MapReduce

http://www.google.org/flutrends/ca/

(2012)
Average Searches Per Day:
5,134,000,000

2
 Motivation
• Process lots of data
• Google processed about 24 petabytes of data per day in
2009.

• A single machine cannot serve all the data

• You need a distributed system to store and process in
parallel
• Parallel programming?
• Threading is hard!
• How do you facilitate communication between nodes?
• How do you scale to more machines?
• How do you handle machine failures?
3
 MapReduce
• MapReduce [OSDI’04] provides
– Automatic parallelization, distribution
– I/O scheduling
Apache Hadoop:
• Load balancing Open source
• Network and data transfer optimization implementation of
MapReduce
– Fault tolerance
• Handling of machine failures

• Need more power: Scale out, not up!

• Large number of commodity servers as opposed to some high
end specialized servers

4
Typical problem solved by MapReduce
• Read a lot of data

• Map: extract something you care about from each

record

• Shuffle and Sort

• Reduce: aggregate, summarize, filter, or transform

• Write the results

5
 MapReduce workflow

Input Data Output Data

Wor Outpu
ker Wor write t
Split local ker File 0
0
Split read Wor write
1
Split ker Outpu
Wor t
2
Wor ker File 1
remote
ker
read,
sort
Map Reduce
extract something you aggregate,
care about from each summarize, filter,
record or transform
6
 Mappers and Reducers
• Need to handle more data? Just add more
Mappers/Reducers!

• No need to handle multithreaded code ☺

– Mappers and Reducers are typically single threaded
and deterministic
• Determinism allows for restarting of failed jobs
– Mappers/Reducers run entirely independent of each other
• In Hadoop, they run in separate JVMs

7
 Example: Word Count

http://kickstarthadoop.blogspot.ca/2011/04/word-count-hadoop-map-reduce-example.html

8
 Mapper
• Reads in input pair <Key,Value>

• Outputs a pair <K’, V’>

– Let’s count number of each word in user queries (or Tweets/Blogs)
– The input to the mapper will be <queryID, QueryText>:
<Q1,“The teacher went to the store. The store was closed;
the store opens in the morning. The store opens at 9am.”
>
– The output would be:
<The, 1> <teacher, 1> <went, 1> <to, 1> <the, 1>
<store,1> <the, 1> <store, 1> <was, 1> <closed, 1> <the,
1> <store,1> <opens, 1> <in, 1> <the, 1> <morning, 1>
<the 1> <store, 1> <opens, 1> <at, 1> <9am, 1>

9
 Reducer
• Accepts the Mapper output, and aggregates
values on the key
– For our example, the reducer input would be:
<The, 1> <teacher, 1> <went, 1> <to, 1> <the, 1> <store, 1>
<the, 1> <store, 1> <was, 1> <closed, 1> <the, 1> <store, 1>
<opens,1> <in, 1> <the, 1> <morning, 1> <the 1> <store, 1>
<opens, 1> <at, 1> <9am, 1>
– The output would be:
<The, 6> <teacher, 1> <went, 1> <to, 1> <store, 3> <was, 1>
<closed, 1> <opens, 1> <morning, 1> <at, 1> <9am, 1>

10
 MapReduce
Hadoo
p
Progra
m
fork fork fork

Mas
assign ter assign
map reduce
Input Data Output Data
Wor Outpu
ker Wor write t
Split Transfer local ker File 0
0 read write
Split peta-sca Wor
1
Split ker Outpu
le data Wor t
2
through Wor ker File 1
remote
network ker
read,
sort
Map Reduce 11
 Google File System (GFS)
Hadoop Distributed File System (HDFS)
• Split data and store 3 replica on commodity
servers

12
 HDFS MapReduce
NameNod
e Where are the chunks
Location of the of input data?
chunks of input data
Mas
assign ter assign
map
Input Data reduce Output Data
Wor Outpu
Split
ker Wor write t
Split 0 local ker File 0
0
Split Wor write
1
Split ker
Split Outpu
1 Wor t
2
Wor ker File 1
remote
ker
Split
read,
2
Read from sort
local disk Map Reduce

13
 Locality Optimization
• Master scheduling policy:
– Asks GFS for locations of replicas of input file blocks
– Map tasks scheduled so GFS input block replica are on
same machine or same rack

• Effect: Thousands of machines read input at local

disk speed
– Eliminate network bottleneck!

14
 Failure in MapReduce
• Failures are norm in commodity hardware

• Worker failure
– Detect failure via periodic heartbeats
– Re-execute in-progress map/reduce tasks

• Master failure
– Single point of failure; Resume from Execution Log

• Robust
– Google’s experience: lost 1600 of 1800 machines once!, but finished fine

15
 Fault tolerance:
Handled via re-execution
• On worker failure:
– Detect failure via periodic heartbeats
– Re-execute completed and in-progress map tasks
– Task completion committed through master

• Robust: [Google’s experience] lost 1600 of

1800 machines, but finished fine

16
 Refinement:
Redundant Execution
• Slow workers significantly lengthen completion
time
– Other jobs consuming resources on machine
– Bad disks with soft errors transfer data very slowly
– Weird things: processor caches disabled (!!)

• Solution: spawn backup copies of tasks

– Whichever one finishes first "wins"

17
 Refinement:
Skipping Bad Records
Map/Reduce functions sometimes fail for particular
inputs

• Best solution is to debug & fix, but not always

possible

• If master sees two failures for the same record:

– Next worker is told to skip the record

18
A MapReduce Job
Mapper

Reducer

Run this program as

a MapReduce job
19
 Mapper

Reducer

Run this program as

a MapReduce job 20
 Summary
• MapReduce
– Programming paradigm for data-intensive computing
– Distributed & parallel execution model
– Simple to program
• The framework automates many tedious tasks (machine
selection, failure handling, etc.)

21
 Running Program in MR
• Apache sources under /opt, the examples will be in
the following directory:

/opt/hadoop-2.6.0/share/hadoop/mapreduce/

• The exact location of the example jar file can be found

using the find command:

$ find / -name "hadoop-mapreduce-examples*.jar"

-print

22
 Running Program in MR
• A list of the available examples can be found
by running the following command.

$ yarn jar
$HADOOP_EXAMPLES/hadoop-mapreduce-exa
mples.jar

23
 Running Program in MR
• The pi example calculates the digits of p using
a quasi-Monte Carlo method.

$ yarn jar
$HADOOP_EXAMPLES/hadoop-mapreduce-exa
mples.jar pi 16 1000000

24
25
26
 Running the Terasort Test
• The terasort benchmark sorts a specified amount of randomly
generated data.

• This benchmark provides combined testing of the HDFS and

MapReduce layers of a

• Hadoop cluster. A full terasort benchmark run consists of the

following three steps:
– Generating the input data via teragen program.
– Running the actual terasort benchmark on the input data.
– Validating the sorted output data via the teravalidate program.

27
 Running the Terasort Test
• Run teragen to generate rows of random data to sort.
$ yarn jar $HADOOP_EXAMPLES/hadoop-mapreduce-examples.jar teragen 500000000
/user/hdfs/TeraGen-50GB
• Run terasort to sort the database.
$ yarn jar $HADOOP_EXAMPLES/hadoop-mapreduce-examples.jar terasort
➥/user/hdfs/TeraGen-50GB /user/hdfs/TeraSort-50GB
• Run teravalidate to validate the sort.
$ yarn jar $HADOOP_EXAMPLES/hadoop-mapreduce-examples.jar teravalidate
➥/user/hdfs/TeraSort-50GB /user/hdfs/TeraValid-50GB

28
 Running the Terasort Test
• The following command will instruct terasort to
use four reducer tasks:

• $ yarn jar
$HADOOP_EXAMPLES/hadoop-mapreduce-examp
les.jar terasort

➥ -Dmapred.reduce.tasks=4
/user/hdfs/TeraGen-50GB
/user/hdfs/TeraSort-50GB 29
 Additional information and background on
each of the examples and benchmarks
• Pi Benchmark
• https://hadoop.apache.org/docs/current/api/org/apache/h
adoop/examples/pi/package-summary.html
• Terasort Benchmark
• https://hadoop.apache.org/docs/current/api/org/apache/h
adoop/examples/terasort/package-summary.html
• Benchmarking and Stress Testing an Hadoop Cluster
• http://www.michael-noll.com/blog/2011/04/09/benchmark
ing-and-stresstesting-an-hadoop-cluster-with-terasort-testd
fsio-nnbench-mrbench (uses Hadoop V1, will work with V2)

30