Hadoop

Hadoop is a distributed computing platform.

Hadoop primarily consists of the Hadoop Distributed FileSystem (HDFS) and an implementation of the Map-Reduce programming paradigm.

Hadoop is a software framework that lets one easily write and run applications that process vast amounts of data. Here's what makes Hadoop especially useful:

Scalable: Hadoop can reliably store and process petabytes.
Economical: It distributes the data and processing across clusters of commonly available computers. These clusters can number into the thousands of nodes.
Efficient: By distributing the data, Hadoop can process it in parallel on the nodes where the data is located. This makes it extremely rapid.
Reliable: Hadoop automatically maintains multiple copies of data and automatically redeploys computing tasks based on failures.

Requirements

Platforms

Hadoop has been demonstrated on GNU/Linux clusters with more than 4000 nodes.
Windows is also a supported platform.

Requisite Software

Java 1.6.x, preferably from Sun. Set JAVA_HOME to the root of your Java installation.
ssh must be installed and sshd must be running to use Hadoop's scripts to manage remote Hadoop daemons.
rsync may be installed to use Hadoop's scripts to manage remote Hadoop installations.

Installing Required Software

If your platform does not have the required software listed above, you will have to install it.

For example on Ubuntu Linux:

$ sudo apt-get install ssh

$ sudo apt-get install rsync

Getting Started

First, you need to get a copy of the Hadoop code.

Edit the file conf/hadoop-env.sh to define at least JAVA_HOME.

Try the following command:

bin/hadoop

This will display the documentation for the Hadoop command script.

Standalone operation

By default, Hadoop is configured to run things in a non-distributed mode, as a single Java process. This is useful for debugging, and can be demonstrated as follows:

mkdir input cp conf/*.xml input bin/hadoop jar hadoop-*-examples.jar grep input output 'dfs[a-z.]+' cat output/*

This will display counts for each match of the regular expression.

Note that input is specified as a directory containing input files and that output is also specified as a directory where parts are written.

Distributed operation

To configure Hadoop for distributed operation you must specify the following:

The NameNode (Distributed Filesystem master) host. This is specified with the configuration property fs.default.name.
The org.apache.hadoop.mapred.JobTracker (MapReduce master) host and port. This is specified with the configuration property mapred.job.tracker.
A workers file that lists the names of all the hosts in the cluster. The default workers file is conf/workers.

Pseudo-distributed configuration

You can in fact run everything on a single host. To run things this way, put the following in:

conf/core-site.xml: <configuration> <property> <name>fs.default.name</name> <value>hdfs://localhost/</value> </property> </configuration> conf/hdfs-site.xml: <configuration> <property> <name>dfs.replication</name> <value>1</value> </property> </configuration> conf/mapred-site.xml: <configuration> <property> <name>mapred.job.tracker</name> <value>localhost:9001</value> </property> </configuration>

(We also set the HDFS replication level to 1 in order to reduce warnings when running on a single node.)

Now check that the command
ssh localhost
does not require a password. If it does, execute the following commands:

ssh-keygen -t dsa -P '' -f ~/.ssh/id_dsa cat ~/.ssh/id_dsa.pub >> ~/.ssh/authorized_keys

Bootstrapping

A new distributed filesystem must be formatted with the following command, run on the master node:

bin/hadoop namenode -format

The Hadoop daemons are started with the following command:

bin/start-all.sh

Daemon log output is written to the logs/ directory.

Input files are copied into the distributed filesystem as follows:

bin/hadoop fs -put input input

Distributed execution

Things are run as before, but output must be copied locally to examine it:

bin/hadoop jar hadoop-*-examples.jar grep input output 'dfs[a-z.]+' bin/hadoop fs -get output output cat output/*

When you're done, stop the daemons with:

bin/stop-all.sh

Fully-distributed operation

Fully distributed operation is just like the pseudo-distributed operation described above, except, specify:

The hostname or IP address of your master server in the value for fs.default.name, as hdfs://master.example.com/ in conf/core-site.xml.
The host and port of the your master server in the value of mapred.job.tracker as master.example.com:port in conf/mapred-site.xml.
Directories for dfs.name.dir and dfs.data.dir in conf/hdfs-site.xml.These are local directories used to hold distributed filesystem data on the master node and worker nodes respectively. Note that dfs.data.dir may contain a space- or comma-separated list of directory names, so that data may be stored on multiple local devices.
mapred.local.dir in conf/mapred-site.xml, the local directory where temporary MapReduce data is stored. It also may be a list of directories.
mapred.map.tasks and mapred.reduce.tasks in conf/mapred-site.xml. As a rule of thumb, use 10x the number of worker processors for mapred.map.tasks, and 2x the number of worker processors for mapred.reduce.tasks.

Finally, list all worker hostnames or IP addresses in your conf/workers file, one per line. Then format your filesystem and start your cluster on your master node, as above.