
Notes in Computer Science, pages 97–109. Springer-Verlag,
September 2011.

[AAB+ 11] Yehuda Afek, Noga Alon, Omer Barad, Eran Hornstein, Naama
Barkai, and Ziv Bar-Joseph. A biological solution to a fundamen-
tal distributed computing problem. science, 331(6014):183–185,
2011.

[AABJ+ 11] Yehuda Afek, Noga Alon, Ziv Bar-Joseph, Alejandro Cornejo,
Bernhard Haeupler, and Fabian Kuhn. Beeping a maximal
independent set. In Proceedings of the 25th International Con-
ference on Distributed Computing, DISC’11, pages 32–50, Berlin,
Heidelberg, 2011. Springer-Verlag.

[AACH+ 11] Dan Alistarh, James Aspnes, Keren Censor-Hillel, Seth Gilbert,
and Morteza Zadimoghaddam. Optimal-time adaptive tight
renaming, with applications to counting. In Proceedings of
the Thirtieth Annual ACM SIGACT-SIGOPS Symposium on
Principles of Distributed Computing, pages 239–248, June 2011.

[AACH12] James Aspnes, Hagit Attiya, and Keren Censor-Hillel. Poly-

logarithmic concurrent data structures from monotone circuits.
Journal of the ACM, 59(1):2:1–2:24, February 2012.

[AACHE15] James Aspnes, Hagit Attiya, Keren Censor-Hillel, and Faith

Ellen. Limited-use snapshots with polylogarithmic step com-
plexity. Journal of the ACM, 62(1):3, February 2015.

453
BIBLIOGRAPHY 454

[AACHE18] James Aspnes, Hagit Attiya, Keren Censor-Hillel, and Faith

Ellen. Erratum: Limited-use atomic snapshots with polyloga-
rithmic step complexity. J. ACM, 65(6):38:1–38:2, November
2018.

[AACV17] Yehuda Afek, James Aspnes, Edo Cohen, and Danny Vain-
stein. Brief announcement: Object oriented consensus. In
Elad Michael Schiller and Alexander A. Schwarzmann, editors,
Proceedings of the ACM Symposium on Principles of Distributed
Computing, PODC 2017, Washington, DC, USA, July 25-27,
2017, pages 367–369. ACM, 2017.

[AAD+ 93] Yehuda Afek, Hagit Attiya, Danny Dolev, Eli Gafni, Michael
Merritt, and Nir Shavit. Atomic snapshots of shared memory.
J. ACM, 40(4):873–890, 1993.

[AAD+ 06] Dana Angluin, James Aspnes, Zoë Diamadi, Michael J. Fischer,
and René Peralta. Computation in networks of passively mobile
finite-state sensors. Distributed Computing, pages 235–253,
March 2006.

[AAE06] Dana Angluin, James Aspnes, and David Eisenstat. Stably

computable predicates are semilinear. In PODC ’06: Proceed-
ings of the twenty-fifth annual ACM symposium on Principles
of distributed computing, pages 292–299, New York, NY, USA,
2006. ACM Press.

[AAE08a] Dana Angluin, James Aspnes, and David Eisenstat. Fast com-
putation by population protocols with a leader. Distributed
Computing, 21(3):183–199, September 2008.

[AAE08b] Dana Angluin, James Aspnes, and David Eisenstat. A sim-

ple population protocol for fast robust approximate majority.
Distributed Computing, 21(2):87–102, July 2008.

[AAG+ 10] Dan Alistarh, Hagit Attiya, Seth Gilbert, Andrei Giurgiu, and
Rachid Guerraoui. Fast randomized test-and-set and renam-
ing. In Nancy A. Lynch and Alexander A. Shvartsman, editors,
Distributed Computing, 24th International Symposium, DISC
2010, Cambridge, MA, USA, September 13-15, 2010. Proceed-
ings, volume 6343 of Lecture Notes in Computer Science, pages
94–108. Springer, 2010.
BIBLIOGRAPHY 455

[AAGG11] Dan Alistarh, James Aspnes, Seth Gilbert, and Rachid Guer-
raoui. The complexity of renaming. In Fifty-Second Annual
IEEE Symposium on Foundations of Computer Science, pages
718–727, October 2011.

[AAGW13] Dan Alistarh, James Aspnes, George Giakkoupis, and Philipp

Woelfel. Randomized loose renmaing in O(log log n) time. In
2013 ACM Symposium on Principles of Distributed Computing,
pages 200–209, July 2013.

[ABND+ 90] Hagit Attiya, Amotz Bar-Noy, Danny Dolev, David Peleg, and
Rüdiger Reischuk. Renaming in an asynchronous environment.
J. ACM, 37(3):524–548, 1990.

[ABND95] Hagit Attiya, Amotz Bar-Noy, and Danny Dolev. Sharing

memory robustly in message-passing systems. Journal of the
ACM, 42(1):124–142, 1995.

[Abr88] Karl Abrahamson. On achieving consensus using a shared

memory. In Proceedings of the 7th Annual ACM Symposium on
Principles of Distributed Computing (PODC), pages 291–302,
1988.

[AC08] Hagit Attiya and Keren Censor. Tight bounds for asynchronous
randomized consensus. Journal of the ACM, 55(5):20, October
2008.

[AC09] James Aspnes and Keren Censor. Approximate shared-memory

counting despite a strong adversary. In SODA ’09: Proceedings
of the Nineteenth Annual ACM -SIAM Symposium on Discrete
Algorithms, pages 441–450, Philadelphia, PA, USA, 2009. Soci-
ety for Industrial and Applied Mathematics.

[ACAH16] James Aspnes, Keren Censor-Hillel, Hagit Attiya, and Danny

Hendler. Lower bounds for restricted-use objects. SIAM J.
Comput., 45(3):734–762, 2016.

[ACH10] Hagit Attiya and Keren Censor-Hillel. Lower bounds for ran-
domized consensus under a weak adversary. SIAM J. Comput.,
39(8):3885–3904, 2010.

[ACH13] James Aspnes and Keren Censor-Hillel. Atomic snapshots in

O(log 3 n) steps using randomized helping. In Yehuda Afek,
BIBLIOGRAPHY 456

editor, Distributed Computing: 27th International Symposium,

DISC 2013, Jerusalem, Israel, October 14–18, 2013. Proceedings,
volume 8205 of Lecture Notes in Computer Science, pages 254–
268. Springer Berlin Heidelberg, 2013.
[ACH18] James Aspnes and Keren Censor-Hillel. Atomic snap-
shots in expected O(log 3 n) steps using randomized
helping. Submitted to Distributed Computing. Available
from http://www.cs.yale.edu/homes/aspnes/papers/
randomized-snapshot-abstract.html., 2018.
[ACHS13] Dan Alistarh, Keren Censor-Hillel, and Nir Shavit. Are lock-
free concurrent algorithms practically wait-free? arXiv preprint
arXiv:1311.3200, 2013.
[AE11] James Aspnes and Faith Ellen. Tight bounds for anonymous
adopt-commit objects. In 23rd Annual ACM Symposium on
Parallelism in Algorithms and Architectures, pages 317–324,
June 2011.
[AEG16] Yehuda Afek, Faith Ellen, and Eli Gafni. Deterministic objects:
Life beyond consensus. In Proceedings of the 2016 ACM Sympo-
sium on Principles of Distributed Computing, PODC ’16, pages
97–106, New York, NY, USA, 2016. ACM.
[AEH75] E. A. Akkoyunlu, K. Ekanadham, and R. V. Huber. Some con-
straints and tradeoffs in the design of network communications.
SIGOPS Oper. Syst. Rev., 9:67–74, November 1975.
[AF01] Hagit Attiya and Arie Fouren. Adaptive and efficient algo-
rithms for lattice agreement and renaming. SIAM Journal on
Computing, 31(2):642–664, 2001.
[AFL83] Eshrat Arjomandi, Michael J. Fischer, and Nancy A. Lynch. Ef-
ficiency of synchronous versus asynchronous distributed systems.
J. ACM, 30(3):449–456, 1983.
[AG91] Yehuda Afek and Eli Gafni. Time and message bounds for
election in synchronous and asynchronous complete networks.
SIAM Journal on Computing, 20(2):376–394, 1991.
[AG95] Sarita V. Adve and Kourosh Gharachorloo. Shared memory
consistency models: A tutorial. Technical Report 95/7, DEC
Western Research Laboratory, 1995.
BIBLIOGRAPHY 457

[AG18] Dan Alistarh and Rati Gelashvili. Recent algorithmic advances

in population protocols. SIGACT News, 49(3):63–73, October
2018.
[AGGT09] Dan Alistarh, Seth Gilbert, Rachid Guerraoui, and Corentin
Travers. Of choices, failures and asynchrony: The many faces
of set agreement. In Yingfei Dong, Ding-Zhu Du, and Os-
car H. Ibarra, editors, ISAAC, volume 5878 of Lecture Notes in
Computer Science, pages 943–953. Springer, 2009.
[AGHK06] Hagit Attiya, Rachid Guerraoui, Danny Hendler, and Petr
Kouznetsov. Synchronizing without locks is inherently expensive.
In PODC ’06: Proceedings of the twenty-fifth annual ACM
symposium on Principles of distributed computing, pages 300–
307, New York, NY, USA, 2006. ACM.
[AGTV92] Yehuda Afek, Eli Gafni, John Tromp, and Paul M. B. Vitányi.
Wait-free test-and-set (extended abstract). In Adrian Segall
and Shmuel Zaks, editors, Distributed Algorithms, 6th Inter-
national Workshop, WDAG ’92, Haifa, Israel, November 2-4,
1992, Proceedings, volume 647 of Lecture Notes in Computer
Science, pages 85–94. Springer, 1992.
[AH90a] James Aspnes and Maurice Herlihy. Fast randomized consensus
using shared memory. Journal of Algorithms, 11(3):441–461,
September 1990.
[AH90b] James Aspnes and Maurice Herlihy. Wait-free data structures
in the asynchronous PRAM model. In Second Annual ACM
Symposium on Parallel Algorithms and Architectures, pages
340–349, July 1990.
[AHM09] Hagit Attiya, Eshcar Hillel, and Alessia Milani. Inherent limita-
tions on disjoint-access parallel implementations of transactional
memory. In Friedhelm Meyer auf der Heide and Michael A.
Bender, editors, SPAA 2009: Proceedings of the 21st Annual
ACM Symposium on Parallelism in Algorithms and Architec-
tures, Calgary, Alberta, Canada, August 11-13, 2009, pages
69–78. ACM, 2009.
[AHR95] Hagit Attiya, Maurice Herlihy, and Ophir Rachman. Atomic
snapshots using lattice agreement. Distributed Computing,
8(3):121–132, 1995.
BIBLIOGRAPHY 458

[AHS94] James Aspnes, Maurice Herlihy, and Nir Shavit. Counting

networks. Journal of the ACM, 41(5):1020–1048, September
1994.

[AHTW18] James Aspnes, Bernhard Haeupler, Alexander Tong, and Philipp

Woelfel. Allocate-on-use space complexity of shared-memory
algorithms. In Ulrich Schmid and Josef Widder, editors, 32nd
International Symposium on Distributed Computing, DISC 2018,
New Orleans, LA, USA, October 15–19, 2018, volume 121 of
LIPIcs, pages 8:1–8:17. Schloss Dagstuhl - Leibniz-Zentrum fuer
Informatik, 2018.

[AHW08] Hagit Attiya, Danny Hendler, and Philipp Woelfel. Tight RMR
lower bounds for mutual exclusion and other problems. In
Proceedings of the 40th annual ACM symposium on Theory of
computing, STOC ’08, pages 217–226, New York, NY, USA,
2008. ACM.

[AKM+ 93] Baruch Awerbuch, Shay Kutten, Yishay Mansour, Boaz Patt-
Shamir, and George Varghese. Time optimal self-stabilizing
synchronization. In Proceedings of the twenty-fifth annual ACM
symposium on Theory of computing, pages 652–661. ACM, 1993.

[AKM+ 07] Baruch Awerbuch, Shay Kutten, Yishay Mansour, Boaz Patt-
Shamir, and George Varghese. A time-optional self-stabilizing
synchronizer using a phase clock. IEEE Transactions on De-
pendable and Secure Computing, 4(3):180–190, July–September
2007.

[AKP+ 06] Hagit Attiya, Fabian Kuhn, C. Greg Plaxton, Mirjam Watten-
hofer, and Roger Wattenhofer. Efficient adaptive collect using
randomization. Distributed Computing, 18(3):179–188, 2006.

[AKS83] M. Ajtai, J. Komlós, and E. Szemerédi. An o(n log n) sorting

network. In Proceedings of the fifteenth annual ACM symposium
on Theory of computing, pages 1–9, New York, NY, USA, 1983.
ACM.

[AM93] James H. Anderson and Mark Moir. Towards a necessary

and sufficient condition for wait-free synchronization (extended
abstract). In André Schiper, editor, Distributed Algorithms, 7th
International Workshop, WDAG ’93, Lausanne, Switzerland,
BIBLIOGRAPHY 459

September 27-29, 1993, Proceedings, volume 725 of Lecture

Notes in Computer Science, pages 39–53. Springer, 1993.
[AM94] Hagit Attiya and Marios Mavronicolas. Efficiency of semisyn-
chronous versus asynchronous networks. Mathematical Systems
Theory, 27(6):547–571, November 1994.
[AM99] Yehuda Afek and Michael Merritt. Fast, wait-free (2k − 1)-
renaming. In PODC, pages 105–112, 1999.
[AMW11] Yehuda Afek, Adam Morrison, and Guy Wertheim. From
bounded to unbounded concurrency objects and back. In Pro-
ceedings of the 30th Annual ACM SIGACT-SIGOPS Symposium
on Principles of Distributed Computing, pages 119–128. ACM,
2011.
[And90] Thomas E. Anderson. The performance of spin lock alternatives
for shared-money multiprocessors. IEEE Trans. Parallel Distrib.
Syst., 1(1):6–16, 1990.
[And94] James H. Anderson. Multi-writer composite registers. Dis-
tributed Computing, 7(4):175–195, 1994.
[Ang80] Dana Angluin. Local and global properties in networks of
processors (extended abstract). In Proceedings of the twelfth
annual ACM symposium on Theory of computing, STOC ’80,
pages 82–93, New York, NY, USA, 1980. ACM.
[AP06] Noa Agmon and David Peleg. Fault-tolerant gathering algo-
rithms for autnonomous mobile robots. SIAM Journal on
Computing, 36(1):56–82, 2006.
[AR09] James Aspnes and Eric Ruppert. An introduction to population
protocols. In Benoît Garbinato, Hugo Miranda, and Luís Ro-
drigues, editors, Middleware for Network Eccentric and Mobile
Applications, pages 97–120. Springer-Verlag, 2009.
[Asp98] James Aspnes. Lower bounds for distributed coin-flipping and
randomized consensus. Journal of the ACM, 45(3):415–450,
May 1998.
[Asp10] James Aspnes. Slightly smaller splitter networks. Technical
Report YALEU/DCS/TR-1438, Yale University Department of
Computer Science, November 2010.
BIBLIOGRAPHY 460

[Asp11] James Aspnes. Notes on randomized algorithms. http://www.

cs.yale.edu/homes/aspnes/classes/469/notes.pdf, July
2011.
[Asp12a] James Aspnes. Faster randomized consensus with an oblivious
adversary. In 2012 ACM Symposium on Principles of Distributed
Computing, pages 1–8, July 2012.
[Asp12b] James Aspnes. A modular approach to shared-memory consen-
sus, with applications to the probabilistic-write model. Dis-
tributed Computing, 25(2):179–188, May 2012.
[ASW88] Hagit Attiya, Marc Snir, and Manfred K. Warmuth. Computing
on an anonymous ring. J. ACM, 35:845–875, October 1988.
[Att14] Hagit Attiya. Lower bounds and impossibility results for transac-
tional memory computing. Bulletin of the European Association
for Computer Science, 112:38–52, February 2014.
[Aum97] Yonatan Aumann. Efficient asynchronous consensus with the
weak adversary scheduler. In PODC ’97: Proceedings of the
Sixteenth Annual ACM Symposium on Principles of Distributed
Computing, pages 209–218, New York, NY, USA, 1997. ACM.
[AW99] Yehuda Afek and Eytan Weisberger. The instancy of snapshots
and commuting objects. J. Algorithms, 30(1):68–105, 1999.
[AW04] Hagit Attiya and Jennifer Welch. Distributed Computing: Fun-
damentals, Simulations, and Advanced Topics. Wiley, second
edition, 2004. On-line version: http://dx.doi.org/10.1002/
0471478210. (This may not work outside Yale.).
[Awe85] Baruch Awerbuch. Complexity of network synchronization. J.
ACM, 32:804–823, October 1985.
[AWW93] Yehuda Afek, Eytan Weisberger, and Hanan Weisman. A com-
pleteness theorem for a class of synchronization objects (ex-
tended abstract). In Proceedings of the Twelfth Annual ACM
Symposium on Principles of Distributed Computing, pages 159–
170, 1993.
[Bat68] K. E. Batcher. Sorting networks and their applications. In
Proceedings of the AFIPS Spring Joint Computer Conference
32, pages 307–314, 1968.
BIBLIOGRAPHY 461

[BBE+ 15] P. Berenbrink, A. Brinkmann, R. Elsässer, T. Friedetzky, and

L. Nagel. Randomized renaming in shared memory systems. In
Parallel and Distributed Processing Symposium (IPDPS), 2015
IEEE International, pages 542–549, May 2015.

[BDLP08] Christian Boulinier, Ajoy K. Datta, Lawrence L. Larmore, and

Franck Petit. Space efficient and time optimal distributed BFS
tree construction. Information Processing Letters, 108(5):273–
278, November 2008. http://dx.doi.org/10.1016/j.ipl.
2008.05.016.

[BDT12] Zohir Bouzid, Shantanu Das, and Sébastien Tixeuil. Wait-free

gathering of mobile robots. CoRR, abs/1207.0226, 2012.

[Bel58] RE Bellman. On a routing problem. Quarterly of Applied

Mathematics, 16:87–90, 1958.

[Bel03] S. Bellovin. The Security Flag in the IPv4 Header. RFC 3514
(Informational), April 2003.

[BEW11] Alex Brodsky, Faith Ellen, and Philipp Woelfel. Fully-adaptive

algorithms for long-lived renaming. Distributed Computing,
24(2):119–134, 2011.

[BG93] Elizabeth Borowsky and Eli Gafni. Generalized flp impossibility

result for t-resilient asynchronous computations. In STOC,
pages 91–100, 1993.

[BG97] Elizabeth Borowsky and Eli Gafni. A simple algorithmically

reasoned characterization of wait-free computations (extended
abstract). In PODC, pages 189–198, 1997.

[BGA94] Elizabeth Borowsky, Eli Gafni, and Yehuda Afek. Consensus

power makes (some) sense! (extended abstract). In PODC,
pages 363–372, 1994.

[BGLR01] E. Borowsky, E. Gafni, N. Lynch, and S. Rajsbaum. The bg

distributed simulation algorithm. Distrib. Comput., 14(3):127–
146, October 2001.

[BGP89] Piotr Berman, Juan A. Garay, and Kenneth J. Perry. To-

wards optimal distributed consensus (extended abstract). In
30th Annual Symposium on Foundations of Computer Science,
BIBLIOGRAPHY 462

30 October-1 November 1989, Research Triangle Park, North

Carolina, USA, pages 410–415, 1989.

[BHMT20] Mirza Ahad Baig, Danny Hendler, Alessia Milani, and Corentin
Travers. Long-lived snapshots with polylogarithmic amortized
step complexity. In Proceedings of the 39th Symposium on
Principles of Distributed Computing, PODC ’20, page 3140, New
York, NY, USA, 2020. Association for Computing Machinery.

[BL93] James E. Burns and Nancy A. Lynch. Bounds on shared memory

for mutual exclusion. Inf. Comput., 107(2):171–184, 1993.

[BND89] A. Bas-Noy and D. Dolev. Shared-memory vs. message-passing

in an asynchronous distributed environment. In Proceedings of
the eighth annual ACM Symposium on Principles of distributed
computing, PODC ’89, pages 307–318, New York, NY, USA,
1989. ACM.

[BO83] Michael Ben-Or. Another advantage of free choice: Completely

asynchronous agreement protocols (extended abstract). In Pro-
ceedings of the Second Annual ACM SIGACT-SIGOPS Sym-
posium on Principles of Distributed Computing, pages 27–30,
Montreal, Quebec, Canada, August 1983.

[Bor95] Elizabeth Borowsky. Capturing the Power of Resiliency and

Set Consensus in Distributed Systems. PhD thesis, University
of California, Los Angeles, 1995.

[BPSV06] Harry Buhrman, Alessandro Panconesi, Riccardo Silvestri, and

Paul Vitányi. On the importance of having an identity or,
is consensus really universal? Distrib. Comput., 18:167–176,
February 2006.

[BR91] Gabriel Bracha and Ophir Rachman. Randomized consensus in

expected O(n2 log n) operations. In Sam Toueg, Paul G. Spi-
rakis, and Lefteris M. Kirousis, editors, Distributed Algorithms,
5th International Workshop, volume 579 of Lecture Notes in
Computer Science, pages 143–150, Delphi, Greece, 7–9 October
1991. Springer, 1992.

[BT15] Quentin Bramas and Sébastien Tixeuil. Wait-free gathering

without chirality. In Christian Scheideler, editor, Structural In-
formation and Communication Complexity: 22nd International
BIBLIOGRAPHY 463

Colloquium, SIROCCO 2015, Montserrat, Spain, July 14-16,

2015. Post-Proceedings, pages 313–327, Cham, 2015. Springer
International Publishing.

[Bur80] James E. Burns. A formal model for message passing systems.

Technical Report 91, Computer Science Department, Indiana
University, September 1980. http://www.cs.indiana.edu/
pub/techreports/TR91.pdf.

[CCN12] Luca Cardelli and Attila Csikász-Nagy. The cell cycle switch
computes approximate majority. Scientific reports, 2, 2012.

[Cha93] Soma Chaudhuri. More choices allow more faults: Set consen-
sus problems in totally asynchronous systems. Inf. Comput.,
105(1):132–158, 1993.

[Cha96] Tushar Deepak Chandra. Polylog randomized wait-free consen-

sus. In Proceedings of the Fifteenth Annual ACM Symposium on
Principles of Distributed Computing, pages 166–175, Philadel-
phia, Pennsylvania, USA, 23–26 May 1996.

[CHKM19] Anne Condon, Monir Hajiaghayi, David Kirkpatrick, and Ján

Maňuch. Approximate majority analyses using tri-molecular
chemical reaction networks. Natural Computing, pages 1–22,
2019.

[CHT96] Tushar Deepak Chandra, Vassos Hadzilacos, and Sam Toueg.

The weakest failure detector for solving consensus. J. ACM,
43:685–722, July 1996.

[CIL94] Benny Chor, Amos Israeli, and Ming Li. Wait-free consensus
using asynchronous hardware. SIAM J. Comput., 23(4):701–712,
1994.

[CK10] Alejandro Cornejo and Fabian Kuhn. Deploying wireless net-

works with beeps. In Proceedings of the 24th International
Conference on Distributed Computing, DISC’10, pages 148–162,
Berlin, Heidelberg, 2010. Springer-Verlag.

[CL85] K. Mani Chandy and Leslie Lamport. Distributed snapshots:

Determining global states of distributed systems. ACM Trans.
Comput. Syst., 3(1):63–75, 1985.
BIBLIOGRAPHY 464

[CLM+ 16] Michael B. Cohen, Yin Tat Lee, Gary L. Miller, Jakub Pachocki,
and Aaron Sidford. Geometric median in nearly linear time.
In Daniel Wichs and Yishay Mansour, editors, Proceedings
of the 48th Annual ACM SIGACT Symposium on Theory of
Computing, STOC 2016, Cambridge, MA, USA, June 18-21,
2016, pages 9–21. ACM, 2016.

[CR79] Ernest Chang and Rosemary Roberts. An improved algorithm

for decentralized extrema-finding in circular configurations of
processes. Commun. ACM, 22:281–283, May 1979.

[CR93] Ran Canetti and Tal Rabin. Fast asynchronous byzantine

agreement with optimal resilience. In S. Rao Kosaraju, David S.
Johnson, and Alok Aggarwal, editors, Proceedings of the Twenty-
Fifth Annual ACM Symposium on Theory of Computing, May
16-18, 1993, San Diego, CA, USA, pages 42–51. ACM, 1993.

[CR08] Armando Castañeda and Sergio Rajsbaum. New combinatorial

topology upper and lower bounds for renaming. In Rida A.
Bazzi and Boaz Patt-Shamir, editors, Proceedings of the Twenty-
Seventh Annual ACM Symposium on Principles of Distributed
Computing, PODC 2008, Toronto, Canada, August 18-21, 2008,
pages 295–304. ACM, 2008.

[CT96] Tushar Deepak Chandra and Sam Toueg. Unreliable failure

detectors for reliable distributed systems. J. ACM, 43:225–267,
March 1996.

[CV86] Richard Cole and Uzi Vishkin. Deterministic coin tossing with
applications to optimal parallel list ranking. Information and
Control, 70(1):32–53, 1986.

[DH04] Robert Danek and Vassos Hadzilacos. Local-spin group mutual

exclusion algorithms. In Rachid Guerraoui, editor, Distributed
Computing, 18th International Conference, DISC 2004, Ams-
terdam, The Netherlands, October 4-7, 2004, Proceedings, vol-
ume 3274 of Lecture Notes in Computer Science, pages 71–85.
Springer, 2004.

[DHW97] Cynthia Dwork, Maurice Herlihy, and Orli Waarts. Contention

in shared memory algorithms. J. ACM, 44(6):779–805, 1997.
BIBLIOGRAPHY 465

[Dij74] Edsger W. Dijkstra. Self-stabilizing systems in spite of dis-

tributed control. Communications of the ACM, 17(11):643–644,
November 1974.

[Dij75] Edsger W. Dijkstra. Guarded commands, non-determinacy and

formal derivation of programs. Communications of the ACM,
18(8):453–457, 1975.

[DLP+ 86] Danny Dolev, Nancy A. Lynch, Shlomit S. Pinter, Eugene W.

Stark, and William E. Weihl. Reaching approximate agreement
in the presence of faults. J. ACM, 33(3):499–516, 1986.

[DLS88] Cynthia Dwork, Nancy A. Lynch, and Larry J. Stockmeyer. Con-

sensus in the presence of partial synchrony. J. ACM, 35(2):288–
323, 1988.

[Dol00] Shlomi Dolev. Self-Stabilization. MIT Press, 2000.

[DS83] Danny Dolev and H. Raymond Strong. Authenticated algo-

rithms for byzantine agreement. SIAM J. Comput., 12(4):656–
666, 1983.

[DS15] David Doty and David Soloveichik. Stable leader election in

population protocols requires linear time. In Yoram Moses,
editor, Distributed Computing: 29th International Symposium,
DISC 2015, Tokyo, Japan, October 7-9, 2015, Proceedings, pages
602–616, Berlin, Heidelberg, 2015. Springer Berlin Heidelberg.

[EGSZ16] Faith Ellen, Rati Gelashvili, Nir Shavit, and Leqi Zhu. A
complexity-based hierarchy for multiprocessor synchronization:
[extended abstract]. In Proceedings of the 2016 ACM Symposium
on Principles of Distributed Computing, PODC ’16, pages 289–
298, New York, NY, USA, 2016. ACM.

[EHS12] Faith Ellen, Danny Hendler, and Nir Shavit. On the inherent se-
quentiality of concurrent objects. SIAM Journal on Computing,
41(3):519–536, 2012.

[ER+ 18] Robert Elsässer, Tomasz Radzik, et al. Recent results in popu-
lation protocols for exact majority and leader election. Bulletin
of EATCS, 3(126), 2018.
BIBLIOGRAPHY 466

[EW13] Faith Ellen and Philipp Woelfel. An optimal implementation

of fetch-and-increment. In Yehuda Afek, editor, Distributed
Computing, pages 284–298, Berlin, Heidelberg, 2013. Springer
Berlin Heidelberg.

[FH07] Keir Fraser and Timothy L. Harris. Concurrent programming

without locks. ACM Trans. Comput. Syst., 25(2), 2007.

[FHS98] Faith Ellen Fich, Maurice Herlihy, and Nir Shavit. On the space
complexity of randomized synchronization. J. ACM, 45(5):843–
862, 1998.

[FHS05] Faith Ellen Fich, Danny Hendler, and Nir Shavit. Linear lower
bounds on real-world implementations of concurrent objects. In
Foundations of Computer Science, Annual IEEE Symposium on,
pages 165–173, Los Alamitos, CA, USA, 2005. IEEE Computer
Society.

[Fic05] Faith Fich. How hard is it to take a snapshot? In Peter Vojtáš,

Mária Bieliková, Bernadette Charron-Bost, and Ondrej Sýkora,
editors, SOFSEM 2005: Theory and Practice of Computer
Science, volume 3381 of Lecture Notes in Computer Science,
pages 28–37. Springer Berlin / Heidelberg, 2005.

[Fid91] Colin J. Fidge. Logical time in distributed computing systems.

IEEE Computer, 24(8):28–33, 1991.

[FK07] Panagiota Fatourou and Nikolaos D. Kallimanis. Time-optimal,

space-efficient single-scanner snapshots & multi-scanner snap-
shots using CAS. In Indranil Gupta and Roger Wattenhofer,
editors, Proceedings of the Twenty-Sixth Annual ACM Sym-
posium on Principles of Distributed Computing, PODC 2007,
Portland, Oregon, USA, August 12-15, 2007, pages 33–42. ACM,
2007.

[FL82] Michael J. Fischer and Nancy A. Lynch. A lower bound for

the time to assure interactive consistency. Inf. Process. Lett.,
14(4):183–186, 1982.

[FL87] Greg N. Frederickson and Nancy A. Lynch. Electing a leader

in a synchronous ring. J. ACM, 34(1):98–115, 1987.
BIBLIOGRAPHY 467

[FL06] Rui Fan and Nancy A. Lynch. An ω(n log n) lower bound on
the cost of mutual exclusion. In Eric Ruppert and Dahlia
Malkhi, editors, Proceedings of the Twenty-Fifth Annual ACM
Symposium on Principles of Distributed Computing, PODC
2006, Denver, CO, USA, July 23-26, 2006, pages 275–284.
ACM, 2006.

[Fle59] Ian Fleming. Goldfinger. Jonathan Cape, 1959.

[FLM86] Michael J. Fischer, Nancy A. Lynch, and Michael Merritt. Easy

impossibility proofs for distributed consensus problems. Dis-
tributed Computing, 1(1):26–39, 1986.

[FLMS05] Faith Ellen Fich, Victor Luchangco, Mark Moir, and Nir Shavit.
Obstruction-free algorithms can be practically wait-free. In
Pierre Fraigniaud, editor, Distributed Computing, 19th Inter-
national Conference, DISC 2005, Cracow, Poland, September
26-29, 2005, Proceedings, volume 3724 of Lecture Notes in Com-
puter Science, pages 78–92. Springer, 2005.

[FLP85] Michael J. Fischer, Nancy A. Lynch, and Michael S. Paterson.

Impossibility of distributed consensus with one faulty process.
Journal of the ACM, 32(2):374–382, April 1985.

[For56] Lester Randolph Ford. Network Flow Theory. Rand Corpora-

tion, 1956.

[FR98] Michael J. Fischer and Michael O. Rabin. Super-exponential

complexity of presburger arithmetic. In Bob F. Caviness and
Jeremy R. Johnson, editors, Quantifier Elimination and Cylin-
drical Algebraic Decomposition, pages 122–135. Springer Vienna,
Vienna, 1998.

[FW10] Roland Flury and Roger Wattenhofer. Slotted programming for

sensor networks. In Tarek F. Abdelzaher, Thiemo Voigt, and
Adam Wolisz, editors, Proceedings of the 9th International Con-
ference on Information Processing in Sensor Networks, IPSN
2010, April 12-16, 2010, Stockholm, Sweden, pages 24–34. ACM,
2010.

[Gaf98] Eli Gafni. Round-by-round fault detectors: Unifying synchrony

and asynchrony (extended abstract). In Proceedings of the Sev-
BIBLIOGRAPHY 468

enteenth Annual ACM Symposium on Principles of Distributed

Computing, pages 143–152, 1998.

[Gaf09] Eli Gafni. The extended BG-simulation and the characteriza-

tion of t-resiliency. In Proceedings of the 41st annual ACM
symposium on Theory of computing, pages 85–92. ACM, 2009.

[Gal82] Robert G. Gallager. Distributed minimum hop algorithms. Tech-

nical Report LIDS-P-1175, M.I.T. Laboratory for Information
and Decision Systems, January 1982.

[Gel15] Rati Gelashvili. On the optimal space complexity of consensus

for anonymous processes. In Yoram Moses, editor, Distributed
Computing: 29th International Symposium, DISC 2015, Tokyo,
Japan, October 7-9, 2015, Proceedings, pages 452–466, Berlin,
Heidelberg, 2015. Springer Berlin Heidelberg.

[GHHW12] Wojciech Golab, Vassos Hadzilacos, Danny Hendler, and Philipp

Woelfel. RMR-efficient implementations of comparison primi-
tives using read and write operations. Distributed Computing,
25(2):109–162, May 2012.

[GHHW13] George Giakkoupis, Maryam Helmi, Lisa Higham, and Philipp

√
Woelfel. An o( n) space bound for obstruction-free leader
election. In Proceedings of the 27th International Symposium
on Distributed Computing (DISC), pages 46–60, October 14–18
2013.

[GHHW15] George Giakkoupis, Maryam Helmi, Lisa Higham, and Philipp

Woelfel. Test-and-set in optimal space. In Proceedings of the
Forty-seventh Annual ACM Symposium on Theory of Comput-
ing, STOC ’15, pages 615–623, New York, NY, USA, 2015.
ACM.

[GHW11] Wojciech Golab, Lisa Higham, and Philipp Woelfel. Lineariz-

able implementations do not suffice for randomized distributed
computation. In Proceedings of the Forty-third Annual ACM
Symposium on Theory of Computing, STOC ’11, pages 373–382,
New York, NY, USA, 2011. ACM.

[GK20] Mohsen Ghaffari and Fabian Kuhn. Deterministic distributed

vertex coloring: Simpler, faster, and without network decompo-
sition. CoRR, abs/2011.04511, 2020.
BIBLIOGRAPHY 469

[GKM+ 19] Rachid Guerraoui, Petr Kuznetsov, Matteo Monti, Matej

Pavlovič, and Dragos-Adrian Seredinschi. The consensus num-
ber of a cryptocurrency. In Proceedings of the 2019 ACM
Symposium on Principles of Distributed Computing, PODC
’19, page 307–316, New York, NY, USA, 2019. Association for
Computing Machinery.

[GM98] Juan A. Garay and Yoram Moses. Fully polynomial byzantine

agreement for n > 3t processors in t + 1 rounds. SIAM J.
Comput., 27(1):247–290, 1998.

[Gol11] Wojciech M. Golab. A complexity separation between the

cache-coherent and distributed shared memory models. In
Cyril Gavoille and Pierre Fraigniaud, editors, Proceedings of
the 30th Annual ACM Symposium on Principles of Distributed
Computing, PODC 2011, San Jose, CA, USA, June 6-8, 2011,
pages 109–118. ACM, 2011.

[Gra78] Jim Gray. Notes on data base operating systems. In Operating

Systems, An Advanced Course, pages 393–481. Springer-Verlag,
London, UK, 1978.

[GRS90] Ronald L. Graham, Bruce L. Rothschild, and Joel H. Spencer.

Ramsey Theory. Wiley-Interscience, 2nd edition, 1990.

[GW12a] George Giakkoupis and Philipp Woelfel. On the time and space
complexity of randomized test-and-set. In Darek Kowalski and
Alessandro Panconesi, editors, ACM Symposium on Princi-
ples of Distributed Computing, PODC ’12, Funchal, Madeira,
Portugal, July 16-18, 2012, pages 19–28. ACM, 2012.

[GW12b] George Giakkoupis and Philipp Woelfel. A tight RMR lower

bound for randomized mutual exclusion. In Proceedings of
the 44th symposium on Theory of Computing, pages 983–1002.
ACM, 2012.

[GW14] G. Giakkoupis and P. Woelfel. Randomized mutual exclusion

with constant amortized RMR complexity on the DSM. In 2014
IEEE 55th Annual Symposium on Foundations of Computer
Science, pages 504–513, Oct 2014.

[GW17] George Giakkoupis and Philipp Woelfel. Randomized abortable

mutual exclusion with constant amortized RMR complexity
BIBLIOGRAPHY 470

on the cc model. In Proceedings of the ACM Symposium on

Principles of Distributed Computing, PODC ’17, pages 221–229,
New York, NY, USA, 2017. ACM.
[Her91a] Maurice Herlihy. Impossibility results for asynchronous PRAM
(extended abstract). In Proceedings of the third annual ACM
symposium on Parallel algorithms and architectures, SPAA ’91,
pages 327–336, New York, NY, USA, 1991. ACM.
[Her91b] Maurice Herlihy. Wait-free synchronization. ACM Trans. Pro-
gram. Lang. Syst., 13(1):124–149, January 1991.
[Her93] Maurice Herlihy. A methodology for implementing highly con-
current objects. ACM Trans. Program. Lang. Syst., 15(5):745–
770, 1993.
[HFP02] Timothy L. Harris, Keir Fraser, and Ian A. Pratt. A practical
multi-word compare-and-swap operation. In Dahlia Malkhi,
editor, Distributed Computing, 16th International Conference,
DISC 2002, Toulouse, France, October 28-30, 2002 Proceedings,
volume 2508 of Lecture Notes in Computer Science, pages 265–
279. Springer, 2002.
[HHT20] Vassos Hadzilacos, Xing Hu, and Sam Toueg. On linearizability
and the termination of randomized algorithms, 2020.
[HK14] Danny Hendler and Vitaly Khait. Complexity tradeoffs for
read and update operations. In Proceedings of the 2014 ACM
Symposium on Principles of Distributed Computing, PODC ’14,
pages 186–195, New York, NY, USA, 2014. ACM.
[HLM03] Maurice Herlihy, Victor Luchangco, and Mark Moir.
Obstruction-free synchronization: Double-ended queues as an
example. In 23rd International Conference on Distributed Com-
puting Systems (ICDCS 2003), 19-22 May 2003, Providence,
RI, USA, pages 522–529. IEEE Computer Society, 2003.
[HM93] Maurice Herlihy and J. Eliot B. Moss. Transactional memory:
Architectural support for lock-free data structures. In ISCA,
pages 289–300, 1993.
[HS80] Daniel S. Hirschberg and J. B. Sinclair. Decentralized extrema-
finding in circular configurations of processors. Commun. ACM,
23(11):627–628, 1980.
BIBLIOGRAPHY 471

[HS99] Maurice Herlihy and Nir Shavit. The topological structure of

asynchronous computability. J. ACM, 46(6):858–923, 1999.
[HW90] Maurice Herlihy and Jeannette M. Wing. Linearizability: A cor-
rectness condition for concurrent objects. ACM Trans. Program.
Lang. Syst., 12(3):463–492, 1990.
[HW11] Danny Hendler and Philipp Woelfel. Randomized mutual exclu-
sion with sub-logarithmic RMR-complexity. Distributed Com-
puting, 24(1):3–19, 2011.
[IMCT94] Michiko Inoue, Toshimitsu Masuzawa, Wei Chen, and Nobuki
Tokura. Linear-time snapshot using multi-writer multi-reader
registers. In Gerard Tel and Paul Vitányi, editors, Distributed
Algorithms, volume 857 of Lecture Notes in Computer Science,
pages 130–140. Springer Berlin / Heidelberg, 1994.
[IR09] Damien Imbs and Michel Raynal. Visiting Gafni’s reduction
land: From the BG simulation to the extended BG simulation.
In Stabilization, Safety, and Security of Distributed Systems,
pages 369–383. Springer, 2009.
[Jay97] Prasad Jayanti. Robust wait-free hierarchies. J. ACM, 44(4):592–
614, 1997.
[Jay98] Prasad Jayanti. A time complexity lower bound for random-
ized implementations of some shared objects. In Proceedings
of the Seventeenth Annual ACM Symposium on Principles of
Distributed Computing, PODC ’98, pages 201–210, New York,
NY, USA, 1998. ACM.
[Jay02] Prasad Jayanti. f -arrays: implementation and applications. In
Proceedings of the twenty-first annual symposium on Principles
of distributed computing, PODC ’02, pages 270–279, New York,
NY, USA, 2002. ACM.
[Jay11] Prasad Jayanti. personal communication, 19 October 2011.
[JT92] Prasad Jayanti and Sam Toueg. Some results on the impos-
sibility, universality, and decidability of consensus. In Adrian
Segall and Shmuel Zaks, editors, Distributed Algorithms: 6th
International Workshop, WDAG ’92 Haifa, Israel, November
2–4, 1992 Proceedings, pages 69–84, Berlin, Heidelberg, 1992.
Springer Berlin Heidelberg.
BIBLIOGRAPHY 472

[JTT00] Prasad Jayanti, King Tan, and Sam Toueg. Time and space
lower bounds for nonblocking implementations. SIAM J. Com-
put., 30(2):438–456, 2000.

[Kaw00] Jawal Y. Kawash. Limitations and Capabilities of Weak Memory

Consistency Systems. PhD thesis, University of Calgary, January
2000.

[LAA87] Michael C. Loui and Hosame H. Abu-Amara. Memory require-

ments for agreement among unreliable asynchronous processes.
In Franco P. Preparata, editor, Parallel and Distributed Com-
puting, volume 4 of Advances in Computing Research, pages
163–183. JAI Press, 1987.

[Lam77] Leslie Lamport. Concurrent reading and writing. Communica-

tions of the ACM, 20(11):806–811, November 1977.

[Lam78] Leslie Lamport. Time, clocks, and the ordering of events in a

distributed system. Commun. ACM, 21(7):558–565, 1978.

[Lam79] L. Lamport. How to make a multiprocessor computer that

correctly executes multiprocess programs. Computers, IEEE
Transactions on, C-28(9):690–691, Sept 1979.

[Lam83] Leslie Lamport. The weak byzantine generals problem. J. ACM,

30(3):668–676, 1983.

[Lam87] Leslie Lamport. A fast mutual exclusion algorithm. ACM Trans.

Comput. Syst., 5(1):1–11, 1987.

[Lam98] Leslie Lamport. The part-time parliament. ACM Trans. Com-

put. Syst., 16(2):133–169, 1998.

[Lam01] Leslie Lamport. Paxos made simple. SIGACT News, 32(4):18–

25, 2001.

[Lin92] Nathan Linial. Locality in distributed graph algorithms. SIAM

J. Comput., 21(1):193–201, 1992.

[LL77] Gérard Le Lann. Distributed systems—towards a formal ap-

proach. In B. Gilchrist, editor, Information Processing 77, pages
155–160. North-Holland, 1977.
BIBLIOGRAPHY 473

[LS14] Juhana Laurinharju and Jukka Suomela. Brief announcement:

Linial’s lower bound made easy. In Magnús M. Halldórsson
and Shlomi Dolev, editors, ACM Symposium on Principles of
Distributed Computing, PODC ’14, Paris, France, July 15-18,
2014, pages 377–378. ACM, 2014.

[LSW09] Christoph Lenzen, Jukka Suomela, and Roger Wattenhofer. Lo-

cal algorithms: Self-stabilization on speed. In Rachid Guerraoui
and Franck Petit, editors, Stabilization, Safety, and Security
of Distributed Systems: 11th International Symposium, SSS
2009, Lyon, France, November 3-6, 2009. Proceedings, volume
5873 of Lecture Notes in Computer Science, pages 17–34, Berlin,
Heidelberg, 2009. Springer Berlin Heidelberg.

[LT87] Nancy A. Lynch and Mark R. Tuttle. Hierarchical correctness

proofs for distributed algorithms. In Proceedings of the Sixth
Annual ACM Symposium on Principles of Distributed Com-
puting, PODC ’87, pages 137–151, New York, NY, USA, 1987.
ACM.

[Lyn96] Nancy A. Lynch. Distributed Algorithms. Morgan Kaufmann,

1996.

[MA95] Mark Moir and James H. Anderson. Wait-free algorithms for

fast, long-lived renaming. Sci. Comput. Program., 25(1):1–39,
1995.

[Mat93] Friedemann Mattern. Efficient algorithms for distributed snap-

shots and global virtual time approximation. J. Parallel Distrib.
Comput., 18(4):423–434, 1993.

[MCS91] John M. Mellor-Crummey and Michael L. Scott. Algorithms

for scalable synchronization on shared-memory multiprocessors.
ACM Trans. Comput. Syst., 9(1):21–65, 1991.

[Mor95] Shlomo Moran. Using approximate agreement to obtain com-

plete disagreement: the output structure of input-free asyn-
chronous computations. In Third Israel Symposium on the
Theory of Computing and Systems, pages 251–257, January
1995.

[MR98] Dahlia Malkhi and Michael K. Reiter. Byzantine quorum sys-

tems. Distributed Computing, 11(4):203–213, 1998.
BIBLIOGRAPHY 474

[MR10] Michael Merideth and Michael Reiter. Selected results from

the latest decade of quorum systems research. In Bernadette
Charron-Bost, Fernando Pedone, and André Schiper, editors,
Replication, volume 5959 of Lecture Notes in Computer Science,
pages 185–206. Springer, 2010.
[MRRT08] Achour Mostefaoui, Sergio Rajsbaum, Michel Raynal, and
Corentin Travers. The combined power of conditions and infor-
mation on failures to solve asynchronous set agreement. SIAM
Journal on Computing, 38(4):1574–1601, 2008.
[MRWW01] Dahlia Malkhi, Michael K. Reiter, Avishai Wool, and Rebecca N.
Wright. Probabilistic quorum systems. Inf. Comput., 170(2):184–
206, 2001.
[MRZ15] Yves Métivier, John Michael Robson, and Akka Zemmari. On
distributed computing with beeps. CoRR, abs/1507.02721,
2015.
[NT87] Gil Neiger and Sam Toueg. Substituting for real time and
common knowledge in asynchronous distributed systems. In
Proceedings of the sixth annual ACM Symposium on Principles
of distributed computing, PODC ’87, pages 281–293, New York,
NY, USA, 1987. ACM.
[NW98] Moni Naor and Avishai Wool. The load, capacity, and avail-
ability of quorum systems. SIAM J. Comput., 27(2):423–447,
1998.
[Oka99] Chris Okasaki. Purely Functional Data Structures. Cambridge
University Press, 1999.
[Pel00] David Peleg. Distributed Computing: A Locality-Sensitive Ap-
proach. SIAM, 2000.
[Pet62] Carl Adam Petri. Kommunikation mit Automaten. PhD thesis,
Technische Hochschule Darmstadt, 1962.
[Pet81] Gary L. Peterson. Myths about the mutual exclusion problem.
Inf. Process. Lett., 12(3):115–116, 1981.
[Pet82] Gary L. Peterson. An O(n log n) unidirectional algorithm for
the circular extrema problem. ACM Trans. Program. Lang.
Syst., 4(4):758–762, 1982.
BIBLIOGRAPHY 475

[PF77] Gary L. Peterson and Michael J. Fischer. Economical solu-

tions for the critical section problem in a distributed system
(extended abstract). In John E. Hopcroft, Emily P. Friedman,
and Michael A. Harrison, editors, Proceedings of the 9th Annual
ACM Symposium on Theory of Computing, May 4-6, 1977,
Boulder, Colorado, USA, pages 91–97. ACM, 1977.

[Plo89] S. A. Plotkin. Sticky bits and universality of consensus. In

Proceedings of the eighth annual ACM Symposium on Principles
of distributed computing, PODC ’89, pages 159–175, New York,
NY, USA, 1989. ACM.

[Pos81] J. Postel. Transmission Control Protocol. RFC 793 (INTERNET

STANDARD), September 1981. Updated by RFCs 1122, 3168,
6093, 6528.

[PR01] Alessandro Panconesi and Romeo Rizzi. Some simple distributed

algorithms for sparse networks. Distributed Comput., 14(2):97–
100, 2001.

[Pre29] M. Presburger. Über die Vollständigkeit eines gewissen Systems

der Arithmetik ganzer Zahlen, in welchem die Addition als
einzige Operation hervortritt. In Comptes-Rendus du I Congrès
de Mathématiciens des Pays Slaves, Warszawa, pages 92–101,
1929.

[Pri01] Giusseppe Principe. CORDA: Distributed coordination of a set

of autonomous mobile robots. In Proceedings of the European
Research Seminar on Advances in Distributed Systems, pages
185–190, 2001.

[PSL80] M. Pease, R. Shostak, and L. Lamport. Reaching agreement

in the presence of faults. Journal of the ACM, 27(2):228–234,
April 1980.

[PW95] David Peleg and Avishai Wool. The availability of quorum

systems. Inf. Comput., 123(2):210–223, 1995.

[PW97a] David Peleg and Avishai Wool. The availability of crumbling

wall quorum systems. Discrete Applied Mathematics, 74(1):69–
83, 1997.
BIBLIOGRAPHY 476

[PW97b] David Peleg and Avishai Wool. Crumbling walls: A class of

practical and efficient quorum systems. Distributed Computing,
10(2):87–97, 1997.
[Rab83] M. O. Rabin. Randomized byzantine generals. In 24th Annual
Symposium on Foundations of Computer Science (sfcs 1983),
pages 403–409, Nov 1983.
[RST01] Yaron Riany, Nir Shavit, and Dan Touitou. Towards a practical
snapshot algorithm. Theor. Comput. Sci., 269(1-2):163–201,
2001.
[Rup00] Eric Ruppert. Determining consensus numbers. SIAM J. Com-
put., 30(4):1156–1168, 2000.
[Sch95] Eric Schenk. Faster approximate agreement with multi-writer
registers. In 36th Annual Symposium on Foundations of Com-
puter Science, Milwaukee, Wisconsin, 23-25 October 1995, pages
714–723. IEEE Computer Society, 1995.
[Spe28] E. Sperner. Neuer Beweis für die Invarianz der Dimensions-
zahl und des Gebietes. Abhandlungen aus dem Mathema-
tischen Seminar der Universität Hamburg, 6:265–272, 1928.
10.1007/BF02940617.
[SRS08] Christian Scheideler, Andréa W. Richa, and Paolo Santi. An
o(log n) dominating set protocol for wireless ad-hoc networks
under the physical interference model. In Xiaohua Jia, Ness B.
Shroff, and Peng-Jun Wan, editors, Proceedings of the 9th ACM
Interational Symposium on Mobile Ad Hoc Networking and
Computing, MobiHoc 2008, Hong Kong, China, May 26-30,
2008, pages 91–100. ACM, 2008.
[SSW91] M. Saks, N. Shavit, and H. Woll. Optimal time randomized
consensus - making resilient algorithms fast in practice. In Proc.
of the 2nd ACM Symposium on Discrete Algorithms (SODA),
pages 351–362, 1991.
[ST97] Nir Shavit and Dan Touitou. Software transactional memory.
Distributed Computing, 10(2):99–116, 1997.
[SY99] Ichiro Suzuki and Masafumi Yamashita. Distributed anonymous
mobile robots: formation of geometric patterns. SIAM Journal
on Computing, 28(4):1347–1363, 1999.
BIBLIOGRAPHY 477

[SZ00] Michael E. Saks and Fotios Zaharoglou. Wait-free k-set agree-

ment is impossible: The topology of public knowledge. SIAM
J. Comput., 29(5):1449–1483, 2000.

[TV02] John Tromp and Paul M. B. Vitányi. Randomized two-process

wait-free test-and-set. Distributed Computing, 15(3):127–135,
2002.

[VL92] George Varghese and Nancy A. Lynch. A tradeoff between

safety and liveness for randomized coordinated attack protocols.
In Proceedings of the Eleventh Annual ACM Symposium on
Principles of Distributed Computing, PODC ’92, pages 241–250,
New York, NY, USA, 1992. ACM.

[Wel87] Jennifer L. Welch. Simulating synchronous processors. Inf.

Comput., 74(2):159–170, 1987.

[YA95] Jae-Heon Yang and James H. Anderson. A fast, scalable mutual

exclusion algorithm. Distributed Computing, 9(1):51–60, 1995.

[Yu06] Haifeng Yu. Signed quorum systems. Distributed Computing,

18(4):307–323, 2006.

[Zhu16] Leqi Zhu. A tight space bound for consensus. In Daniel Wichs
and Yishay Mansour, editors, Proceedings of the 48th Annual
ACM SIGACT Symposium on Theory of Computing, STOC
2016, Cambridge, MA, USA, June 18-21, 2016, pages 345–350.
ACM, 2016.
Index

2-component max array, 190 adaptive, 208

δ-high-quality quorum, 110 content-oblivious, 60, 208
-agreement, 292 intermediate, 208
-intersecting quorum system, 109 location-oblivious, 208
τ -register, 237 oblivious, 60, 208
b-disseminating quorum system, 108 strong, 208
b-masking quorum system, 109 value-oblivious, 208
f -array, 194 weak, 208
k-ary c-coloring function, 311 aggregator, 205
k-connectivity, 290 agreement, 53, 58, 63, 81
k-neighborhood, 39 -, 292
k-set agreement, 273, 274, 431 k-set, 273, 274, 431
0-valent, 82 approximate, 292, 421
1-valent, 82 Byzantine, 69
2-hop coloring, 333 probabilistic, 211
randomized, 60
absolute coordinates, 326 safe, 267
abstract simplicial complex, 276 simplex, 287
accepter, 86 synchronous, 63
accuracy, 95 alpha synchronizer, 26, 50
action, 444 anonymity, 326
input, 444 anonymous, 30
internal, 444 anti-consensus, 440
output, 444 append, 436
active round, 39 append-and-fetch, 436
adaptive, 225, 233 approximate agreement, 292, 421
adaptive adversary, 208 approximate majority, 321
adaptive collect, 234 arithmetic
admissible, 3, 10 Presburger, 319
adopt-commit object, 210 array
adopt-commit protocol, 210 max, 190
adversary 2-component, 190

478
INDEX 479

ASYNC, 326 buffer, 8

asynchronous, 10 busy-waiting, 117
mobile robots, 326 Byzantine agreement, 69
Asynchronous Computability Theo- weak, 72
rem, 286 Byzantine failure, 3, 69
asynchronous message-passing, 3, 10 Byzantine fault
ATOM, 326 mobile robots, 327
atomic, 120, 240
atomic queue, 4 cache-coherent, 141
atomic register, 4, 113 capacity, 106
atomic snapshot object, 161 CAS, 152
average-case complexity, 36 causal ordering, 41
causal shuffle, 42
bank account, 153 central daemon, 300
barrier Chandra-Toueg consensus protocol,
memory, 4 100
beeping model, 332 channel, 8
beta synchronizer, 26, 51 FIFO, 13
BFS, 23 chemical reaction network, 314
BG simulation, 266 chemical reaction networks, 4
extended, 270 chirality, 326
big-step, 117 chromatic subdivision, 286
binary consensus, 67 class G, 259
biologically inspired systems, 4 client, 11
birthday paradox, 233 client-server, 11
bisimulation, 452 clock
bit complexity, 118 logical, 41
bivalence, 82 Lamport, 44
bivalent, 82 Neiger-Toueg-Welch, 45
mobile robots, 328 phase, 323
Borowsky-Gafni simulation, 266 coherence, 210
bounded, 58 coin
bounded bypass, 127 common, 212
bounded fetch-and-subtract, 433 weak shared, 212
bounded wait-free, 292 collect, 125, 161
breadth-first search, 23 adaptive, 234
broadcast coordinated, 175
ordered partial, 364 coloring
reliable, 101 2-hop, 333
terminating reliable, 104 interval, 333
INDEX 480

coloring function idle, 139

k-ary c-, 311 initial, 8
colorless task, 270 connected, 277
command simply, 290
guarded, 300 consensus, 57, 207
common coin, 212 binary, 67
common node, 76 Chandra-Toeug, 100
common2, 197 randomized, 207
common2 conjecture, 197 synchronous, 63
communication graph, 8 universality of, 158
communication pattern, 66 consensus number, 146
communication register, 300 consensus object, 157
commuting object, 197 consistency property, 116
commuting operations, 150 consistent cut, 46
comparability, 167 consistent scan, 162
comparators, 236 consistent snapshot, 46
compare-and-swap, 4, 119, 152 content-oblivious adversary, 60, 208
comparison-based algorithm, 39 contention, 118, 258
complement, 108 contention management, 258
completeness, 95, 197 contention manager, 248
complex continuous function, 286
input, 277 convergecast, 20
output, 277 convergence, 210
protocol, 287 coordinated attack, 57
complexity randomized, 60
bit, 118 coordinated collect, 175
message, 15 coordinator, 100
obstruction-free step, 250 copy
space, 118 memory-to-memory, 153
step, 14 CORDA, 325
individual, 14, 117 counter, 406
per-process, 117 slow, 370
total, 14, 117 counting network, 236
time, 14, 117 cover, 139
composite register, 162 crash failure, 3, 65
computation event, 8 Crash fault
conciliator, 211 mobile robots, 327
concurrency detector, 407 crashes fully, 66
concurrent execution, 116 critical, 126
configuration, 2, 8, 316 critical section, 126
INDEX 481

CRN, 314 eventually perfect failure detector, 95,

97
daemon eventually strong failure detector, 97,
central, 300 431
distributed, 300 execution, 2, 9, 445
synchronous, 300 concurrent, 116
deadlock, 127 fair, 446
decidability, 319 execution segment, 8
decision bit, 210 exiting, 126
delivery event, 8 exponential information gathering, 73
dense, 324 extended BG simulation, 270
depth
sorting network, 236 failure
deque, 251 Byzantine, 3, 69
detector crash, 3, 65
failure, 94 omission, 3
deterministic, 9, 30 failure detector, 3, 92, 94
deterministic renaming, 226 eventually perfect, 95, 97
diameter, 24 eventually strong, 97, 431
direct scan, 162, 166 perfect, 97
disk, 289 strong, 97
distributed breadth-first search, 19 failure probability
distributed shared memory, 120, 142 quorum system, 106
distributed shared-memory, 4 fair, 446
distributed systems, 1 fair execution, 446
diverging, 329 fairness, 5, 10
downward validity, 167 global, 314
dual graph, 108 fast path, 137
dynamic transaction, 240 fat robot, 326
fault-tolerance
eccentricity, 380 quorum system, 106
Elias gamma code, 188 faulty, 65
enabled, 3, 300, 444 fence, 4
equivalence relation, 30 memory, 4
event, 2, 8 fetch-and-add, 119, 151
computation, 8 fetch-and-cons, 119, 152, 160
delivery, 8 fetch-and-increment, 427
receive, 41 fetch-and-subtract
send, 41 bounded, 433
FIFO channel, 13
INDEX 482

fire (Petri net), 315 immediacy, 279

flooding, 16, 23 impossibility, 5, 6
Frankenexecution, 70 independent set, 335
FSYNC, 327 maximal, 335
full-information protocol, 39 indirect scan, 162, 166
fully synchronous indistinguishability, 6, 65
mobile robots, 327 indistinguishability proof, 58
function indistinguishability proofs, 9
k-ary c-coloring, 311 indistinguishable, 58
continuous, 286 individual step complexity, 14, 117
individual work, 117
gamma synchronizer, 51 initial configuration, 8
geometric median, 330 initiator, 23, 316
ghost root, 306 input action, 444
global fairness, 314 input complex, 277
global synchronizer, 49 interfering operations, 150
Gnutella, 17 intermediate adversary, 208
graph internal action, 444
communication, 8 interval, 116
interaction, 314, 316 interval coloring, 333
guard, 300 invariant, 5, 6
guarded command, 300 invariants, 449
handshake, 165 invocation, 115, 120
happens-before, 41, 50 iterated immediate snapshot, 279
hierarchy Iverson bracket, 302
robust, 147 join, 167
wait-free, 146
high quality quorum, 110 König’s lemma, 58
historyless, 415
historyless object, 181, 197 Lamport clock, 44
homeomorphism, 275 lattice, 167
homotopy, 290 lattice agreement, 167
hyperactive, 329 leader election, 15
for population protocols, 317
I/O automaton, 444 learner, 86
identifier left null, 251
local, 380 legal, 299
identity, 31 limit-closed, 449
idle configuration, 139 line
IIS, 279 world, 12
INDEX 483

linear set, 320 geometric, 330

linearizability, 116, 123 median register, 388
linearizable, 116, 120 meet, 167
linearization, 123 memory barrier, 4
linearization point, 116, 164 memory fence, 4
link register, 300 memory stall, 259
liveness, 5, 6 memory-to-memory copy, 153
liveness property, 448, 449 memory-to-memory swap, 153
LL/SC, 152, 243 message complexity, 15
load, 106 message-passing, 3
load balancing, 233 asynchronous, 3, 10
load-linked, 152, 243 semi-synchronous, 3
load-linked/store-conditional, 4, 174, synchronous, 3, 12
243 MIS, 335
load-linked/stored-conditional, 152 model
LOCAL, 307 beeping, 332
local coin, 207 monoid, 321
local identifier, 380 multi-Paxos, 85, 93
local synchronizer, 49 multi-writer multi-reader register, 114
location-oblivious adversary, 208 multi-writer register, 114
lock-free, 247 multiplicity
lockable register, 429 mobile robots, 326
lockout, 127 mutual exclusion, 119, 127
lockout-freedom, 127 mutual exclusion protocol, 126
logical clock, 41, 44
Lamport, 44 Neiger-Toueg-Welch clock, 45
Neiger-Toueg-Welch, 45 net
long-lived renaming, 228, 233 Petri, 315
long-lived strong renaming, 233 network, 8, 10
look-compute-move, 325 chemical reaction, 314
lower bound, 6 counting, 236
overlay, 10
map renaming, 235
simplicial, 285 sorting, 235
max array, 190 node
2-component, 190 common, 76
max minus one, 305 non-blocking, 240, 247
max register, 185 non-triviality, 64
maximal independent set, 335 nonce, 122
median nondeterminism, 1
INDEX 484

nondeterministic solo termination, 222participating set, 280, 287

null path
left, 251 null, 290
right, 251 Paxos, 85
null path, 290 per-process step complexity, 117
null-homotopic, 290 per-process work, 117
number perfect failure detector, 97
port, 313 period, 334
sequence, 13 persona, 218
Perturbable, 181
object, 113 perturbable, 181, 183
commuting, 197 Petri net, 315
historyless, 181, 197 phase, 323
resilient, 179 phase clock, 323
ring buffer, 437 phase king, 77
snapshot, 161, 179 place, 315
swap, 182 population, 316
oblivious population protocol, 314, 315
mobile robots, 326 population protocols, 4
oblivious adversary, 60, 208 port number, 313
obstruction-free, 247 preference, 295
obstruction-free step complexity, 250 prefix code, 188
Oliver Twist, 5 prefix-closed, 449
omission failure, 3 Presburger arithmetic, 319
one-time building blocks, 135 probabilistic agreement, 211
operations probabilistic quorum system, 109
commuting, 150 probabilistic termination, 207
interfering, 150 process, 8
overwriting, 150 product, 450
oracle, 251 progress, 127
order-equivalent, 38 progress function, 449
order-preserving renaming, 226 progress measure, 6
ordered partial broadcast, 364 proof
ordered set register, 385 impossibility, 6
ordering invariant, 6
causal, 41 liveness, 6
output action, 444 lower bound, 6
output complex, 277 safety, 6
overlay network, 10 termination, 6
overwriting operations, 150 property
INDEX 485

stable, 48 atomic, 4, 113

proposer, 86 communication, 300
protocol composite, 162
population, 314, 315 link, 300
protocol complex, 287 lockable, 429
max, 185
quantifier elimination, 319 median, 388
queue multi-writer, 114
atomic, 4 ordered set, 385
wait-free, 151 second-to-last, 393
with peek, 152 set, 385
quiesce, 18 single-reader, 115
quiescent, 38, 445 single-writer, 114
quorum tamper-proof, 376
δ-high-quality, 110 relation
high quality, 110 simulation, 451
quorum size, 106 reliable broadcast, 101
quorum system, 105 terminating, 104
-intersecting, 109 remainder, 126
b-disseminating, 108 remote memory reference, 118, 141
b-masking, 109 renaming, 136, 224
probabilistic, 109 deterministic, 226
signed, 111 long-lived, 228
strict, 109 long-lived strong, 233
order-preserving, 226
racing counters, 249
strong, 225, 233
Ramsey theory, 40
tight, 225
Ramsey’s Theorem, 40
renaming network, 235
randomization, 3, 31
replica, 93
randomized agreement, 60
replicated state machine, 93
randomized consensus, 207
representative, 219
randomized coordinated attack, 60
request, 11
randomized splitter, 234
reset, 128
RatRace, 234
ReShuffle, 235
reaction network
resilience, 179
chemical, 314
resilient object, 179
read, 120
responder, 316
read-modify-write, 118, 127
response, 11, 115, 120
receive event, 41
restriction, 9
register, 113, 120
right null, 251
INDEX 486

ring, 30 server, 11
ring buffer object, 437 session, 54
RMR, 118, 141 session problem, 54
RMW, 118 set
robot indepedent, 335
fat, 326 linear, 320
robust hierarchy, 147 maximal indepedent, 335
rock-paper-scissors object, 352 semilinear, 320
root set register, 385
ghost, 306 ordered, 385
roster, 373 shared coin
round, 12, 89, 117 weak, 212
shared memory, 3
safe agreement, 267 distributed, 120
safety, 5 shutdown mechanism, 362
safety property, 6, 447, 448 sifter, 216
scan signature, 444
direct, 162, 166 signed quorum system, 111
indirect, 162, 166 silent self-stabilization, 299
schedule, 3, 9 similar, 39, 41
admissible, 10 simplex, 273, 275, 276
second-to-last register, 393 simplex agreement, 287
self-stabilization, 4, 299 simplicial complex, 275
silent, 299 abstract, 276
self-stabilizing, 299 simplicial map, 285
semi-lattice, 168 simply connected, 290
semi-synchronous simulation, 5, 448
mobile robots, 326 simulation relation, 451
semi-synchronous message-passing, 3 single-writer multi-reader register, 114
semilinear set, 320 single-writer register, 114
semisynchrony single-writer single-reader register, 115
unknown-bound, 253 slow counter, 370
send event, 41 slow path, 137
sense of direction, 30 snapshot, 161
mobile robots, 326 snapshot object, 179
sense of scale software transactional memory, 240
mobile robots, 326 solo termination, 181
sequence number, 13 solo-terminating, 181, 247
sequential consistency, 116 sorting network, 235
sequential execution, 120 space complexity, 118
INDEX 487

special action, 54 alpha, 26, 50

species, 314 beta, 26, 51
Sperner’s Lemma, 274 gamma, 51
sphere, 289 global, 49
splitter, 135 local, 49
randomized, 234 synchronizers, 5
splitters, 229 synchronous agreement, 63
spread, 293 synchronous daemon, 300
SSYNC, 326 synchronous message-passing, 3, 12
stable, 299
stable property, 48 tamper-proof register, 376
stably computable, 316 task
stall, 259 colorless, 270
starvation, 5 terminating reliable broadcast, 104
state, 2, 8 termination, 6, 58, 63, 81, 207
static transaction, 240 solo, 181
step complexity, 14 test-and-set, 118, 128, 151, 198
individual, 117 tight renaming, 225
obstruction-free, 250 time complexity, 14, 117
per-process, 117 time-to-live, 17
total, 117 timestamp, 43
sticky bit, 119, 152 token (Petri net), 315
two-writer, 409 torus, 419
sticky register, 119 total step complexity, 14, 117
STM, 240 total work, 117
store-conditional, 153, 243 trace, 445
strict quorum system, 109 trace property, 448
strong adversary, 208 transaction, 240
strong failure detector, 97 dynamic, 240
strong renaming, 225, 233 static, 240
subdivision, 279 transactional memory
chromatic, 286 software, 240
suspect, 94 transition, 316
Suzuki-Yamashita model, 325 transition (Petri net), 315
swap, 151 transition function, 9
memory-to-memory, 153 transition relation, 444
swap object, 182 transitive closure, 42
symmetry, 30 triangulation, 280
symmetry breaking, 30 trying, 126
synchronizer, 12, 23, 44, 49 Twist
INDEX 488

Oliver, 5
Two Generals, 5, 57
two-writer sticky bit, 409

unidirectional ring, 31
uniform, 36
univalent, 82
universality of consensus, 158
unknown-bound semisynchrony, 253
unsafe, 267
upward validity, 167

validity, 58, 63, 81

downward, 167
upward, 167
value-oblivious adversary, 208
values, 210
vector clock, 44, 46

wait-free, 115, 146, 247

bounded, 292
wait-free hierarchy, 146
wait-free queue, 151
wait-freedom, 168, 247
weak adversary, 208
weak Byzantine agreement, 72
weak shared coin, 212
weird condition, 262
width, 118
sorting network, 236
wire, 235
work
individual, 117
per-process, 117
total, 117
world line, 12
write, 120