An Adaptive Sensor Network Architecture for Multi-scale Communication

Santashil PalChaudhuri PhD Proposal

May 12, 2005

Sensor Node Applications

Sensor nodes can sense, process and communicate A few applications

Source: www.intel.com

Distributed surveillance Preventive maintenance Environment monitoring Industrial automation Smart Buildings Pervasive computing
2

Computer Science | Rice University | PhD Proposal

What is Multi-scale?

Multi-scale implies that data is sensed, processed and communicated at different scales Many sensor applications require support for multi-scale approach

Collaborative Monitoring, Wavelet Compression

Illustrative example follow

Computer Science | Rice University | PhD Proposal 3

Multi-scale Monitoring

Computer Science | Rice University | PhD Proposal

Finer Scale Drilling

Computer Science | Rice University | PhD Proposal

Thesis Contributions

Propose a COMPASS networking architecture Design and evaluate specific protocols of this architecture to support multi-scale applications
1. 2. 3.

Data Service Synchronization Medium Access

Computer Science | Rice University | PhD Proposal

1. Architecture

COMPASS Networking Architecture

Requirements Design

Computer Science | Rice University | PhD Proposal

Traditional Network Architectures

Layering is a design principle, and is important for complex systems

Each layer is independent of another layer Service provided by one layer is solely dependent on service by layer directly below

Applications and networking are developed separately with no interaction

Computer Science | Rice University | PhD Proposal

Sensor Networks Characteristics

Resource Constraints

Energy, computational and communication capability Node is not important, content is Limited set of application requirements Hierarchy is a well-known way for scalability In-network processing (E.g,: Averaging)
Computer Science | Rice University | PhD Proposal 9

Data-Centric Routing

Adaptability to applications

Large Scale

Data Fusion Capability

COMPASS Design Goals

1.

Cross-layering for relevant information

Example: MAC knows communication schedule

2.

Adaptable for specific application

Example: Routing has cost and latency tradeoff

3.

Network Programming Interface (NPI)

Example: Logical naming instead of physical naming

Computer Science | Rice University | PhD Proposal

10

COMPASS Network Architecture

Application Synchronization Service Localization Service
11

Data Fusion Data Service Medium Access

Information Exchange Service (IES) Radio

Computer Science | Rice University | PhD Proposal

COMPASS Network Architecture

Enables multi-scale application support Application Adaptive clock Synchronization Publishes routing hierarchy Subscribes to synchronization and NPI needs To be published at DCOSS, Data Fusion Published at IPSN, April 2004 Information June 2005 Exchange Data Service Scheduled medium access Service Publish-Subscribe based for the multi-scale data (IES) notification service Medium Access service Modules register with Subscribes to hierarchy, interest in a specific resource Radio NPI, fusion functions E.g.: App -> MAC To be submitted

Synchronization Service

Computer Science | Rice University | PhD Proposal

Localization Service

12

2. Multi-Scale Data Service

Related Work Hierarchical Overlay Network Programming Interfaces

Computer Science | Rice University | PhD Proposal

13

Related Work

Traditional sensor routing One Scale

Directed Diffusion, Trickle Dimensions, Fractional Cascading Regular grid structure, needs localization Abstract Region neighborhood communication

Multi-scale approaches

Communication Abstractions

Computer Science | Rice University | PhD Proposal

14

COMPASS Hierarchy Formation

Similar to Landmark, L+, Safari 1. 2. 3.

Nodes choose to become cellheads Starts broadcasting beacons Other nodes associate with the heads Repeat this process hierarchically
Choose next hop to parent randomly Heterogeneous or random

4.

Computer Science | Rice University | PhD Proposal

15

Adaptivity using Selectors

Hop proximity is relevant for some applications But, collaborative set of nodes might be based on other criterion sensors with similar value might need to coordinate more often Beacons have selectors associated with it

Selector = (attribute, value, operator) Selectors = Selector (AND|OR) Selector Nodes associates with cluster-heads if Selectors evaluate to
TRUE

Computer Science | Rice University | PhD Proposal

16

Communication Interfaces

Get & Put support for both types

Low-rate data generation, high interest => put High-rate data generation, low interest => get Parent, Cell, Peer
PutPeer(Level,Selectors,Attribute,Value)

3 different kinds of addressing

Reduction interface

Max. temp in cell

Computer Science | Rice University | PhD Proposal 17

Implementation of Interfaces

With the parent node

Reverse path of beacon packet of parent Reverse path of beacon packet to peers Optimal cell flooding using approximate Connected Dominating Set (CDS)

With peer nodes

Within the cell

Computer Science | Rice University | PhD Proposal

18

4. Adaptive Synchronization

Overview

Computer Science | Rice University | PhD Proposal

19

Clock Synchronization Service

Important service in sensor networks Applications

Medium access scheduling Sensor reading fusion Synchronized energy savings Time-sensitive routing

All related works provide best possible syncronization

Computer Science | Rice University | PhD Proposal 20

Need for Adaptation

Synchronization requires messaging, and hence energy

Provide only necessary synchronization

Different applications have different synchronization requirements

Apps publish requirements to IES, and IES notifies synchronization service of common requirement
Computer Science | Rice University | PhD Proposal 21

Summary of work

Given service specification

Maximum Error in synchronization Confidence probability - p

max

Derive receiver-based synchronization protocol parameters to meet these requirements

Number of messages per interval n Interval length - Tsync

Computer Science | Rice University | PhD Proposal 22

4. Hierarchical Medium

Access

Survey of MAC techniques Design of MAC for COMPASS

Computer Science | Rice University | PhD Proposal

23

Fundamental Tasks

Single-hop data transfer Provides efficient access to shared media Controls access to radio

Computer Science | Rice University | PhD Proposal

24

Sensor Network: Requirements and Opportunities

Higher significance: Energy efficiency & scalability Lower significance: Throughput, utilization, latency, fairness Adapt to data patterns using information subscribed from IES

Limited communication abstractions Known fusion function characteristics

Computer Science | Rice University | PhD Proposal 25

Typical approaches

Contention-Based: Random Access & Slotted

9

Flexible with increased traffic fluctuations and node failures or additions Sources of overhead Idle listening, Packet collisions, Control packet overhead Collision-free and no idle listening Reduced flexibility to handle variable traffic and changing neighborhood Combination of both
Computer Science | Rice University | PhD Proposal 26

Schedule-Base
9

Hybrid Approach

Random Access Protocols

Preamble notifies receivers of upcoming transfer Shifts cost from receivers to transmitters Low Power Listening(LPL) and Preamble Sampling [Hill,Micro02][WiseMAC,SenSys03]

Computer Science | Rice University | PhD Proposal 27

Slotted Protocols

Synchronize nodes into slots Implement duty cycle within each slot, which determines energy saved 802.11 style data transfer using RTS/CTS Overhearing avoidance & Streaming sequence of messages[SMAC, Infocom02] Automatically adjusts to fluctuations in traffic [TMAC, SenSys03]
Computer Science | Rice University | PhD Proposal 28

Sink-Based Scheduling

A central base-station, or sink or clusterhead computes the TDMA schedule Sensors inform cluster-heads of traffic demands, which are addressed in the next scheduled frame [IBM Systems Journal,95] Cluster-heads can be rotated to save energy for that node [LEECH, PACT, BMA]
Computer Science | Rice University | PhD Proposal 29

Distributed Scheduling

Nodes know about their

2-hop neighborhood 1-hop future traffic

Sufficient information to choose one transmitter in a collision-free way

Priority given by hash func of node and slot id

Completely general communication assumed, no hierarchy, and no ordering constraints [NAMA, MobiCom01][TRAMA, SenSys03]
Computer Science | Rice University | PhD Proposal 30

Aims for COMPASS Medium Access

Design a medium access schedule to provide energy efficiency for multi-scale applications

Routing hierarchy known Communication pattern known Fusion functions known For Parent and local communication
Computer Science | Rice University | PhD Proposal 31

Optimize the common case

My Assumptions
1.

Bi-directionality of links
True for limited range

2.

Interference range = Transmission range

Techniques such as Radio Interference Detection (RID) [Infocom 05] provides interference range

3.

Clock synchronization present

Provided necessary clock synchronization

4.

Have 2-hop node information

Easier to have than mobile networks
Computer Science | Rice University | PhD Proposal 32

Time-Frame Design
Time-frame divided into two parts Synchronized Contention Free Period (CFP)

Periodic deterministic application traffic Example: Periodic sample of temperature Event-driven traffic Example: Intrusion detection, Fire, etc

Event-driven Contention Access Period (CAP)

I design a protocol for CFP, and any random access technique can be used for CAP
Computer Science | Rice University | PhD Proposal 33

Hierarchical Communication
Intra-Cluster (L1)

Inter-Cluster (L2, L3)

Computer Science | Rice University | PhD Proposal

34

Time Frame Structure

Frame Beacons

L1
Clock Sync

L2 L3

Time Contention Access Period

Contention Free Period

Computer Science | Rice University | PhD Proposal

35

Intra-cluster Structure
a

Hierarchy is achieved by beacons

Computer Science | Rice University | PhD Proposal 36

Design Goals
1.

2.

Spatially conflict-free slot allocation in 2hop neighborhood Parents allocated slots after children,
For latency reduction Avoids switching states
Partial Order Depth-First Search

3.

Slots of children grouped together Fusion function characteristics known

4.

No. of slots to allocate per node

Computer Science | Rice University | PhD Proposal 37

Centralized Algorithm - I

DFS gives a topological sort Conflict-free But, inefficient

b 4 c 1 d 2 e 3 g 5

f 7 j 8

i 10

h 6

k 9

Computer Science | Rice University | PhD Proposal

38

Centralized Algorithm - II

Complete graph knowledge Allocate minimum slotnumber without conflict Slotparent > max(slotchild)
b 4 c 1 g 1

f 5 j 3

i 6

d 2

Assume complete graph knowledge Inter-cluster interference not accounted

e 3

h 2

k 1

Computer Science | Rice University | PhD Proposal

39

Distributed Protocol

Token sent in DFS-order Parent sends token to each child, child send back Nodes know its 2-hop neighborhood b 4
c 1 d 2 e 3 g 1

f 5 j 3

i 6

h 2

k 1

Computer Science | Rice University | PhD Proposal

40

Inter-cluster Interference

To support parent->child & child->child communication If node j was near node g, it cannot know about 2/3 All nodes locally broadcast after sub-tree is done Node j now knows about allocated slots

f 3 1 g m 2

Computer Science | Rice University | PhD Proposal

41

Fusion Function Characteristics

Nodes know about fusion function through IES Determines no. of slots to allocate for forwarding Fusion functions assumed to be simple Does not depend on input, or has a bound

3,4,5 (all data) f 3 (maximum) 1 g 2

Computer Science | Rice University | PhD Proposal

42

Similarity with graph coloring

Network is taken as graph, G 2 Graph G is formed

V: same vertex set as G E: 2-hop connected in G

2

Minimal graph coloring of G is similar to my scheduling NP-hard problem, so approximate chromaticity of graph is relevant
Computer Science | Rice University | PhD Proposal 43

Thesis Outline

Introduction An Adaptive Sensor Network Architecture Multi-scale Data Service [Published]

TODO: Toy applications with interfaces, for MAC evaluation, as well as to demonstrate ease of programmability TODO: Complete design description. Simulate and jointly evaluate with routing

Hierarchical Medium Access [Work in Progress]

Adaptive Clock Synchronization [Published] Conclusion

Computer Science | Rice University | PhD Proposal 44

The End
Comments
Questions

Feedback

Critique

Computer Science | Rice University | PhD Proposal

45