blob: be75058c948721477282567307e539f4e4bec006 [file] [log] [blame]
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
<head>
<title>Hadoop</title>
</head>
<body>
Hadoop is a distributed computing platform.
<p>Hadoop primarily consists of the <a
href="org/apache/hadoop/hdfs/package-summary.html">Hadoop Distributed FileSystem
(HDFS)</a> and an
implementation of the <a href="org/apache/hadoop/mapred/package-summary.html">
Map-Reduce</a> programming paradigm.</p>
<p>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:</p>
<ul>
<li>
<b>Scalable</b>: Hadoop can reliably store and process petabytes.
</li>
<li>
<b>Economical</b>: It distributes the data and processing across clusters
of commonly available computers. These clusters can number into the thousands
of nodes.
</li>
<li>
<b>Efficient</b>: By distributing the data, Hadoop can process it in parallel
on the nodes where the data is located. This makes it extremely rapid.
</li>
<li>
<b>Reliable</b>: Hadoop automatically maintains multiple copies of data and
automatically redeploys computing tasks based on failures.
</li>
</ul>
<h2>Requirements</h2>
<h3>Platforms</h3>
<ul>
<li>
Hadoop was been demonstrated on GNU/Linux clusters with 2000 nodes.
</li>
<li>
Win32 is supported as a <i>development</i> platform. Distributed operation
has not been well tested on Win32, so this is not a <i>production</i>
platform.
</li>
</ul>
<h3>Requisite Software</h3>
<ol>
<li>
Java 1.6.x, preferably from
<a href="http://java.sun.com/javase/downloads/">Sun</a>.
Set <tt>JAVA_HOME</tt> to the root of your Java installation.
</li>
<li>
ssh must be installed and sshd must be running to use Hadoop's
scripts to manage remote Hadoop daemons.
</li>
<li>
rsync may be installed to use Hadoop's scripts to manage remote
Hadoop installations.
</li>
</ol>
<h4>Additional requirements for Windows</h4>
<ol>
<li>
<a href="http://www.cygwin.com/">Cygwin</a> - Required for shell support in
addition to the required software above.
</li>
</ol>
<h3>Installing Required Software</h3>
<p>If your platform does not have the required software listed above, you
will have to install it.</p>
<p>For example on Ubuntu Linux:</p>
<p><blockquote><pre>
$ sudo apt-get install ssh<br>
$ sudo apt-get install rsync<br>
</pre></blockquote></p>
<p>On Windows, if you did not install the required software when you
installed cygwin, start the cygwin installer and select the packages:</p>
<ul>
<li>openssh - the "Net" category</li>
<li>rsync - the "Net" category</li>
</ul>
<h2>Getting Started</h2>
<p>First, you need to get a copy of the Hadoop code.</p>
<p>Edit the file <tt>conf/hadoop-env.sh</tt> to define at least
<tt>JAVA_HOME</tt>.</p>
<p>Try the following command:</p>
<tt>bin/hadoop</tt>
<p>This will display the documentation for the Hadoop command script.</p>
<h2>Standalone operation</h2>
<p>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:</p>
<tt>
mkdir input<br>
cp conf/*.xml input<br>
bin/hadoop jar hadoop-examples-*.jar grep input output 'dfs[a-z.]+'<br>
cat output/*
</tt>
<p>This will display counts for each match of the <a
href="http://java.sun.com/j2se/1.4.2/docs/api/java/util/regex/Pattern.html">
regular expression.</a></p>
<p>Note that input is specified as a <em>directory</em> containing input
files and that output is also specified as a directory where parts are
written.</p>
<h2>Distributed operation</h2>
To configure Hadoop for distributed operation you must specify the
following:
<ol>
<li>The NameNode (Distributed Filesystem master) host. This is
specified with the configuration property <tt><a
href="../core-default.html#fs.default.name">fs.default.name</a></tt>.
</li>
<li>The {@link org.apache.hadoop.mapred.JobTracker} (MapReduce master)
host and port. This is specified with the configuration property
<tt><a
href="../mapred-default.html#mapred.job.tracker">mapred.job.tracker</a></tt>.
</li>
<li>A <em>slaves</em> file that lists the names of all the hosts in
the cluster. The default slaves file is <tt>conf/slaves</tt>.
</ol>
<h3>Pseudo-distributed configuration</h3>
You can in fact run everything on a single host. To run things this
way, put the following in:
<br/>
<br/>
conf/core-site.xml:
<xmp><configuration>
<property>
<name>fs.default.name</name>
<value>hdfs://localhost/</value>
</property>
</configuration></xmp>
conf/hdfs-site.xml:
<xmp><configuration>
<property>
<name>dfs.replication</name>
<value>1</value>
</property>
</configuration></xmp>
conf/mapred-site.xml:
<xmp><configuration>
<property>
<name>mapred.job.tracker</name>
<value>localhost:9001</value>
</property>
</configuration></xmp>
<p>(We also set the HDFS replication level to 1 in order to
reduce warnings when running on a single node.)</p>
<p>Now check that the command <br><tt>ssh localhost</tt><br> does not
require a password. If it does, execute the following commands:</p>
<p><tt>ssh-keygen -t dsa -P '' -f ~/.ssh/id_dsa<br>
cat ~/.ssh/id_dsa.pub >> ~/.ssh/authorized_keys
</tt></p>
<h3>Bootstrapping</h3>
<p>A new distributed filesystem must be formatted with the following
command, run on the master node:</p>
<p><tt>bin/hadoop namenode -format</tt></p>
<p>The Hadoop daemons are started with the following command:</p>
<p><tt>bin/start-all.sh</tt></p>
<p>Daemon log output is written to the <tt>logs/</tt> directory.</p>
<p>Input files are copied into the distributed filesystem as follows:</p>
<p><tt>bin/hadoop fs -put input input</tt></p>
<h3>Distributed execution</h3>
<p>Things are run as before, but output must be copied locally to
examine it:</p>
<tt>
bin/hadoop jar hadoop-examples-*.jar grep input output 'dfs[a-z.]+'<br>
bin/hadoop fs -get output output
cat output/*
</tt>
<p>When you're done, stop the daemons with:</p>
<p><tt>bin/stop-all.sh</tt></p>
<h3>Fully-distributed operation</h3>
<p>Fully distributed operation is just like the pseudo-distributed operation
described above, except, specify:</p>
<ol>
<li>The hostname or IP address of your master server in the value
for <tt><a
href="../core-default.html#fs.default.name">fs.default.name</a></tt>,
as <tt><em>hdfs://master.example.com/</em></tt> in <tt>conf/core-site.xml</tt>.</li>
<li>The host and port of the your master server in the value
of <tt><a href="../mapred-default.html#mapred.job.tracker">mapred.job.tracker</a></tt>
as <tt><em>master.example.com</em>:<em>port</em></tt> in <tt>conf/mapred-site.xml</tt>.</li>
<li>Directories for <tt><a
href="../hdfs-default.html#dfs.name.dir">dfs.name.dir</a></tt> and
<tt><a href="../hdfs-default.html#dfs.data.dir">dfs.data.dir</a>
in <tt>conf/hdfs-site.xml</tt>.
</tt>These are local directories used to hold distributed filesystem
data on the master node and slave nodes respectively. Note
that <tt>dfs.data.dir</tt> may contain a space- or comma-separated
list of directory names, so that data may be stored on multiple local
devices.</li>
<li><tt><a href="../mapred-default.html#mapred.local.dir">mapred.local.dir</a></tt>
in <tt>conf/mapred-site.xml</tt>, the local directory where temporary
MapReduce data is stored. It also may be a list of directories.</li>
<li><tt><a
href="../mapred-default.html#mapred.map.tasks">mapred.map.tasks</a></tt>
and <tt><a
href="../mapred-default.html#mapred.reduce.tasks">mapred.reduce.tasks</a></tt>
in <tt>conf/mapred-site.xml</tt>.
As a rule of thumb, use 10x the
number of slave processors for <tt>mapred.map.tasks</tt>, and 2x the
number of slave processors for <tt>mapred.reduce.tasks</tt>.</li>
</ol>
<p>Finally, list all slave hostnames or IP addresses in your
<tt>conf/slaves</tt> file, one per line. Then format your filesystem
and start your cluster on your master node, as above.
</body>
</html>