Data-Intensive Text Processing with MapReduce

Data-Intensive Text Processing
with MapReduce
Tutorial at the 32nd Annual International ACM SIGIR Conference on Research
and Development in Information Retrieval (SIGIR 2009)

Jimmy Lin
The iSchool
University of Maryland

Sunday, July 19, 2009

This work is licensed under a Creative Commons Attribution-Noncommercial-Share Alike 3.0 United States
See http://creativecommons.org/licenses/by-nc-sa/3.0/us/ for details. PageRank slides adapted from slides by
Christophe Bisciglia, Aaron Kimball, & Sierra Michels-Slettvet, Google Distributed Computing Seminar, 2007
(licensed under Creation Commons Attribution 3.0 License)

Who am I?

1

Why big data?
Information retrieval is fundamentally:
Experimental and iterative
Concerned with solving real-world problems
“Big data” is a fact of the real world
Relevance of academic IR research hinges on:
The extent to which we can tackle real-world problems
The extent to which our experiments reflect reality

How much data?
Google processes 20 PB a day (2008)
Wayback Machine has 3 PB + 100 TB/month (3/2009)
Facebook has 2.5 PB of user data + 15 TB/day (4/2009)
f / ( / )
eBay has 6.5 PB of user data + 50 TB/day (5/2009)
CERN’s LHC will generate 15 PB a year (??)

640K ought to be
g
enough for anybody.

2

No data like more data!
s/knowledge/data/g;

How do we get here if we’re not Google?

(Banko and Brill, ACL 2001)
(Brants et al., EMNLP 2007)

Academia vs. Industry
“Big data” is a fact of life
Resource gap between academia and industry
Access to computing resources
Access to data
This is changing:
Commoditization of data-intensive cluster computing
Availability of large datasets for researchers

3

MapReduce
e.g., Amazon Web Services

cheap commodity clusters (or utility computing)
+ simple distributed programming models
+ availability of large datasets
= data-intensive IR research for the masses!

ClueWeb09

ClueWeb09
NSF-funded project, led by Jamie Callan (CMU/LTI)
It’s big!
1 billion web pages crawled in Jan /Feb 2009
Jan./Feb.
10 languages, 500 million pages in English
5 TB compressed, 25 uncompressed
It’s available!
Available to the research community
Test collection coming (TREC 2009)

4

Ivory and SMRF
Collaboration between:
Yahoo! Research
Reference implementation for a Web-scale IR toolkit
Designed around Hadoop from the ground up
Written specifically for the ClueWeb09 collection
Implements some of the algorithms described in this tutorial
Features SMRF query engine based on Markov Random Fields
Open source
Initial release available now!

Cloud9
Set of libraries originally developed for teaching
MapReduce at the University of Maryland
Demos, exercises, etc.
“Eat you own dog food”
Actively used for a variety of research projects

5

Topics: Morning Session
Why is this different?
Introduction to MapReduce
Graph algorithms
G
MapReduce algorithm design
Indexing and retrieval
Case study: statistical machine translation
Case study: DNA sequence alignment
Concluding thoughts

Topics: Afternoon Session
Hadoop “Hello World”
Running Hadoop in “standalone” mode
Running Hadoop in distributed mode
Running Hadoop on EC2
Hadoop “nuts and bolts”
Hadoop ecosystem tour
Exercises and “office hours”
office hours

6

Graph algorithms
g
Concluding thoughts

Divide and Conquer

“Work”
Partition

w1 w2 w3

“worker” “worker” “worker”

r1 r2 r3

“Result” Combine

7

It’s a bit more complex…
Fundamental issues
scheduling, data distribution, synchronization, Different programming models
inter-process communication, robustness, fault Message Passing Shared Memory
tolerance, …

y
Memory
P1 P2 P3 P4 P5 P1 P2 P3 P4 P5
Architectural issues
Flynn’s taxonomy (SIMD, MIMD, etc.),
network typology, bisection bandwidth
UMA vs. NUMA, cache coherence Different programming constructs
mutexes, conditional variables, barriers, …
masters/slaves, producers/consumers, work queues, …
Common problems
livelock, deadlock, data starvation, priority inversion…
dining philosophers, sleeping barbers, cigarette smokers, …

The reality: programmer shoulders the burden
of managing concurrency…

Source: Ricardo Guimarães Herrmann

8

Source: MIT Open Courseware

Source: MIT Open Courseware

9

Source: Harper’s (Feb, 2008)


Graph algorithms
y
Concluding thoughts

10

Typical Large-Data Problem
Iterate over a large number of records
Extract something of interest from each
Shuffle
S ff and sort intermediate results
Aggregate intermediate results
Generate final output

Key idea: provide a f
K id id functional abstraction f these
ti l b t ti for th
two operations

(Dean and Ghemawat, OSDI 2004)

MapReduce ~ Map + Fold from functional programming!

Map f f f f f

Fold g g g g g

11

MapReduce
Programmers specify two functions:
map (k, v) → <k’, v’>*
reduce (k’, v’) → <k’, v’>*
All values with th same k are reduced t
l ith the key d d together
th
The runtime handles everything else…

k1 v1 k2 v2 k3 v3 k4 v4 k5 v5 k6 v6

map map map map

a 1 b 2 c 3 c 6 a 5 c 2 b 7 c 8

Shuffle and Sort: aggregate values by keys
a 1 5 b 2 7 c 2 3 6 8

reduce reduce reduce

r1 s1 r2 s2 r3 s3

12

MapReduce
Programmers specify two functions:
map (k, v) → <k’, v’>*
reduce (k’, v’) → <k’, v’>*
All values with th same k are reduced t
l ith the key d d together
th
The runtime handles everything else…
Not quite…usually, programmers also specify:
partition (k’, number of partitions) → partition for k’
Often a simple hash of the key, e.g., hash(k’) mod n
Divides up key space for parallel reduce operations
combine (k’, v’) → <k’, v’>*
Mini-reducers that run in memory after the map phase
Used as an optimization to reduce network traffic

k1 v1 k2 v2 k3 v3 k4 v4 k5 v5 k6 v6

map map map map

a 1 b 2 c 3 c 6 a 5 c 2 b 7 c 8

combine combine combine combine

a 1 b 2 c 9 a 5 c 2 b 7 c 8

partitioner partitioner partitioner partitioner

a 1 5 b 2 7 c 2 9 8

reduce reduce reduce

r1 s1 r2 s2 r3 s3

13

MapReduce Runtime
Handles scheduling
Assigns workers to map and reduce tasks
Handles “data distribution”
Moves processes to data
Handles synchronization
Gathers, sorts, and shuffles intermediate data
Handles faults
Detects worker failures and restarts
Everything h
E thi happens on t of a di t ib t d FS (l t )
top f distributed (later)

“Hello World”: Word Count

Map(String docid, String text):
for
f each word w i text:
h d in
Emit(w, 1);

Reduce(String term, Iterator<Int> values):
int sum = 0;
for each v in values:
sum += v;
Emit(term, value);

14

MapReduce Implementations
MapReduce is a programming model
Google has a proprietary implementation in C++
Bindings in Java Python
Java,
Hadoop is an open-source implementation in Java
Project led by Yahoo, used in production
Rapidly expanding software ecosystem

User
Program

(1) fork (1) fork (1) fork

Master

(2) assign map
(2) assign reduce

worker
split 0
(6) write output
split 1 (5) remote read worker
(3) read file 0
split 2 (4) local write
worker
split 3
split 4 output
worker
file 1

worker

Input Map Intermediate files Reduce Output
files phase (on local disk) phase files

Redrawn from (Dean and Ghemawat, OSDI 2004)

15

How do we get data to the workers?
NAS

SAN

Compute Nodes

What’s the problem here?

Distributed File System
Don’t move data to workers… move workers to the data!
Store data on the local disks of nodes in the cluster
Start up the workers on the node that has the data local
Why?
Not enough RAM to hold all the data in memory
Disk access is slow, but disk throughput is reasonable
A distributed file system is the answer
GFS (Google File System)
HDFS for Hadoop (= GFS clone)

16

GFS: Assumptions
Commodity hardware over “exotic” hardware
Scale out, not up
High component failure rates
Inexpensive commodity components fail all the time
“Modest” number of HUGE files
Files are write-once, mostly appended to
Perhaps concurrently
Large streaming reads over random access
High sustained throughput over low latency

GFS slides adapted from material by (Ghemawat et al., SOSP 2003)

GFS: Design Decisions
Files stored as chunks
Fixed size (64MB)
Reliability through replication
Each chunk replicated across 3+ chunkservers
Single master to coordinate access, keep metadata
Simple centralized management
No data caching
Little benefit due to large datasets, streaming reads
Simplify the API
Push some of the issues onto the client

17

Application GFS master
(file name, chunk index) /foo/bar
GSF Client File namespace chunk 2ef0
(chunk handle, chunk location)

Instructions to chunkserver

Chunkserver state
(chunk handle, byte range)
GFS chunkserver GFS chunkserver
chunk data
Linux file system Linux file system

… …

Redrawn from (Ghemawat et al., SOSP 2003)

Master’s Responsibilities
Metadata storage
Namespace management/locking
Periodic communication with chunkservers
Chunk creation, re-replication, rebalancing
Garbage collection

18

Questions?


Graph Algorithms
y q g
Concluding thoughts

19

Graph Algorithms: Topics
Introduction to graph algorithms and graph
representations
Single Source Shortest Path (SSSP) problem
Refresher: Dijkstra’s algorithm
Breadth-First Search with MapReduce
PageRank

What’s a graph?
G = (V,E), where
V represents the set of vertices (nodes)
E represents the set of edges (links)
Both vertices and edges may contain additional information
Different types of graphs:
Directed vs. undirected edges
Presence or absence of cycles
...

20

Some Graph Problems
Finding shortest paths
Routing Internet traffic and UPS trucks
Finding minimum spanning trees
Telco laying down fiber
Finding Max Flow
Airline scheduling
Identify “special” nodes and communities
Breaking up terrorist cells, spread of avian flu
Bipartite matching
Monster.com, Match.com
And of course... PageRank

Representing Graphs
G = (V, E)
Two common representations
Adjacency matrix
Adjacency list

21

Adjacency Matrices
Represent a graph as an n x n square matrix M
n = |V|
Mijj = 1 means a link from node i to j

2
1 2 3 4
1 0 1 0 1 1

2 1 0 1 1 3

3 1 0 0 0
4 1 0 1 0 4

Adjacency Lists
Take adjacency matrices… and throw away all the zeros

1 2 3 4
1 0 1 0 1 1: 2, 4
2: 1, 3, 4
2 1 0 1 1
3: 1
3 1 0 0 0 4: 1, 3
1
4 1 0 1 0

22

Single Source Shortest Path
Problem: find shortest path from a source node to one or
more target nodes
First,
First a refresher: Dijkstra s Algorithm
Dijkstra’s

Dijkstra’s Algorithm Example

∞ 1 ∞

10

0 2 3 9 4 6

5 7

∞ ∞
2

Example from CLR

23


10 1 ∞

10

0 2 3 9 4 6

5 7

5 ∞
2

Example from CLR


8 1 14

10

0 2 3 9 4 6

5 7

5 7
2

Example from CLR

24


8 1 13

10

0 2 3 9 4 6

5 7

5 7
2

Example from CLR


8 1 9

10

0 2 3 9 4 6

5 7

5 7
2

Example from CLR

25


8 1 9

10

0 2 3 9 4 6

5 7

5 7
2

Example from CLR

Single Source Shortest Path
Problem: find shortest path from a source node to one or
more target nodes
Single processor machine: Dijkstra s Algorithm
Dijkstra’s
MapReduce: parallel Breadth-First Search (BFS)

26

Finding the Shortest Path
Consider simple case of equal edge weights
Solution to the problem can be defined inductively
Here’s the intuition:
DISTANCETO(startNode) = 0
For all nodes n directly reachable from startNode,
DISTANCETO (n) = 1
For all nodes n reachable from some other set of nodes S,
DISTANCETO(n) = 1 + min(DISTANCETO(m), m ∈ S)

cost1 m1
…
cost2
… n
m2

… cost3
m3

From Intuition to Algorithm
Mapper input
Key: node n
Value: D (distance from start), adjacency list (list of nodes
reachable from n)
Mapper output
∀p ∈ targets in adjacency list: emit( key = p, value = D+1)
The reducer gathers possible distances to a given p and
selects the minimum one
Additional b kk
Additi l bookkeeping needed t k
i d d to keep t k of actual path
track f t l th

27

Multiple Iterations Needed
Each MapReduce iteration advances the “known frontier”
by one hop
Subsequent iterations include more and more reachable nodes as
frontier expands
Multiple iterations are needed to explore entire graph
Feed output back into the same MapReduce task
Preserving graph structure:
Problem: Where did the adjacency list go?
Solution: mapper emits (n, adjacency list) as well
pp ( j y )

Visualizing Parallel BFS

3
1 2

2 2
3

3
3
4

4

28

Weighted Edges
Now add positive weights to the edges
Simple change: adjacency list in map task includes a
weight w for each edge
emit (p, D+wp) instead of (p, D+1) for each node p

Comparison to Dijkstra
Dijkstra’s algorithm is more efficient
At any step it only pursues edges from the minimum-cost path
inside the frontier
MapReduce explores all paths in parallel

29

Random Walks Over the Web
Model:
User starts at a random Web page
User randomly clicks on links, surfing from page to page
PageRank = the amount of time that will be spent on any
given page

PageRank: Defined
Given page x with in-bound links t1…tn, where
C(t) is the out-degree of t
α is probability of random jump
N is the total number of nodes in the graph

⎛1⎞ n
PR(ti )
PR ( x) = α ⎜ ⎟ + (1 − α )∑
⎝N⎠ i =1 C (ti )

t1

X

t2

…
tn

30

Computing PageRank
Properties of PageRank
Can be computed iteratively
Effects at each iteration is local
Sketch of algorithm:
Start with seed PRi values
Each page distributes PRi “credit” to all pages it links to
Each target page adds up “credit” from multiple in-bound links to
compute PRi+1
Iterate until values converge
g

PageRank in MapReduce

Map: distribute PageRank “credit” to link targets

Reduce: gather up PageRank “credit” from multiple sources
to compute new PageRank value

Iterate until
convergence

...

31

PageRank: Issues
Is PageRank guaranteed to converge? How quickly?
What is the “correct” value of α, and how sensitive is the
algorithm to it?
What about dangling links?
How do you know when to stop?

Graph Algorithms in MapReduce
General approach:
Store graphs as adjacency lists
Each map task receives a node and its adjacency list
Map task compute some function of the link structure, emits value
with target as the key
Reduce task collects keys (target nodes) and aggregates
Perform multiple MapReduce iterations until some
termination condition
Remember to “pass” g p structure from one iteration to next
p graph

32

Questions?

Graph algorithms

MapReduce Algorithm Design
Concluding thoughts
g g

33

Managing Dependencies
Remember: Mappers run in isolation
You have no idea in what order the mappers run
You have no idea on what node the mappers run
You have no idea when each mapper finishes
Tools for synchronization:
Ability to hold state in reducer across multiple key-value pairs
Sorting function for keys
Partitioner
Cleverly constructed
Cleverly-constructed data structures

Slides in this section adapted from work reported in (Lin, EMNLP 2008)

Motivating Example
Term co-occurrence matrix for a text collection
M = N x N matrix (N = vocabulary size)
Mijj: number of times i and j co-occur in some context
(for concreteness, let’s say context = sentence)
Why?
Distributional profiles as a way of measuring semantic distance
Semantic distance useful for many language processing tasks

34

MapReduce: Large Counting Problems
Term co-occurrence matrix for a text collection
= specific instance of a large counting problem
A large event space (number of terms)
A large number of observations (the collection itself)
Goal: keep track of interesting statistics about the events
Basic approach
Mappers generate partial counts
Reducers aggregate partial counts

How do we aggregate partial counts efficiently?

First Try: “Pairs”
Each mapper takes a sentence:
Generate all co-occurring term pairs
For all pairs, emit (a, b) → count
Reducers sums up counts associated with these pairs
Use combiners!

35

“Pairs” Analysis
Advantages
Easy to implement, easy to understand
Disadvantages
Lots of pairs to sort and shuffle around (upper bound?)

Another Try: “Stripes”
Idea: group together pairs into an associative array
(a, b) → 1
(a, c) → 2
(a,
(a d) → 5 a → { b: 1, c: 2, d: 5, e: 3, f: 2 }
1 2 5 3
(a, e) → 3
(a, f) → 2

Each mapper takes a sentence:
Generate all co-occurring term pairs
For each term, emit a → { b: countb, c: countc, d: countd … }
Reducers perform element wise sum of associative arrays
element-wise
a → { b: 1, d: 5, e: 3 }
+ a → { b: 1, c: 2, d: 2, f: 2 }
a → { b: 2, c: 2, d: 7, e: 3, f: 2 }

36

“Stripes” Analysis
Advantages
Far less sorting and shuffling of key-value pairs
Can make better use of combiners
Disadvantages
More difficult to implement
Underlying object is more heavyweight
Fundamental limitation in terms of size of event space

Cluster size: 38 cores
Data Source: Associated Press Worldstream (APW) of the English Gigaword Corpus (v3),
which contains 2.27 million documents (1.8 GB compressed, 5.7 GB uncompressed)

37

Conditional Probabilities
How do we estimate conditional probabilities from counts?

count ( A, B) count ( A, B)
P( B | A) = =
count ( A) ∑ count( A, B' )
B'

Why do we want to do this?
How do we do this with MapReduce?

P(B|A): “Stripes”

a → {b1:3, b2 :12, b3 :7, b4 :1, … }

Easy!
One pass to compute (a, *)
Another pass to directly compute P(B|A)

38

P(B|A): “Pairs”

(a, *) → 32 Reducer holds this value in memory

(a, b1) → 3 (a, b1) → 3 / 32
(a,
(a b2) → 12 (a,
(a b2) → 12 / 32
(a, b3) → 7 (a, b3) → 7 / 32
(a, b4) → 1 (a, b4) → 1 / 32
… …

For this to work:
Must emit extra (a, *) for every bn in mapper
)
Must make sure all a’s get sent to same reducer (use partitioner)
Must make sure (a, *) comes first (define sort order)
Must hold state in reducer across different key-value pairs

Synchronization in Hadoop
Approach 1: turn synchronization into an ordering problem
Sort keys into correct order of computation
Partition key space so that each reducer gets the appropriate set
of partial results
Hold state in reducer across multiple key-value pairs to perform
computation
Illustrated by the “pairs” approach
Approach 2: construct data structures that “bring the
pieces together”
Each reducer receives all the data it needs to complete the
computation
Illustrated by the “stripes” approach

39

Issues and Tradeoffs
Number of key-value pairs
Object creation overhead
Time for sorting and shuffling pairs across the network
Size of each key-value pair
De/serialization overhead
Combiners make a big difference!
RAM vs. disk vs. network
Arrange data to maximize opportunities to aggregate partial results

Questions?

40

Graph algorithms

Indexing and Retrieval
Concluding thoughts

Abstract IR Architecture

Query Documents

online offline
Representation Representation
Function Function

Query Representation Document Representation

Comparison
p
Function Index
I d

Hits

41

MapReduce it?
The indexing problem
Scalability is critical
Must be relatively fast, but need not be real time
Fundamentally a batch operation
Incremental updates may or may not be important
For the web, crawling is a challenge in itself
The retrieval problem
Must have sub-second response time
For the web, only need relatively few results

Counting Words…

Documents

case folding, tokenization, stopword removal, stemming

Bag of
Words syntax, semantics, word knowledge, etc.

Inverted
Index

42

Inverted Index: Boolean Retrieval
Doc 1 Doc 2 Doc 3 Doc 4
one fish, two fish red fish, blue fish cat in the hat green eggs and ham

1 2 3 4

blue 1 blue 2

cat 1 cat 3

egg 1 egg 4

fish 1 1 fish 1 2

green 1 green 4

ham 1 ham 4

hat 1 hat 3

one 1 one 1

red 1 red 2

two 1 two 1

Inverted Index: Ranked Retrieval

tf
1 2 3 4 df
blue 1 1 blue 1 2,1

cat 1 1 cat 1 3,1

egg 1 1 egg 1 4,1

fish 2 2 2 fish 2 1,2 2,2

green 1 1 green 1 4,1

ham 1 1 ham 1 4,1

hat 1 1 hat 1 3,1

one 1 1 one 1 1,1

red 1 1 red 1 2,1

two 1 1 two 1 1,1

43

Inverted Index: Positional Information

blue 1 2,1 blue 1 2,1,[3]

cat 1 3,1 cat 1 3,1,[1]

egg 1 4,1 egg 1 4,1,[2]

fish 2 1,2 2,2 fish 2 1,2,[2,4] 1,2,[2,4]

green 1 4,1 green 1 4,1,[1]

ham 1 4,1 ham 1 4,1,[3]

hat 1 3,1 hat 1 3,1,[2]

one 1 1,1 one 1 1,1,[1]

red 1 2,1 red 1 2,1,[1]

two 1 1,1 two 1 1,1,[3]

Indexing: Performance Analysis
Fundamentally, a large sorting problem
Terms usually fit in memory
Postings usually don’t
How is it done on a single machine?
How can it be done with MapReduce?
First, let’s characterize the problem size:
Size of vocabulary
Size of postings

44

Vocabulary Size: Heaps’ Law

M = kT
M is vocabulary size
b T is collection size (number of documents)
k and b are constants
d t t

Typically, k is between 30 and 100, b is between 0.4 and 0.6

Heaps’ Law: linear in log-log space
Vocabulary size grows unbounded!

Heaps’ Law for RCV1

k = 44
b = 0.49

First 1,000,020 terms:
Predicted = 38,323
Actual = 38,365

Reuters-RCV1 collection: 806,791 newswire documents (Aug 20, 1996-August 19, 1997)

Manning, Raghavan, Schütze, Introduction to Information Retrieval (2008)

45

Postings Size: Zipf’s Law

c
cf i = cf is the collection frequency of i-th common term
c is a constant
i th

i

Zipf’s Law: (also) linear in log-log space
Specific case of Power Law distributions
In other words:
A few elements occur very frequently
Many elements occur very infrequently

Zipf’s Law for RCV1

Fit isn’t that good
good…
but good enough!

Reuters-RCV1 collection: 806,791 newswire documents (Aug 20, 1996-August 19, 1997)

Manning, Raghavan, Schütze, Introduction to Information Retrieval (2008)

46

Figure from: Newman, M. E. J. (2005) “Power laws, Pareto
distributions and Zipf's law.” Contemporary Physics 46:323–351.

MapReduce: Index Construction
Map over all documents
Emit term as key, (docno, tf) as value
Emit other information as necessary (e.g., term position)
Sort/shuffle: group postings by term
Reduce
Gather and sort the postings (e.g., by docno or tf)
Write postings to disk
MapReduce does all the heavy lifting!
p y g

47

Inverted Indexing with MapReduce
Doc 1 Doc 2 Doc 3
one fish, two fish red fish, blue fish cat in the hat

one 1 1 red 2 1 cat 3 1

Map two 1 1 blue 2 1 hat 3 1

fish 1 2 fish 2 2


cat 3 1
blue 2 1
Reduce fish 1 2 2 2
hat 3 1
one 1 1
two 1 1
red 2 1

Inverted Indexing: Pseudo-Code

48

You’ll implement this in the afternoon!

Positional Indexes
Doc 1 Doc 2 Doc 3
one fish, two fish red fish, blue fish cat in the hat

one 1 1 [1] red 2 1 [1] cat 3 1 [1]

Map two 1 1 [3] blue 2 1 [3] hat 3 1 [2]

fish 1 2 [2,4] fish 2 2 [2,4]


cat 3 1 [1]
blue 2 1 [3]

Reduce fish 1 2 [2,4] 2 2 [2,4]
hat 3 1 [2]
one 1 1 [1]
two 1 1 [3]
red 2 1 [1]

49

Inverted Indexing: Pseudo-Code

Scalability Bottleneck
Initial implementation: terms as keys, postings as values
Reducers must buffer all postings associated with key (to sort)
What if we run out of memory to buffer postings?
Uh oh!

50

Another Try…
(key) (values) (keys) (values)

fish 1 2 [2,4] fish 1 [2,4]

34 1 [23] fish
fi h 9 [9]

21 3 [1,8,22] fish 21 [1,8,22]

35 2 [8,41] fish 34 [23]

80 3 [2,9,76] fish 35 [8,41]

9 1 [9] fish 80 [2,9,76]

How is this different?
• Let the framework do the sorting
• Term frequency implicitly stored
• Directly write postings to disk!

Wait, there’s more!
(but first, an aside)

51

Postings Encoding
Conceptually:

fish 1 2 9 1 21 3 34 1 35 2 80 3 …

In Practice:
• Don’t encode docnos, encode gaps (or d-gaps)
• But it’s not obvious that this save space…

fish 1 2 8 1 12 3 13 1 1 2 45 3 …

Overview of Index Compression
Non-parameterized
Unary codes
γ codes
δ codes
Parameterized
Golomb codes (local Bernoulli model)

Want more detail? Read Managing Gigabytes by Witten, Moffat, and Bell!

52

Unary Codes
x ≥ 1 is coded as x-1 one bits, followed by 1 zero bit
3 = 110
4 = 1110
Great for small numbers… horrible for large numbers
Overly-biased for very small gaps

Watch out! Slightly different definitions in Witten et al.,
compared to Manning et al. and Croft et al.!

γ codes
x ≥ 1 is coded in two parts: length and offset
Start with binary encoded, remove highest-order bit = offset
Length is number of binary digits, encoded in unary code
Concatenate length + offset codes
Example: 9 in binary is 1001
Offset = 001
Length = 4, in unary code = 1110
γ code = 1110:001
Analysis
Offset = ⎣log x⎦
Length = ⎣log x⎦ +1
Total = 2 ⎣log x⎦ +1

53

δ codes
Similar to γ codes, except that length is encoded in γ code
Example: 9 in binary is 1001
Offset = 001
Length = 4, in γ code = 11000
δ code = 11000:001
γ codes = more compact for smaller numbers
δ codes = more compact for larger numbers

Golomb Codes
x ≥ 1, parameter b:
q + 1 in unary, where q = ⎣( x - 1 ) / b⎦
r in binary, where r = x - qb - 1, in ⎣log b⎦ or ⎡log b⎤ bits
Example:
b = 3, r = 0, 1, 2 (0, 10, 11)
b = 6, r = 0, 1, 2, 3, 4, 5 (00, 01, 100, 101, 110, 111)
x = 9, b = 3: q = 2, r = 2, code = 110:11
x = 9, b = 6: q = 1, r = 2, code = 10:100
Optimal b ≈ 0 69 (N/df)
0.69
Different b for every term!

54

Comparison of Coding Schemes

Unary γ δ Golomb
b=3 b=6

1 0 0 0 0:0 0:00
2 10 10:0 100:0 0:10 0:01
3 110 10:1 100:1 0:11 0:100
4 1110 110:00 101:00 10:0 0:101
5 11110 110:01 101:01 10:10 0:110
6 111110 110:10 101:10 10:11 0:111
7 1111110 110:11 101:11 110:0 10:00
8 11111110 1110:000 11000:000 110:10 10:01
9 111111110 1110:001 11000:001 110:11 10:100
10 1111111110 1110:010 11000:010 1110:0 10:101

Witten, Moffat, Bell, Managing Gigabytes (1999)

Index Compression: Performance

Comparison of Index Size (bits per pointer)

Bible TREC

Unary 262 1918
Binary 15 20
γ 6.51 6.63
δ 6.23 6.38
Golomb 6.09 5.84 Recommend best practice

Bible: King James version of the Bible; 31,101 verses (4.3 MB)
TREC: TREC disks 1+2; 741,856 docs (2070 MB)

Witten, Moffat, Bell, Managing Gigabytes (1999)

55

Chicken and Egg?

(key) (value)

fish 1 [2,4]
But
B t wait! H
it! How d we set th
do t the
fish 9 [9]
Golomb parameter b?
fish 21 [1,8,22]
Recall: optimal b ≈ 0.69 (N/df)
fish 34 [23]
We need the df to set b…
fish 35 [8,41] But we don’t know the df until we’ve
seen all postings!
fish 80 [2,9,76]

…

Write directly to disk

Getting the df
In the mapper:
Emit “special” key-value pairs to keep track of df
In the reducer:
Make sure “special” key-value pairs come first: process them to
determine df

56

Getting the df: Modified Mapper

Doc 1
one fish, two fish Input document…

(key) (value)

fish 1 [2,4] Emit normal key-value pairs…

one 1 [1]

two 1 [3]

fish [1] Emit “special” key-value pairs to keep track of df…

one [1]

two [1]

Getting the df: Modified Reducer
(key) (value)

fish [1]

First, compute the df by summing contributions from
p y g
fish
fi h [1]
all “special” key-value pair…
fish [1]

… Compute Golomb parameter b…
fish 1 [2,4]

fish 9 [9]

fish 21 [1,8,22] Important: properly define sort order to make
sure “
“special” k
i l” key-value pairs come fi t!
l i first!
fish 34 [23]

fish 35 [8,41]

fish 80 [2,9,76]

… Write postings directly to disk

57

MapReduce it?
The indexing problem Just covered
Scalability is paramount
Must be relatively fast, but need not be real time
Fundamentally a batch operation
Incremental updates may or may not be important
For the web, crawling is a challenge in itself
The retrieval problem Now
Must have sub-second response time
For the web, only need relatively few results

Retrieval in a Nutshell
Look up postings lists corresponding to query terms
Traverse postings for each query term
Store partial query-document scores in accumulators
S
Select top k results to return

58

Retrieval: Query-At-A-Time
Evaluate documents one query at a time
Usually, starting from most rare term (often with tf-scored postings)

blue 9 2 21 1 35 1 …
Score{q=x}(doc n) = s Accumulators
(e.g., hash)

fish 1 2 9 1 21 3 34 1 35 2 80 3 …

Tradeoffs
Early termination heuristics (good)
Large memory footprint (bad), but filtering heuristics possible

Retrieval: Document-at-a-Time
Evaluate documents one at a time (score all query terms)
blue 9 2 21 1 35 1 …

fish
fi h 1 2 9 1 21 3 34 1 35 2 80 3 …

Document score in top k?
Accumulators Yes: Insert document score, extract-min if queue too large
(e.g. priority queue) No: Do nothing

Tradeoffs
Small memory footprint (good)
Must read through all postings (bad), but skipping possible
More disk seeks (bad), but blocking possible

59

Retrieval with MapReduce?
MapReduce is fundamentally batch-oriented
Optimized for throughput, not latency
Startup of mappers and reducers is expensive
MapReduce is not suitable for real-time queries!
Use separate infrastructure for retrieval…

Important Ideas
Partitioning (for scalability)
Replication (for redundancy)
Caching (f speed)
C (for )
Routing (for load balancing)

The rest is just details!

60

Term vs. Document Partitioning
D
T1

D T2
Term …
Partitioning
T3

T
Document
Partitioning
P titi i … T

D1 D2 D3

Katta Architecture
(Distributed Lucene)

http://katta.sourceforge.net/

61

Batch ad hoc Queries
What if you cared about batch query evaluation?
MapReduce can help!

Parallel Queries Algorithm
Assume standard inner-product formulation:

score(q, d ) = ∑ wt ,q wt ,d
t∈V
V

Algorithm sketch:
Load queries into memory in each mapper
Map over postings, compute partial term contributions and store in
accumulators
Emit accumulators as intermediate output
p
Reducers merge accumulators to compute final document scores

Lin (SIGIR 2009)

62

Parallel Queries: Map

blue 9 2 21 1 35 1

Mapper query id = 1, “blue fish
blue fish”
Compute score contributions for term

key = 1, value = { 9:2, 21:1, 35:1 }

fish 1 2 9 1 21 3 34 1 35 2 80 3

Mapper query id = 1, “blue fish”

key = 1, value = { 1:2, 9:1, 21:3, 34:1, 35:2, 80:3 }

Parallel Queries: Reduce

key = 1, value = { 9:2, 21:1, 35:1 }
key = 1, value = { 1:2, 9:1, 21:3, 34:1, 35:2, 80:3 }

Reducer Element-wise sum of associative arrays

key = 1, value = { 1:2, 9:3, 21:4, 34:1, 35:3, 80:3 }

Sort accumulators to generate final ranking

Query: “blue fish”
doc 21, score=4
doc 2, score=3
doc 35, score=3
doc 80, score=3
doc 1, score=2
doc 34, score=1

63

A few more details…

fish 1 2 9 1 21 3 34 1 35 2 80 3

Mapper query id = 1 “blue fish
1, blue fish”

key = 1, value = { 1:2, 9:1, 21:3, 34:1, 35:2, 80:3 }

Evaluate multiple queries within each mapper
Approximations by accumulator limiting
Complete independence of mappers makes this problematic

Ivory and SMRF
Collaboration between:
Yahoo! Research
Reference implementation for a Web-scale IR toolkit
Designed around Hadoop from the ground up
Written specifically for the ClueWeb09 collection
Implements some of the algorithms described in this tutorial
Features SMRF query engine based on Markov Random Fields
Open source
Initial release available now!

64

Questions?

Graph l ith
G h algorithms

Case Study:
Statistical Machine Translation
Concluding thoughts

65

Statistical Machine Translation
Conceptually simple:
(translation from foreign f into English e)

e = arg max P( f | e) P(e)
ˆ
e

Difficult in practice!
Phrase-Based Machine Translation (PBMT) :
Break up source sentence into little pieces (phrases)
Translate each phrase individually

Dyer et al. (Third ACL Workshop on MT, 2008)

Translation as a “Tiling” Problem

Maria no dio una bofetada a la bruja verde

Mary not give a slap to the witch green

did not a slap by green witch

no slap to the

did not give to

the

slap the witch

Example from Koehn (2006)

66

MT Architecture

Training Data Word Alignment Phrase Extraction

i saw the small table (vi,
(vi i saw)
vi la mesa pequeña (la mesa pequeña, the small table)
Parallel Sentences
…

he sat at the table Language Translation
the service was good Model Model
Target-Language Text

Decoder

maria no daba una bofetada a la bruja verde mary did not slap the green witch
Foreign Input Sentence English Output Sentence

The Data Bottleneck

67

MT Architecture
There are MapReduce Implementations of
these two components!


(vi i saw)
Parallel Sentences
…


Decoder


HMM Alignment: Giza

Single-core
Single core commodity server

68

HMM Alignment: MapReduce

Single-core
Single core commodity server

38 processor cluster

HMM Alignment: MapReduce

38 processor cluster

1/38 Single-core commodity server

69

MT Architecture
There are MapReduce Implementations of
these two components!


(vi i saw)
Parallel Sentences
…


Decoder


Phrase table construction

Single-core commodity server

70



38 proc. cluster



38 proc. cluster

1/38 of single-core

71

What’s the point?
The optimally-parallelized version doesn’t exist!
It’s all about the right level of abstraction

Questions?

72

Graph algorithms

Case Study:
DNA Sequence Alignment
Concluding thoughts

From Text to DNA Sequences
Text processing: [0-9A-Za-z]+
DNA sequence processing: [ATCG]+

(Nope, not really)

The following describes the work of Michael Schatz; thanks also to Ben Langmead…

73

Analogy
(And two disclaimers)

Strangely-Formatted Manuscript
Dickens: A Tale of Two Cities
Text written on a long spool

It was the best of times, it was the worst of times, it was the age of wisdom, it was the age of foolishness, …

74

… With Duplicates
Dickens: A Tale of Two Cities
“Backup” on four more copies






Shredded Book Reconstruction
Dickens accidently shreds the manuscript

It was the b thef best of times, it was worst
It was of
h best times, it was the the worst f times, itit was h age of wisdom, itit was
h of times, was the
of the age of wisdom, was the age age of foolishness, …
f d the of foolishness, …
h ff l h

It was the best besttimes, it waswas the worst of times, it was the the agewisdom, it was the agethe foolishness, …
It was the of of times, it the age of of wisdom, it was of age of foolishness,

It was the best best of times, it was the worst times, it it was the age of wisdom, it it was the age of foolishness, …
It was the of times, it was the worst of of times, was the age of wisdom, was the age of foolishness, …

It was It the best of times, it was the worst of times, it was the age of wisdom, it was the age of foolishness, … …
was the best of times, it was the worst of times, it was the age of wisdom, it was the age of foolishness,

It was the best best of times, was the worst of of times, it was the age of of wisdom, it was the of foolishness, … …
It was the of times, it it was the worst wisdom, it was the age age of foolishness,

How can he reconstruct the text?
5 copies x 138,656 words / 5 words per fragment = 138k fragments
The short fragments from every copy are mixed together
Some fragments are identical

75

Overlaps
It was the best of
It was the best of
age of wisdom, it was 4 word overlap
was the best of times,
best of times, it was

it was the age of

it was the age of It was the best of
1 word overlap
it was the worst of of times, it was the

of times, it was the

of times, it was the It was the best of
1 word overlap
of wisdom, it was the of wisdom, it was the

the age of wisdom, it

the best of times, it
, Generally prefer longer overlaps to shorter overlaps
the worst of times, it
In the presence of error, we might allow the
times, it was the age
overlapping fragments to differ by a small amount
times, it was the worst

was the age of wisdom,

was the age of foolishness,


th t f ti

Greedy Assembly
It was the best of

age of wisdom, it was
It was the best of
it was the age of
it was the age of
it was the worst of of times, it was the
of times, it was the of times, it was the
of times, it was the times, it was the worst
of wisdom, it was the times, it was the age
the age of wisdom, it

,
The
Th repeated sequence makes th correct
t d k the t
the worst of times, it
reconstruction ambiguous
times, it was the age

times, it was the worst

was the age of wisdom,

was the age of foolishness,


th t f ti

76

The Real Problem
(The easier version)

GATGCTTACTATGCGGGCCCC
CGGTCTAATGCTTACTATGC
GCTTACTATGCGGGCCCCTT
AATGCTTACTATGCGGGCCCCTT

?
TAATGCTTACTATGC
AATGCTTAGCTATGCGGGC
CGGTCTAGATGCTTACTATGC
CGGTCTAATGCTTAGCTATGC
ATGCTTACTATGCGGGCCCCTT

Reads

Subject
genome

Sequencer

77

DNA Sequencing
Genome of an organism encodes genetic information
in long sequence of 4 DNA nucleotides: ATCG
Bacteria: ~5 million bp
Humans: ~3 billion bp
Current DNA sequencing machines can generate 1-2
Gbp of sequence per day, in millions of short reads
(25-300bp)
Shorter reads, but much higher throughput
Per-base error rate estimated at 1-2%
Recent studies of entire human genomes have used
3.3 - 4.0 billion 36bp reads
~144 GB of compressed sequence d t
144 f d data

ATCTGATAAGTCCCAGGACTTCAGT

GCAAGGCAAACCCGAGCCCAGTTT

TCCAGTTCTAGAGTTTCACATGATC

GGAGTTAGTAAAAGTCCACATTGAG

How do we put humpty dumpty back together?

78

Human Genome
A complete human DNA sequence was published in
2003, marking the end of the Human Genome Project

11 years, cost $3 billion… your tax dollars at work!

Subject reads
CTATGCGGGC
CTAGATGCTT AT CTATGCGG
TCTAGATGCT GCTTAT CTAT AT CTATGCGG
AT CTATGCGG AT CTATGCGG
TTA T CTATGC GCTTAT CTAT CTATGCGGGC

Alignment

CGGTCTAGATGCTTAGCTATGCGGGCCCCTT

Reference sequence

79

Subject reads

ATGCGGGCCC
CTAGATGCTT CTATGCGGGC
TCTAGATGCT ATCTATGCGG
CGGTCTAG ATCTATGCGG CTT
CGGTCT TTATCTATGC CCTT
CGGTC GCTTATCTAT GCCCCTT
CGG GCTTATCTAT GGCCCCTT
CGGTCTAGATGCTTATCTATGCGGGCCCCTT

Reference sequence

Reference: ATGAACCACGAACACTTTTTTGGCAACGATTTAT…
Query: ATGAACAAAGAACACTTTTTTGGCCACGATTTAT…

Insertion Deletion Mutation

80

CloudBurst
1. Map: Catalog K‐mers
• Emit every k‐mer in the genome and non‐overlapping k‐mers in the reads
• Non‐overlapping k‐mers sufficient to guarantee an alignment will be found
2. Shuffle: Coalesce Seeds
• Hadoop internal shuffle groups together k‐mers shared by the reads and the reference
• Conceptually build a hash table of k‐mers and their occurrences
3. Reduce: End‐to‐end alignment
• Locally extend alignment beyond seeds by computing “match distance”
• If read aligns end‐to‐end, record the alignment

Map shuffle Reduce
Human chromosome 1

Read 1, Chromosome 1, 12345-12365
…
Read 1

…
Read 2
Read 2, Chromosome 1, 12350-12370

Running Time vs Number of Reads on Chr 1
16000
14000 0
1
12000 2
Runtime (s)

10000 3
8000 4
6000
4000
2000
0
0 2 4 6 8
Millions of Reads

Running Time vs Number of Reads on Chr 22
3000
0
2500
1
Runtime (s)

2000 2
1500 3
4
1000

500

0
0 2 4 6 8
Millions of Reads

Results from a small, 24-core cluster, with different number of mismatches

Michael Schatz. CloudBurst: Highly Sensitive Read Mapping
with MapReduce. Bioinformatics, 2009, in press.

81

Running Time on EC2
High-CPU Medium Instance Cluster
1800
1600
1400

ime (s)
1200
1000

Running ti
800
600
400
200
0
24 48 72 96
Number of Cores

CloudBurst running times for mapping 7M reads to human
Cl dB t i ti f i d t h
chromosome 22 with at most 4 mismatches on EC2

Michael Schatz. CloudBurst: Highly Sensitive Read Mapping
with MapReduce. Bioinformatics, 2009, in press.

Wait, no reference?

82

de Bruijn Graph Construction
Dk = (V,E)
V = All length-k subfragments (k > l)
E = Directed edges between consecutive subfragments
Nodes overlap by k-1 words
k1

Original Fragment Directed Edge
It was the best of It was the best was the best of

Locally constructed graph reveals the global sequence
structure
Overlaps implicitly computed

(de Bruijn, 1946; Idury and Waterman, 1995; Pevzner, Tang, Waterman, 2001)

de Bruijn Graph Assembly
It was the best

was the best of
it was the worst
the best of times,
was the worst of
best of times, it
the worst of times,
of times, it was

worst of times, it
times, it was the

it was the age the age of foolishness

was the age of
the age of wisdom,

age of wisdom, it

of wisdom, it was

wisdom, it was the

83

Compressed de Bruijn Graph
It was the best of times, it

it was the worst of times, it

of times, it was the
the age of foolishness

it was the age of
the age of wisdom, it was the

Unambiguous non-branching paths replaced by single nodes
An Eulerian traversal of the graph spells a compatible reconstruction
of the original text
There may be many traversals of the graph
Different sequences can have the same string graph
It was the best of times, it was the worst of times, it was the worst of times,
it was the age of wisdom, it was the age of foolishness, …

Questions?

84

Graph algorithms

Concluding Thoughts

When is MapReduce appropriate?
Lots of input data
(e.g., compute statistics over large amounts of text)
Take advantage of distributed storage, data locality, aggregate
disk throughput
Lots of intermediate data
(e.g., postings)
Take advantage of sorting/shuffling, fault tolerance
Lots of output data
(e.g., web crawls)
( b l )
Avoid contention for shared resources
Relatively little synchronization is necessary

85

When is MapReduce less appropriate?
Data fits in memory
Large amounts of shared data is necessary
Fine-grained synchronization is needed
Individual operations are processor-intensive

Alternatives to Hadoop

Pthreads Open MPI Hadoop
Programming model shared memory message-passing MapReduce
Job scheduling none with PBS limited
Synchronization fine only any coarse only
Distributed storage no no yes
Fault tolerance no no yes
Shared memory yes limited (MPI-2) no
Scale dozens of threads 10k+ of cores 10k+ cores

86

cheap commodity clusters (or utility computing)
+ simple distributed programming models
+ availability of large datasets
= data-intensive IR research for the masses!

What’s next?
Web-scale text processing: luxury → necessity
Don’t get dismissed as working on “toy problems”!
Fortunately, cluster computing is being commoditized
It’s all about the right level of abstractions:
MapReduce is only the beginning…

87

Applications
(NLP IR ML, etc.)
(NLP, IR, ML t )

Programming Models
(MapReduce…)

Systems
(architecture, network, etc.)
( hit t t k t )

Questions?
Comments?
Thanks to the organizations who support our work:

88

Topics: Afternoon Session
Running Hadoop in “standalone” mode
Running Hadoop in distributed mode
Running Hadoop on EC2
Hadoop ecosystem tour
Exercises and “office hours”
office hours

Source: Wikipedia “Japanese rock garden”

89

Hadoop Zen
Thinking at scale comes with a steep learning curve
Don’t get frustrated (take a deep breath)…
Remember this when you experience those W$*#T@F! moments
Hadoop is an immature platform…
Bugs, stability issues, even lost data
To upgrade or not to upgrade (damned either way)?
Poor documentation (read the fine code)
But… here lies the path to data nirvana
p

Cloud9
Set of libraries originally developed for teaching
MapReduce at the University of Maryland
Demos, exercises, etc.
“Eat you own dog food”
Actively used for a variety of research projects

90


Hadoop in “standalone” mode

91

Hadoop in distributed mode

Hadoop Cluster Architecture

Job submission node HDFS master

Client JobTracker NameNode

TaskTracker DataNode TaskTracker DataNode TaskTracker DataNode

Slave node Slave node Slave node

92

Hadoop Development Cycle
1. Scp data to cluster
2. Move data into HDFS

3. Develop code locally

4. Submit MapReduce job
4a. Go back to Step 3
Hadoop Cluster
You

5. Move data out of HDFS
6. Scp data from cluster

Hadoop on EC2

93

On Amazon: With EC2
0. Allocate Hadoop cluster
1. Scp data to cluster
2. Move data into HDFS

EC2
3. Develop code locally

4. Submit MapReduce job
4a. Go back to Step 3
Your Hadoop Cluster
You

5. Move data out of HDFS
6. Scp data from cluster
7. Clean up!

Uh oh. Where did the data go?

On Amazon: EC2 and S3

Copy from S3 to HDFS

EC2 S3
(Compute Facility) (Persistent Store)

Your Hadoop Cluster

Copy from HFDS to S3

94


What version should I use?

95

InputFormat

Slide from Cloudera basic training

Mapper Mapper Mapper Mapper

(intermediates) (intermediates) (intermediates) (intermediates)

Partitioner Partitioner Partitioner Partitioner
shuffling

(intermediates) (intermediates) (intermediates)

Reducer Reducer Reducer


96

OutputFormat


Data Types in Hadoop

Writable Defines a de/serialization protocol.
Every data type in Hadoop is a Writable.

WritableComprable Defines a sort order. All keys must be
of this type (but not values).

IntWritable Concrete classes for different data types.
LongWritable
Text
…

97

Complex Data Types in Hadoop
How do you implement complex data types?
The easiest way:
Encoded it as Text e.g., (a, b) = “a:b”
Text, e g (a a:b
Use regular expressions (or manipulate strings directly) to parse
and extract data
Works, but pretty hack-ish
The hard way:
Define a custom implementation of WritableComprable
Must i l
M t implement: readFields, write, compareTo
t dFi ld it T
Computationally efficient, but slow for rapid prototyping
Alternatives:
Cloud9 offers two other choices: Tuple and JSON

Hadoop Ecosystem Tour

98

Hadoop Ecosystem
Vibrant open-source community growing around Hadoop
Can I do foo with hadoop?
Most likely someone s already thought of it
likely, someone’s
… and started an open-source project around it
Beware of toys!

Starting Points…
Hadoop streaming
HDFS/FUSE
EC2/S3/EMR/EBS
C /S / / S

99

Pig and Hive
Pig: high-level scripting language on top of Hadoop
Open source; developed by Yahoo
Pig “compiles down” to MapReduce jobs
Hive: a data warehousing application for Hadoop
Open source; developed by Facebook
Provides SQL-like interface for querying petabyte-scale datasets

It’s all about data flows!

M R
MapReduce
p

What if you need…

M M R M

Join, Union Split Chains

… and filter, projection, aggregates, sorting,
distinct, etc.

Pig Slides adapted from Olston et al. (SIGMOD 2008)

100

Source: Wikipedia

Example: Find the top 10 most visited pages in each category

Visits Url Info

User Url Time Url Category PageRank

Amy cnn.com 8:00 cnn.com News 0.9

Amy bbc.com 10:00 bbc.com News 0.8

Amy flickr.com
f 10:05 flickr.com
f Photos 0.7

Fred cnn.com 12:00 espn.com Sports 0.9


101

Load Visits

Group by url

Foreach url
Load Url Info
generate count

Join on url

Group by category
Group by category

Foreach category
generate top10(urls)


visits = load ‘/data/visits’ as (user, url, time);
gVisits = group visits by url;
visitCounts = foreach gVisits generate url, count(visits);

urlInfo = load ‘/data/urlInfo’ as (url, category, pRank);
visitCounts = join visitCounts by url, urlInfo by url;

gCategories = group visitCounts by category;
topUrls = foreach gCategories generate top(visitCounts,10);

store topUrls into ‘/data/topUrls’;


102

Map1
Load Visits

Group by url
Reduce1
Map2
Foreach url
Load Url Info
generate count

Join on url
Reduce2
Map3
Group by category
Group by category

Foreach category Reduce3
generate top10(urls)


Other Systems
Zookeeper
HBase
Mahout
Hamma
Cassandra
Dryad
…

103

Questions?
Comments?
Thanks to the organizations who support our work:

104

Data-Intensive Text Processing with MapReduce

More Related Content

What's hot (9)

Viewers also liked (20)

Similar to Data-Intensive Text Processing with MapReduce (20)

More from George Ang (20)

Recently uploaded (20)

Data-Intensive Text Processing with MapReduce