MapReduce

Simplified Data Processing on Large Clusters

by Jeffrey Dean and Sanjay Ghemawa

Presented by Jon Logan
Outline

Problem Statement / Motivation

An Example Program
MapReduce vs Hadoop
GFS / HDFS
MapReduce Fundamentals
Example Code
Workflows
Conclusion / Questions
Why MapReduce?
Before MapReduce
Large Concurrent Systems
Grid Computing
Rolling Your Own Solution
Considerations
Threading is hard!
How do you scale to more machines?
How do you handle machine failures?
How do you facilitate communication between nodes?
Does your solution scale?

Scale out, not up!

An Example Program

I will present the concepts of MapReduce using the typical example of MR,
Word Count
The input of this program is a volume of raw text, of unspecified size (could
be KB, MB, TB, it doesnt matter!)
The output is a list of words, and their occurrence count. Assume that words
are split correctly, ignoring capitalization and punctuation.
Example
The doctor went to the store. =>
The, 2
Doctor, 1
Went, 1
To, 1
Store, 1
Map? Reduce?

Mappers read in data from the filesystem, and output (typically) modified data
Reducers collect all of the mappers output on the keys, and output (typically)
reduced data
The outputted data is written to disk

All data is in terms of key value pairs

Outline

Problem Statement / Motivation

An Example Program
MapReduce vs Hadoop
GFS / HDFS
MapReduce Fundamentals
Example Code
Workflows
Conclusion / Questions
MapReduce vs Hadoop

The paper is written by two researchers at Google, and describes their

programming paradigm
Unless you work at Google, or use Google App Engine, you wont use it!
Open Source implementation is Hadoop MapReduce
Not developed by Google
Started by Yahoo

Googles implementation (at least the one described) is written in C++

Hadoop is written in Java
GFS/HDFS

This is not a GFS/HDFS presentation! (But the following presentation is)

A few concepts are key to MapReduce though:

Google File System (GFS) and Hadoop Distributed File System (HDFS) are essentially
distributed filesystems
Are fault tolerant through replication
Allows data to be local to computation
Outline

Problem Statement / Motivation

An Example Program
MapReduce vs Hadoop
GFS / HDFS
MapReduce Fundamentals
Example Code
Workflows
Conclusion / Questions
 Major Components

User Components:
Mapper
Reducer
Combiner (Optional)
Partitioner (Optional) (Shuffle)
Writable(s) (Optional)

System Components:
Master
Input Splitter*
Output Committer*
* You can use your own if you really want!

Image source: http://www.ibm.com/developerworks/java/library/l-hadoop-3/index.html

Key Notes

Mappers and Reducers are typically single threaded and deterministic

Determinism allows for restarting of failed jobs, or speculative execution
Need to handle more data? Just add more Mappers/Reducers!
No need to handle multithreaded code
Since theyre all independent of each other, you can run (almost) arbitrary number of nodes
Mappers/Reducers run on arbitrary machines. A machine typically multiple map and
reduce slots available to it, typically one per processor core
Mappers/Reducers run entirely independent of each other
In Hadoop, they run in separate JVMs
Basic Concepts

All data is represented in key value pairs of an arbitrary type

Data is read in from a file or list of files, from HDFS
Data is chunked based on an input split
A typical chunk is 64MB (more or less can be configured depending on your use case)

Mappers read in a chunk of data

Mappers emit (write out) a set of data, typically derived from its input
Intermediate data (the output of the mappers) is split to a number of reducers
Reducers receive each key of data, along with ALL of the values associated with it
(this means each key must always be sent to the same reducer)
Essentially, <key, set<value>>
Reducers emit a set of data, typically reduced from its input which is written to disk
 Data Flow

Split
Mapper 0
0
Out
Reducer 0
0
Split
Input

Mapper 1
1
Out
Reducer 1
1
Split
Mapper 2
2
Input Splitter

Is responsible for splitting your input into multiple chunks

These chunks are then used as input for your mappers
Splits on logical boundaries. The default is 64MB per chunk
Depending on what youre doing, 64MB might be a LOT of data! You can change it
Typically, you can just use one of the built in splitters, unless you are reading
in a specially formatted file
Mapper

Reads in input pair <K,V> (a section as split by the input splitter)

Outputs a pair <K, V>

Ex. For our Word Count example, with the following input: 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>
Reducer

Accepts the Mapper output, and collects values on the key

All inputs with the same key must go to the same reducer!
Input is typically sorted, output is output exactly as is
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>
Combiner

Essentially an intermediate reducer

Is optional
Reduces output from each mapper, reducing bandwidth and sorting
Cannot change the type of its input
Input types must be the same as output types
Output Committer

Is responsible for taking the reduce output, and committing it to a file

Typically, this committer needs a corresponding input splitter (so that another
job can read the input)
Again, usually built in splitters are good enough, unless you need to output a
special kind of file
Partitioner (Shuffler)

Decides which pairs are sent to which reducer

Default is simply:
Key.hashCode() % numOfReducers

User can override to:

Provide (more) uniform distribution of load between reducers
Some values might need to be sent to the same reducer
Ex. To compute the relative frequency of a pair of words <W1, W2> you would need to
make sure all of word W1 are sent to the same reducer

Binning of results
Master

Responsible for scheduling & managing jobs

Scheduled computation should be close to the data if possible

Bandwidth is expensive! (and slow)
This relies on a Distributed File System (GFS / HDFS)!

If a task fails to report progress (such as reading input, writing output, etc),
crashes, the machine goes down, etc, it is assumed to be stuck, and is killed,
and the step is re-launched (with the same input)

The Master is handled by the framework, no user code is necessary

Master Cont.

HDFS can replicate data to be local if necessary for scheduling

Because our nodes are (or at least should be) deterministic
The Master can restart failed nodes
Nodes should have no side effects!

If a node is the last step, and is completing slowly, the master can launch a second
copy of that node
This can be due to hardware isuses, network issues, etc.
First one to complete wins, then any other runs are killed
Writables

Are types that can be serialized / deserialized to a stream

Are required to be input/output classes, as the framework will serialize your
data before writing it to disk
User can implement this interface, and use their own types for their
input/output/intermediate values
There are default for basic values, like Strings, Integers, Longs, etc.
Can also handle store, such as arrays, maps, etc.
Your application needs at least six writables
2 for your input
2 for your intermediate values (Map <-> Reduce)
2 for your output
Outline

Problem Statement / Motivation

An Example Program
MapReduce vs Hadoop
GFS / HDFS
MapReduce Fundamentals
Example Code
Workflows
Conclusion / Questions
Mapper Code
Our input to our mapper is <LongWritable, Text>
The key (the LongWritable) can be assumed to be the position in the document our
input is in. This doesnt matter for this example.

Our output is a bunch of <Text, LongWritable>. The key is the token, and the value
is the count. This is always 1.

For the purpose of this demonstration, just assume Text is a fancy String, and
LongWritable is a fancy Long. In reality, theyre just the Writable equivalents.
Reducer Code

Our input is the output of our Mapper, a <Text, LongWritable> pair

Our output is still a <Text,LongWritable>, but it reduces N inputs for token T,
into one output <T, N>
Combiner Code

Do we need a combiner?
No, but it reduces bandwidth.

Our reducer can actually be our combiner in this case though!

Thats it!

All that is needed to run the above code is an extremely simple runner class.
Simply specifies which components to use, and your input/output directories
Workflows

Sometimes you need multiple steps to express your design

MapReduce does not directly allow for this, but there are solutions that do
Hadoop YARN allows for a Directed Acyclic Graph of nodes
Oozie also allows for a graph of nodes
Handling Data By Type

Process
Data A

Input Fetch
Merge Output
Data

Process
Data B
Conclusion

MapReduce provides a simple way to scale your application

Scales out to more machines, rather than scaling up
Effortlessly scale from a single machine to thousands
Fault tolerant & High performance
If you can fit your use case to its paradigm, scaling is handled by the
framework