Tag Archives: Hadoop Cluster

How-to: Analyze Twitter Data with Apache Hadoop “http://blog.cloudera.com/blog/2012/09/analyzing-twitter-data-with-hadoop/”

Social media has gained immense popularity with marketing teams, and Twitter is an effective tool for a company to get people excited about its products. Twitter makes it easy to engage users and communicate directly with them, and in turn, users can provide word-of-mouth marketing for companies by discussing the products. Given limited resources, and knowing we may not be able to talk to everyone we want to target directly, marketing departments can be more efficient by being selective about whom we reach out to.

In this post, we’ll learn how we can use Apache FlumeApache HDFSApache Oozie, and Apache Hive to design an end-to-end data pipeline that will enable us to analyze Twitter data. This will be the first post in a series. The posts to follow to will describe, in more depth, how each component is involved and how the custom code operates. All the code and instructions necessary to reproduce this pipeline is available on the Cloudera Github.

Who is Influential?

To understand whom we should target, let’s take a step back and try to understand the mechanics of Twitter. A user – let’s call him Joe – follows a set of people, and has a set of followers. When Joe sends an update out, that update is seen by all of his followers. Joe can also retweet other users’ updates. A retweet is a repost of an update, much like you might forward an email. If Joe sees a tweet from Sue, and retweets it, all of Joe’s followers see Sue’s tweet, even if they don’t follow Sue. Through retweets, messages can get passed much further than just the followers of the person who sent the original tweet. Knowing that, we can try to engage users whose updates tend to generate lots of retweets. Since Twitter tracks retweet counts for all tweets, we can find the users we’re looking for by analyzing Twitter data.

Now we know the question we want to ask: Which Twitter users get the most retweets? Who is influential within our industry?

How Do We Answer These Questions?

SQL queries can be used to answer this question: We want to look at which users are responsible for the most retweets, in descending order of most retweeted. However, querying Twitter data in a traditional RDBMS is inconvenient, since the Twitter Streaming API outputs tweets in a JSON format which can be arbitrarily complex. In the Hadoop ecosystem, the Hive project provides a query interface which can be used to query data that resides in HDFS. The query language looks very similar to SQL, but allows us to easily model complex types, so we can easily query the type of data we have. Seems like a good place to start. So how do we get Twitter data into Hive? First, we need to get Twitter data into HDFS, and then we’ll be able to tell Hive where the data resides and how to read it.

The diagram above shows a high-level view of how some of the CDH (Cloudera’s Distribution Including Apache Hadoop) components can be pieced together to build the data pipeline we need to answer the questions we have. The rest of this post will describe how these components interact and the purposes they each serve.

Gathering Data with Apache Flume

The Twitter Streaming API will give us a constant stream of tweets coming from the service. One option would be to use a simple utility like curl to access the API and then periodically load the files. However, this would require us to write code to control where the data goes in HDFS, and if we have a secure cluster, we will have to integrate with security mechanisms. It will be much simpler to use components within CDH to automatically move the files from the API to HDFS, without our manual intervention.

Apache Flume is a data ingestion system that is configured by defining endpoints in a data flow called sources and sinks. In Flume, each individual piece of data (tweets, in our case) is called an event; sources produce events, and send the events through a channel, which connects the source to the sink. The sink then writes the events out to a predefined location. Flume supports some standard data sources, such as syslog or netcat. For this use case, we’ll need to design a custom source that accesses the Twitter Streaming API, and sends the tweets through a channel to a sink that writes to HDFS files. Additionally, we can use the custom source to filter the tweets on a set of search keywords to help identify relevant tweets, rather than a pure sample of the entire Twitter firehose. The custom Flume source code can be found here.

Partition Management with Oozie

Once we have the Twitter data loaded into HDFS, we can stage it for querying by creating an external table in Hive. Using an external table will allow us to query the table without moving the data from the location where it ends up in HDFS. To ensure scalability, as we add more and more data, we’ll need to also partition the table. A partitioned table allows us to prune the files that we read when querying, which results in better performance when dealing with large data sets. However, the Twitter API will continue to stream tweets and Flume will perpetually create new files. We can automate the periodic process of adding partitions to our table as the new data comes in.

Apache Oozie is a workflow coordination system that can be used to solve this problem. Oozie is an extremely flexible system for designing job workflows, which can be scheduled to run based on a set of criteria. We can configure the workflow to run an ALTER TABLE command that adds a partition containing the last hour’s worth of data into Hive, and we can instruct the workflow to occur every hour. This will ensure that we’re always looking at up-to-date data.

The configuration files for the Oozie workflow are located here.

Querying Complex Data with Hive

Before we can query the data, we need to ensure that the Hive table can properly interpret the JSON data. By default, Hive expects that input files use a delimited row format, but our Twitter data is in a JSON format, which will not work with the defaults. This is actually one of Hive’s biggest strengths. Hive allows us to flexibly define, and redefine, how the data is represented on disk. The schema is only really enforced when we read the data, and we can use the Hive SerDe interface to specify how to interpret what we’ve loaded.

SerDe stands for Serializer and Deserializer, which are interfaces that tell Hive how it should translate the data into something that Hive can process. In particular, the Deserializer interface is used when we read data off of disk, and converts the data into objects that Hive knows how to manipulate. We can write a custom SerDe that reads the JSON data in and translates the objects for Hive. Once that’s put into place, we can start querying. The JSON SerDe code can be found here. The SerDe will take a tweet in JSON form, like the following:

and translate the JSON entities into queryable columns:

which will result in:

We’ve now managed to put together an end-to-end system, which gathers data from the Twitter Streaming API, sends the tweets to files on HDFS through Flume, and uses Oozie to periodically load the files into Hive, where we can query the raw JSON data, through the use of a Hive SerDe.

Some Results

In my own testing, I let Flume collect data for about three days, filtering on a set of keywords:

hadoop, big data, analytics, bigdata, cloudera, data science, data scientist, business intelligence, mapreduce, data warehouse, data warehousing, mahout, hbase, nosql, newsql, businessintelligence, cloudcomputing

The collected data was about half a GB of JSON data, and here is an example of what a tweet looks like. The data has some structure, but certain fields may or may not exist. The retweeted_status field, for example, will only be present if the tweet was a retweet. Additionally, some of the fields may be arbitrarily complex. The hashtags field is an array of all the hashtags present in the tweets, but most RDBMS’s do not support arrays as a column type. This semi-structured quality of the data makes the data very difficult to query in a traditional RDBMS. Hive can handle this data much more gracefully.

The query below will find usernames, and the number of retweets they have generated across all the tweets that we have data for:

For the few days of data, I found that these were the most retweeted users for the industry:

From these results, we can see whose tweets are getting heard by the widest audience, and also determine whether these people are communicating on a regular basis or not. We can use this information to more carefully target our messaging in order to get them talking about our products, which, in turn, will get other people talking about our products.


In this post we’ve seen how we can take some of the components of CDH and combine them to create an end-to-end data management system. This same architecture could be used for a variety of applications designed to look at Twitter data, such as identifying spam accounts, or identifying clusters of keywords. Taking the system even further, the general architecture can be used across numerous applications. By plugging in different Flume sources and Hive SerDes, this application can be customized for many other applications, like analyzing web logs, to give an example. Grab the code, and give it a shot yourself.

CDH 5.3: Apache Sentry Integration with HDFS

Starting in CDH 5.3, Apache Sentry integration with HDFS saves admins a lot of work by centralizing access control permissions across components that utilize HDFS.

It’s been more than a year and a half since a couple of my colleagues here at Cloudera shipped the first version of Sentry (now Apache Sentry (incubating)). This project filled a huge security gap in the Apache Hadoop ecosystem by bringing truly secure and dependable fine grained authorization to the Hadoop ecosystem and provided out-of-the-box integration for Apache Hive. Since then the project has grown significantly–adding support for Impala and Search and the wonderful Hue App to name a few significant additions.

In order to provide a truly secure and centralized authorization mechanism, Sentry deployments have been historically set up so that all Hive’s data and metadata are accessible only by HiveServer2 and every other user is cut out. This has been a pain point for Sqoop users as Sqoop does not use the HiveServer2 interface. Hence users with a Sentry-secured Hive deployment were forced to split the import task into two steps: simple HDFS import followed by manually loading the data into Hive.

With the inclusion of HDFS ACLs and the integration of Sentry into the Hive metastore in CDH 5.1, users were able to improve this situation and get the direct Hive import working again. However, this approach required manual administrator intervention to configure HDFS ACLs according to the Sentry configuration and needed a manual refresh to keep both systems in sync.

One of the large features included in the recently released CDH 5.3 is Sentry integration with HDFS, which enables customers to easily share data between Hive, Impala and all the other Hadoop components that interact with HDFS (MapReduce, Spark, Pig, and Sqoop, and so on) while ensuring that user access permissions only need to be set once, and that they are uniformly enforced.

The rest of this post focuses on the example of using Sqoop together with this Sentry feature. Sqoop data can now be imported into Hive without any additional administrator intervention. By exposing Sentry policies—what tables from which a user can select and to what tables they can insert—directly in HDFS, Sqoop will re-use the same policies that have been configured via GRANT/REVOKE statements or the Hue Sentry App and will import data into Hive without any trouble.


In order for Sqoop to seamlessly import into a Sentry Secured Hive instance, the Hadoop administrator needs to follow a few configuration steps to enable all the necessary features. First, your cluster needs to be using the Sentry Service as backend for storing authorization metadata and not rely on the older policy files.

If you are already using Sentry Service and GRANT/REVOKE statements, you can directly jump to step 3).

  1. Make sure that you have Sentry service running on your cluster. You should see it in the service list:

Ekran Resmi 2015-10-12 01.58.03

    2.  And that Hive is configured to use this service as a backend for Sentry metadata:

Ekran Resmi 2015-10-12 01.58.14

     3.  Finally enable HDFS Integration with Sentry:

Ekran Resmi 2015-10-12 01.58.27

Example Sqoop Import

Let’s assume that we have user jarcec who needs to import data into a Hive database named default. User jarcec is part of a group that is also called jarcec – in real life the name of the group doesn’t have to be the same as the username and that is fine.

With an unsecured Hive installation, the Hadoop administrator would have to jump in and grant writing privilege to user jarcec for directory /user/hive/warehouse or one of its subdirectories. With Sentry and HDFS integration, the Hadoop administrator no longer needs to jump in. Instead Sqoop will reuse the same authorization policies that has been configured through Hive SQL or via the Sentry Hue Application. Let’s assume that user bc is jarcec‘s Manager and already has privileges to grant privileges in the default database.

    1. bc starts by invoking beeline and connecting to HiveServer2:




[bc@sqoopsentry-1 ~]$ beeline


1: jdbc:hive2://sqoopsentry-1.vpc.cloudera.co> !connect jdbc:hive2://sqoopsentry-1.vpc.cloudera.com:10000/default;principal=hive/sqoopsentry-1.vpc.cloudera.com@ENT.CLOUDERA.COM

    1. In case that user jarcec is not part of any role yet, we need to create a role for him:



1: jdbc:hive2://sqoopsentry-1.vpc.cloudera.co> CREATE ROLE jarcec_role;


No rows affected (0.769 seconds)

    1. And this new role jarcec_role needs to be granted to jarcec‘s group jarcec.



1: jdbc:hive2://sqoopsentry-1.vpc.cloudera.co> GRANT ROLE jarcec_role to GROUP jarcec;


No rows affected (0.651 seconds)

    1. And finally bc can grant access to database default (or any other) to the role jarcec_role;



1: jdbc:hive2://sqoopsentry-1.vpc.cloudera.co> GRANT ALL ON DATABASE default TO ROLE jarcec_role;


No rows affected (0.16 seconds)

By executing the steps above, user jarcec has been given privilege to do any action (insert or select) with all objects inside database default. That includes the ability to create new tables, insert data or simply querying existing tables. With those privileges user jarcec can run the following Sqoop command as he was used to:




[jarcec@sqoopsentry-1 ~]$ sqoop import –connect jdbc:mysql://mysql.ent.cloudera.com/sqoop –username sqoop –password sqoop –table text <strong>–hive-import</strong>

14/12/14 15:37:38 INFO sqoop.Sqoop: Running Sqoop version: 1.4.5-cdh5.3.0


14/12/14 15:38:58 INFO mapreduce.ImportJobBase: Transferred 249.7567 MB in 75.8448 seconds (3.293 MB/sec) 

14/12/14 15:38:58 INFO mapreduce.ImportJobBase: Retrieved 1000000 records.

14/12/14 15:38:58 INFO manager.SqlManager: Executing SQL statement: SELECT t.* FROM `text` AS t LIMIT 1

14/12/14 15:38:58 INFO hive.HiveImport: Loading uploaded data into Hive

14/12/14 15:39:09 INFO hive.HiveImport: 14/12/14 15:39:09 WARN conf.HiveConf: DEPRECATED: Configuration property hive.metastore.local no longer has any effect. Make sure to provide a valid value for hive.metastore.uris if you are connecting to a remote metastore.

14/12/14 15:39:09 INFO hive.HiveImport:

14/12/14 15:39:09 INFO hive.HiveImport: Logging initialized using configuration in jar:file:/opt/cloudera/parcels/CDH-5.3.0-1.cdh5.3.0.p0.26/jars/hive-common-0.13.1-cdh5.3.0.jar!/hive-log4j.properties 

14/12/14 15:39:12 INFO hive.HiveImport: OK

14/12/14 15:39:12 INFO hive.HiveImport: Time taken: 1.079 seconds

14/12/14 15:39:12 INFO hive.HiveImport: Loading data to table default.text

14/12/14 15:39:12 INFO hive.HiveImport: setfacl: Permission denied. user=jarcec is not the owner of inode=part-m-00000

14/12/14 15:39:12 INFO hive.HiveImport: setfacl: Permission denied. user=jarcec is not the owner of inode=part-m-00001

14/12/14 15:39:12 INFO hive.HiveImport: setfacl: Permission denied. user=jarcec is not the owner of inode=part-m-00002

14/12/14 15:39:13 INFO hive.HiveImport: setfacl: Permission denied. user=jarcec is not the owner of inode=part-m-00003

14/12/14 15:39:13 INFO hive.HiveImport: Table default.text stats: [numFiles=4, numRows=0, totalSize=261888896, rawDataSize=0]

14/12/14 15:39:13 INFO hive.HiveImport: OK

14/12/14 15:39:13 INFO hive.HiveImport: Time taken: 0.719 seconds

14/12/14 15:39:13 INFO hive.HiveImport: <strong>Hive import complete</strong>.

14/12/14 15:39:13 INFO hive.HiveImport: Export directory is not empty, keeping it.

And jarcec can easily confirm in beeline that data have been indeed imported into Hive:



0: jdbc:hive2://sqoopsentry-1.vpc.cloudera.co> show tables from default;


|  tab_name  |


| text       |


1 row selected (0.177 seconds)

0: jdbc:hive2://sqoopsentry-1.vpc.cloudera.co> select count(*) from text;


|   _c0    |


| 1000000  |


1 row selected (72.188 seconds)

If Hive is configured to inherit permissions, you might notice that Sqoop will print out several warnings similar to this one:


14/12/14 15:39:12 INFO hive.HiveImport: setfacl: Permission denied. user=jarcec is not the owner of inode=part-m-00000

As there is no need to inherit HDFS permissions when Sentry is enabled in HDFS, you can safely ignore such messages.


Managing a Hadoop Cluster


  1. Introduction
  2. Goals for this Module
  3. Outline
  4. Basic Setup
    1. Java Requirements
    2. Operating System
    3. Downloading and Installing Hadoop
  5. Important Directories
  6. Selecting Machines
  7. Cluster Configurations
    1. Small Clusters: 2-10 Nodes
    2. Medium Clusters: 10-40 Nodes
    3. Large Clusters: Multiple Racks
  8. Performance Monitoring
    1. Ganglia
    2. Nagios
  9. Additional Tips
  10. References & Resources

Basic Setup

This section discusses the general platform requirements for Hadoop.


Hadoop is a Java-based system. Recent versions of Hadoop require Sun Java 1.6.

Compiling Java programs to run on Hadoop can be done with any number of commonly-used Java compilers. Sun’s compiler is fine, as is ecj, the Eclipse Compiler for Java. A bug in gcj, the GNU Compiler for Java, causes incompatibility between generated classes and Hadoop; it should not be used.


As Hadoop is written in Java, it is mostly portable between different operating systems. Developers can and do run Hadoop under Windows. The various scripts used to manage Hadoop clusters are written in a UNIX shell scripting language that assumes sh– or bash-like behavior. Thus running Hadoop under Windows requires cygwin to be installed. The Hadoop documentation stresses that a Windows/cygwin installation is for development only. The vast majority of server deployments today are on Linux. (Other POSIX-style operating systems such as BSD may also work. Some Hadoop users have reported successfully running the system on Solaris.) The instructions on this page assume a command syntax and system design similar to Linux, but can be readily adapted to other systems.


Hadoop is available for download from the project homepage athttp://hadoop.apache.org/core/releases.html. Here you will find several versions of Hadoop available.

The versioning strategy used is major.minor.revision. Increments to the major version number represent large differences in operation or interface and possibly significant incompatible changes. At the time of this writing (September 2008), there have been no major upgrades; all Hadoop versions have their major version set to 0. The minor version represents a large set of feature improvements and enhancements. Hadoop instances with different minor versions may use different versions of the HDFS file formats and protocols, requiring a DFS upgrade to migrate from one to the next. Revisions are used to provide bug fixes. Within a minor version, the most recent revision contains the most stable patches.

Within the releases page, two or three versions of Hadoop will be readily available, corresponding to the highest revision number in the most recent two or three minor version increments. The stable version is the highest revision number in the second most recent minor version. Production clusters should use this version. The most recent minor version may include improved performance or new features, but may also introduce regressions that will be fixed in ensuing revisions.

At the time of this writing, 0.18.0 is the most recent version, with 0.17.2 being the “stable” release. These example instructions assume that version 0.18.0 is being used; the directions will not change significantly for any other version, except by substituting the new version number where appropriate.

To install Hadoop, first download and install prerequisite software. This includes Java 6 or higher. Distributed operation requires ssh and sshd. Windows users must install and configure cygwin as well. Then download a Hadoop version using a web browser, wget, or curl, and then unzip the package:

gunzip hadoop-0.18.0.tar.gz
tar vxf hadoop-0.18.0.tar

Within the hadoop-0.18.0/ directory which results, there will be several subdirectories. The most interesting of these are bin/, where scripts to run the cluster are located, and conf/ where the cluster’s configuration is stored.

Enter the conf/ directory and modify hadoop-env.sh. The JAVA_HOME variable must be set to the base directory of your Java installation. It is recommended that you install Java in the same location on all machines in the cluster, so this file can be replicated to each machine without modification.

The hadoop-site.xml file must also be modified to contain a number of configuration settings. The sections below address the settings which should be included here.

If you are interested in setting up a development installation, running Hadoop on a single machine, the Hadoop documentation includes getting started instructions which will configure Hadoop for standalone or “pseudo-distributed” operation.

Standalone installations run all of Hadoop and your application inside a single Java process. The distributed file system is not used; file are read from and written to the local file system. Such a setup can be helpful for debugging Hadoop applications.

Pseudo-distributed operation refers to the use of several separate processes representing the different daemons (NameNode, DataNode, JobTracker, TaskTracker) and a separate task process to perform a Hadoop job, but with all processes running on a single machine. A pseudo-distributed instance will have a functioning NameNode/DataNode managing a “DFS” of sorts. Files in HDFS are in a separate namespace from the local file system, and are stored as block objects in a Hadoop-managed directory. However, it is not truly distributed, as no processing or data storage is performed on remote notes. A pseudo-distributed instance can be extended into a fully distributed cluster by adding more machines to function as Task/DataNodes, but more configuration settings are usually required to deploy a Hadoop cluster for multiple users.

The rest of this document deals with configuring Hadoop clusters of multiple nodes, intended for use by one or more developers.

After the conf/hadoop-site.xml is configured according to one of the models in the getting started, the sections below, or your own settings, two more files must be written.

The conf/masters file contains the hostname of the SecondaryNameNode. This should be changed from “localhost” to the fully-qualified domain name of the node to run the SecondaryNameNode service. It does not need to contain the hostname of the JobTracker/NameNode machine; that service is instantiated on whichever node is used to run bin/start-all.sh, regardless of the masters file. The conf/slaves file should contain the hostname of every machine in the cluster which should start TaskTracker and DataNode daemons. One hostname should be written per line in each of these files, e.g.:


The master node does not usually also function as a slave node, except in installations across only 1 or 2 machines.

If the nodes on your cluster do not support passwordless ssh, you should configure this now:

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

This will enable passwordless ssh login to the local machine. (You can verify that this works by executingssh localhost.) The ~/.ssh/id_dsa.pub and authorized_keys files should be replicated on all machines in the cluster.

At this point, the configuration must be replicated across all nodes in the cluster. Small clusters may use rsync or copy the configuration directory to each node. Larger clusters should use a configuration management system such as bcfg2, smartfrog, or puppet. NFS should be avoided as much as is possible, as it is a scalability bottleneck. DataNodes should never share block storage or other high-bandwidth responsibilities over NFS, and should avoid sharing configuration information over NFS if possible.

Various directories should be created on each node. The NameNode requires the NameNode metadata directory:

$ mkdir -p /home/hadoop/dfs/name

And every node needs the Hadoop tmp directory and DataNode directory created. Rather than logging in to each node and performing the steps multiple times manually, the file bin/slaves.sh allows a command to be executed on all nodes in the slaves file. For example, we can create these directories by executing the following commands on the NameNode:

$ mkdir -p /tmp/hadoop  # make the NameNode's tmp dir
$ ${HADOOP_HOME}/bin/slaves.sh "mkdir -p /tmp/hadoop"
$ ${HADOOP_HOME}/bin/slaves.sh "mkdir -p /home/hadoop/dfs/data"

The environment variables $HADOOP_CONF_DIR and $HADOOP_SLAVES are used by the bin/slaves.shscript to find the slave machines list. The provided command is then executed over ssh. If you need particular ssh options, the contents of the $HADOOP_SSH_OPTS variable are passed to ssh as arguments.

We then format HDFS by executing the following command on the NameNode:

$ bin/hadoop namenode -format

And finally, start the cluster:

$ bin/start-all.sh

Now it is time to load in data and start processing it with Hadoop! Good luck!

The remainder of this document discusses various trade-offs in cluster configurations for different sizes, and reviews the settings which may be placed in the hadoop-site.xml file.

Important Directories

One of the basic tasks involved in setting up a Hadoop cluster is determining where the several various Hadoop-related directories will be located. Where they go is up to you; in some cases, the default locations are inadvisable and should be changed. This section identifies these directories.

Directory Description Default location Suggested location
HADOOP_LOG_DIR Output location for log files from daemons ${HADOOP_HOME}/logs /var/log/hadoop
hadoop.tmp.dir A base for other temporary directories /tmp/hadoop-${user.name} /tmp/hadoop
dfs.name.dir Where the NameNode metadata should be stored ${hadoop.tmp.dir}/dfs/name /home/hadoop/dfs/name
dfs.data.dir Where DataNodes store their blocks ${hadoop.tmp.dir}/dfs/data /home/hadoop/dfs/data
mapred.system.dir The in-HDFS path to shared MapReduce system files ${hadoop.tmp.dir}/mapred/system /hadoop/mapred/system

This table is not exhaustive; several other directories are listed in conf/hadoop-defaults.xml. The remaining directories, however, are initialized by default to reside under hadoop.tmp.dir, and are unlikely to be a concern.

It is critically important in a real cluster that dfs.name.dir and dfs.data.dir be moved out fromhadoop.tmp.dir. A real cluster should never consider these directories temporary, as they are where all persistent HDFS data resides. Production clusters should have two paths listed for dfs.name.dir which are on two different physical file systems, to ensure that cluster metadata is preserved in the event of hardware failure.

A multi-user configuration should also definitely adjust mapred.system.dir. Hadoop’s default installation is designed to work for standalone operation, which does not use HDFS. Thus it conflates HDFS and local file system paths. When enabling HDFS, however, MapReduce will store shared information about jobs in mapred.system.dir on the DFS. If this path includes the current username (as the defaulthadoop.tmp.dir does), this will prevent proper operation. The current username on the submitting node will be the username who actually submits the job, e.g., “alex.” All other nodes will have the current username set to the username used to launch Hadoop itself (e.g., “hadoop”). If these do not match, the TaskTrackers will be unable to find the job information and run the MapReduce job.

For this reason, it is also advisable to remove ${user.name} from the general hadoop.tmp.dir.

While most of the directories listed above (all the ones with names in “foo.bar.baz” form) can be relocated via the conf/hadoop-site.xml file, the HADOOP_LOG_DIR directory is specified in conf/hadoop-env.sh as an environment variable. Relocating this directory requires editing this script.

Selecting Machines

Before diving into the details of configuring nodes, we include a brief word on choosing hardware for a cluster. While the processing demands of different organizations will dictate a different machine configuration for optimum efficiency, there are are commonalities associated with most Hadoop-based tasks.

Hadoop is designed to take advantage of whatever hardware is available. Modest “beige box” PCs can be used to run small Hadoop setups for experimentation and debugging. Providing greater computational resources will, to a point, result in increased performance by your Hadoop cluster. Many existing Hadoop deployments include Xeon processors in the 1.8-2.0GHz range. Hadoop jobs written in Java can consume between 1 and 2 GB of RAM per core. If you use HadoopStreaming to write your jobs in a scripting language such as Python, more memory may be advisable. Due to the I/O-bound nature of Hadoop, adding higher-clocked CPUs may not be the most efficient use of resources, unless the intent is to run HadoopStreaming. Big data clusters, of course, can use as many large and fast hard drives as are available. However, too many disks in a single machine will result in many disks not being used in parallel. It is better to have three machines with 4 hard disks each than one machine with 12 drives. The former configuration will be able to write to more drives in parallel and will provide greater throughput. Finally, gigabit Ethernet connections between machines will greatly improve performance over a cluster connected via a slower network interface.

It should be noted that the lower limit on minimum requirements for running Hadoop is well below the specifications for modern desktop or server class machines. However, multiple pages on the Hadoop wiki suggest similar specifications to those posted here for high-performance cluster design. (See [1],  [2].)

Cluster Configurations

This section provides cluster configuration advice and specific settings for clusters of varying sizes. These sizes were picked to demonstrate basic categories of clusters; your own installation may be a hybrid of different aspects of these profiles. Here we suggest various properties which should be included in theconf/hadoop-site.xml file to most effectively use a cluster of a given size, as well as other system configuration elements. The next section describes how to finish the installation after implementing the configurations described here. You should read through each of these configurations in order, as configuration suggestions for larger deployments are based on the preceding ones.


Setting up a small cluster for development purposes is a very straightforward task. When using two nodes, one node will act as both NameNode/JobTracker and a DataNode/TaskTracker; the other node is only a DataNode/TaskTracker. Clusters of three or more machines typically use a dedicated NameNode/JobTracker, and all other nodes are workers.

A relatively minimalist configuration in conf/hadoop-site.xml will suffice for this installation:


Clusters closer to the 8-10 node range may want to set dfs.replication to 3. Values higher than 3 are usually not necessary. Individual files which are heavily utilized by a large number of nodes may have their particular replication factor manually adjusted upward independent of the cluster default.


This category is for clusters that occupy the majority of a single rack. Additional considerations for high availability and reliability come into play at this level.

The single point of failure in a Hadoop cluster is the NameNode. While the loss of any other machine (intermittently or permanently) does not result in data loss, NameNode loss results in cluster unavailability. The permanent loss of NameNode data would render the cluster’s HDFS inoperable.

Therefore, another step should be taken in this configuration to back up the NameNode metadata. One machine in the cluster should be designated as the NameNode’s backup. This machine does not run the normal Hadoop daemons (i.e., the DataNode and TaskTracker). Instead, it exposes a directory via NFS which is only mounted on the NameNode (e.g., /mnt/namenode-backup/). The cluster’s hadoop-site.xml file should then instruct the NameNode to write to this directory as well:


The NameNode will write its metadata to each directory in the comma-separated list of dfs.name.dir. If/mnt/namenode-backup is NFS-mounted from the backup machine, this will ensure that a redundant copy of HDFS metadata is available. The backup node should serve /mnt/namenode-backup from/home/hadoop/dfs/name on its own drive. This way, if the NameNode hardware completely dies, the backup machine can be brought up as the NameNode with no reconfiguration of the backup machine’s software. To switch the NameNode and backup nodes, the backup machine should have its IP address changed to the original NameNode’s IP address, and the server daemons should be started on that machine. The IP address must be changed to allow the DataNodes to recognize it as the “original” NameNode for HDFS. (Individual DataNodes will cache the DNS entry associated with the NameNode, so just changing the hostname is insufficient; the name reassignment must be performed at the IP address level.)

The backup machine still has Hadoop installed and configured on it in the same way as every other node in the cluster, but it is not listed in the slaves file, so normal daemons are not started there.

One function that the backup machine can be used for is to serve as the SecondaryNameNode. Note that this is not a failover NameNode process. The SecondaryNameNode process connects to the NameNode and takes periodic snapshots of its metadata (though not in real time). The NameNode metadata consists of a snapshot of the file system called the fsimage and a series of deltas to this snapshot called theeditlog. With these two files, the current state of the system can be determined exactly. The SecondaryNameNode merges the fsimage and editlog into a new fsimage file that is a more compact representation of the file system state. Because this process can be memory intensive, running it on the backup machine (instead of on the NameNode itself) can be advantageous.

To configure the SecondaryNameNode daemon to run on the backup machine instead of on the master machine, edit the conf/masters file so that it contains the name of the backup machine. Thebin/start-dfs.sh and bin/start-mapred.sh (and by extension, bin/start-all.sh) scripts will actually always start the master daemons (NameNode and JobTracker) on the local machine. The slavesfile is used for starting DataNodes and TaskTrackers. The masters file is used for starting the SecondaryNameNode. This filename is used despite the fact that the master node may not be listed in the file itself.

A cluster of this size may also require nodes to be periodically decommissioned. As noted in Module 2, several machines cannot be turned off simultaneously, or data loss may occur. Nodes must be decommissioned on a schedule that permits replication of blocks being decommissioned. To prepare for this eventuality in advance, an excludes file should be added to the conf/hadoop-site.xml:


This property should provide the full path to the excludes file (the actual location of the file is up to you). You should then create an empty file with this name:

$ touch /home/hadoop/excludes

While the dfs.hosts.exclude property allows the definition of a list of machines which are explicitly barred from connecting to the NameNode (and similarly, mapred.hosts.exclude for the JobTracker), a large cluster may want to explicitly manage a list of machines which are approved to connect to a given JobTracker or NameNode.

The dfs.hosts and mapred.hosts properties allow an administrator to supply a file containing an approved list of hostnames. If a machine is not in this list, it will be denied access to the cluster. This can be used to enforce policies regarding which teams of developers have access to which MapReduce sub-clusters. These are configured in exactly the same way as the excludes file.

Of course, at this scale and above, 3 replicas of each block are advisable; the hadoop-site.xml file should contain:


By default, HDFS does not preserve any free space on the DataNodes; the DataNode service will continue to accept blocks until all free space on the disk is exhausted, which may cause problems. The following setting will require each DataNode to reserve at least 1 GB of space on the drive free before it writes more blocks, which helps preserve system stability:


Another parameter to watch is the heap size associated with each task. Hadoop caps the heap of each task process at 200 MB, which is too small for most data processing tasks. This cap is set as a parameter passed to the child Java process. It is common to override this with a higher cap by specifying:


This will provide each child task with 512 MB of heap. It is not unreasonable in some cases to specify -Xmx1024m instead. In the interest of providing only what is actually required, it may be better to leave this set to 512 MB by default, and allowing applications to manually configure for a full GB of RAM/task themselves.

Using multiple drives per machine

While small clusters often have only one hard drive per machine, more high-performance configurations may include two or more disks per node. Slight configuration changes are required to make Hadoop take advantage of additional disks.

DataNodes can be configured to write blocks out to multiple disks via the dfs.data.dir property. It can take on a comma-separated list of directories. Each block is written to one of these directories. E.g., assuming that there are four disks, mounted on /d1/d2/d3, and /d4, the following (or something like it) should be in the configuration for each DataNode:


MapReduce performance can also be improved by distributing the temporary data generated by MapReduce tasks across multiple disks on each machine:


Finally, if there are multiple drives available in the NameNode, they can be used to provide additional redundant copies of the NameNode metadata in the event of the failure of one drive. Unlike the above two properties, where one drive out of many is selected to write a piece of data, the NameNode writes to eachcomma-separated path in dfs.name.dir. If too many drives are listed here it may adversely affect the performance of the NameNode, as the probability of blocking on one or more I/O operations increases with the number of devices involved, but it is imperative that the sole copy of the metadata does not reside on a single drive.


Configuring multiple racks of machines for Hadoop requires further advance planning. The possibility of rack failure now exists, and operational racks should be able to continue even if entire other racks are disabled. Naive setups may result in large cross-rack data transfers which adversely affect performance. Furthermore, in a large cluster, the amount of metadata under the care of the NameNode increases. This section proposes configuring several properties to help Hadoop operate at very large scale, but the numbers used in this section are just guidelines. There is no single magic number which works for all deployments, and individual tuning will be necessary. These will, however, provide a starting point and alert you to settings which will be important.

The NameNode is responsible for managing metadata associated with each block in the HDFS. As the amount of information in the rack scales into the 10’s or 100’s of TB, this can grow to be quite sizable. The NameNode machine needs to keep the blockmap in RAM to work efficiently. Therefore, at large scale, this machine will require more RAM than other machines in the cluster. The amount of metadata can also be dropped almost in half by doubling the block size:


This changes the block size from 64MB (the default) to 128MB, which decreases pressure on the NameNode’s memory. On the other hand, this potentially decreases the amount of parallelism that can be achieved, as the number of blocks per file decreases. This means fewer hosts may have sections of a file to offer to MapReduce tasks without contending for disk access. The larger the individual files involved (or the more files involved in the average MapReduce job), the less of an issue this is.

In the medium configuration, the NameNode wrote HDFS metadata through to another machine on the rack via NFS. It also used that same machine to checkpoint the NameNode metadata and compact it in the SecondaryNameNode process. Using this same setup will result in the cluster being dependent on a single rack’s continued operation. The NFS-mounted write-through backup should be placed in a different rack from the NameNode, to ensure that the metadata for the file system survives the failure of an individual rack. For the same reason, the SecondaryNameNode should be instantiated on a separate rack as well.

With multiple racks of servers, RPC timeouts may become more frequent. The NameNode takes a continual census of DataNodes and their health via heartbeat messages sent every few seconds. A similar timeout mechanism exists on the MapReduce side with the JobTracker. With many racks of machines, they may force one another to timeout because the master node is not handling them fast enough. The following options increase the number of threads on the master machine dedicated to handling RPC’s from slave nodes:


These settings were used in clusters of several hundred nodes. They should be scaled up accordingly with larger deployments.

The following settings provide additional starting points for optimization. These are based on the reported configurations of actual clusters from 250 to 2000 nodes.

Property Range Description
io.file.buffer.size 32768-131072 Read/write buffer size used in SequenceFiles (should be in multiples of the hardware page size)
io.sort.factor 50-200 Number of streams to merge concurrently when sorting files during shuffling
io.sort.mb 50-200 Amount of memory to use while sorting data
mapred.reduce.parallel.copies 20-50 Number of concurrent connections a reducer should use when fetching its input from mappers
tasktracker.http.threads 40-50 Number of threads each TaskTracker uses to provide intermediate map output to reducers
mapred.tasktracker.map.tasks.maximum 1/2 * (cores/node) to 2 * (cores/node) Number of map tasks to deploy on each machine.
mapred.tasktracker.reduce.tasks.maximum 1/2 * (cores/node) to 2 * (cores/node) Number of reduce tasks to deploy on each machine.

Rack awareness

In a multi-rack configuration, it is important to ensure that replicas of blocks are placed on multiple racks to minimize the possibility of data loss. Thus, a rack-aware placement policy should be used. The guidelines there suggest how to set up a basic rack awareness policy; due to the heterogeneity of network topologies, a definitive general-purpose solution cannot be provided here.

This tutorial targets Hadoop version 0.18.0. While most of the interfaces described will work on other, older versions of Hadoop, rack-awareness underwent a major overhaul in version 0.17. Thus, the following does not apply to version 0.16 and before.

One major consequence of the upgrade is that while rack-aware block replica placement has existed in Hadoop for some time, rack-aware task placement has only been added in version 0.17. If Hadoop MapReduce cannot place a task on the same node as the block of data which the task is scheduled to process, then it picks an arbitrary different node on which to schedule the task. Starting with 0.17.0, tasks will be placed (when possible) on the same rack as at least one replica of an input data block for a job, which should further minimize the amount of inter-rack data transfers required to perform a job.

Hadoop includes an interface called DNSToSwitchMapping which allows arbitrary Java code to be used to map servers onto a rack topology. The configuration key topology.node.switch.mapping.impl can be used to specify a class which meets this interface. More straightforward than writing a Java class for this purpose, however, is to use the default mapper, which executes a user-specified script (or other command) on each node of the cluster, which returns the rack id for that node. These rack ids are then aggregated and sent back to the NameNode.

Note that the rack mapping script used by this system is incompatible with the 0.16 method of usingdfs.network.script. Whereas dfs.network.script runs on each DataNode, a new script specified bytopology.script.file.name is run by the master node only. To set the rack mapping script, specify the key topology.script.file.name in conf/hadoop-site.xml.

Cluster contention

If you are configuring a large number of machines, it is likely that you have a large number of users who wish to submit jobs to execute on it. Hadoop’s job scheduling algorithm is based on a simple FIFO scheduler. Using this in a large deployment without external controls or policies agreed upon by all users can lead to lots of contention for the JobTracker, causing short jobs to be delayed by other long-running tasks and frustrating users.

An advanced technique to combat this problem is to configure a single HDFS cluster which spans all available machines, and configure several separate MapReduce clusters with their own JobTrackers and pools of TaskTrackers. All MapReduce clusters are configured to use the same DFS and the same NameNode; but separate groups of machines have a different machine acting as JobTracker (i.e., subclusters have different settings for mapred.job.tracker). Breaking machines up into several smaller clusters, each of which contains 20-40 TaskTrackers, provides users with lower contention for the system. Users may be assigned to different clusters by policy, or they can use the JobTracker status web pages (a web page exposed on port 50030 of each JobTracker) to determine which is underutilized.

Multiple strategies exist for this assignment process. It is considered best practice to stripe the TaskTrackers associated with each JobTracker across all racks. This maximizes the availability of each cluster (as they are all resistant to individual rack failure), and works with the HDFS replica placement policy to ensure that each MapReduce cluster can find rack-local replicas of all files used in any MapReduce jobs.

Performance Monitoring

Multiple tools exist to monitor large clusters for performance and troubleshooting. This section briefly highlights two such tools.


Ganglia is a performance monitoring framework for distributed systems. Ganglia provides a distributed service which collects metrics on individual machines and forwards them to an aggregator which can report back to an administrator on the global state of a cluster.

Ganglia is designed to be integrated into other applications to collect statistics about their operation. Hadoop includes a performance monitoring framework which can use Ganglia as its backend. Instructions are available on the Hadoop wiki as to how to enable Ganglia metrics in Hadoop. Instructions are also included below.

After installing and configuring Ganglia on your cluster, to direct Hadoop to output its metric reports to Ganglia, create a file named hadoop-metrics.properties in the $HADOOP_HOME/conf directory. The file should have the following contents:



This assumes that gmond is running on each machine in the cluster. Instructions on the Hadoop wiki note that (in the experience of the wiki article author) this may result in all nodes reporting their results as “localhost” instead of with their individual hostnames. If this problem affects your cluster, an alternate configuration is proposed, in which all Hadoop instances speak directly with gmetad:



Where @GMETAD@ is the hostname of the server on which the gmetad service is running. If deploying Ganglia and Hadoop on a very large number of machines, the impact of this configuration (vs. the standard Ganglia configuration where individual services talk to gmond on localhost) should be evaluated.


While Ganglia will monitor Hadoop-specific metrics, general information about the health of the cluster should be monitored with an additional tool.

Nagios is a machine and service monitoring system designed for large clusters. Nagios will provide useful diagnostic information for tuning your cluster, including network, disk, and CPU utilization across machines.

Additional Tips

The following are a few additional pieces of small advice:

  • Create a separate user named “hadoop” to run your instances; this will separate the Hadoop processes from any users on the system. Do not run Hadoop as root.
  • If Hadoop is installed in /home/hadoop/hadoop-0.18.0, link /home/hadoop/hadoop to/home/hadoop/hadoop-0.18.0. When upgrading to a newer version in the future, the link can be moved to make this process easier on other scripts that depend on the hadoop/bin directory.

References & Resources

Hadoop Getting Started – Single-node configuration instructions
Hadoop Cluster Setup – Official Hadoop configuration instructions
Michael Noll’s Hadoop configuration tutorials for single and multiple node configurations.

How to Set Up a Hadoop Cluster Using Oracle Solaris

About Hadoop and Oracle Solaris Zones

The Apache Hadoop software is a framework that allows for the distributed processing of large data sets across clusters of computers using simple programming models.

The following are benefits of using Oracle Solaris Zones for a Hadoop cluster:

  • Fast provision of new cluster members using the zone cloning feature
  • Very high network throughput between the zones for data node replication
  • Optimized disk I/O utilization for better I/O performance with ZFS built-in compression
  • Secure data at rest using ZFS encryption

To store data, Hadoop uses the Hadoop Distributed File System (HDFS), which provides high-throughput access to application data and is suitable for applications that have large data sets. For more information about Hadoop and HDFS see http://hadoop.apache.org/.

The Hadoop cluster building blocks are as follows:

  • NameNode: The centerpiece of HDFS, which stores file system metadata, directs the slave DataNode daemons to perform the low-level I/O tasks, and also runs the JobTracker process.
  • Secondary NameNode: Performs internal checks of the NameNode transaction log.
  • DataNodes: Nodes that store the data in the HDFS file system, which are also known as slaves and run the TaskTracker process.

In the example presented in this article, all the Hadoop cluster building blocks will be installed using the Oracle Solaris Zones, ZFS, and network virtualization technologies. Figure 1 shows the architecture:

Figure 1

Figure 1. Architecture

Download and Install Hadoop

  1. To get a Hadoop distribution, download a recent stable release from one of the Apache download mirrors.For this article, I used the “12 October, 2012 Release 1.0.4” release.
  2. On the global zone, create the /usr/local directory if it doesn’t exist; later, we will share this directory with the zones.
    root@global_zone:~#  mkdir -p /usr/local
  3. Download the Hadoop tarball and copy it into /usr/local:
    root@global_zone:~#  cp /tmp/hadoop-1.0.4.tar.gz  /usr/local
  4. Unpack the tarball:
    root@global_zone:~# cd /usr/local
    root@global_zone:~# gzip -dc /usr/local/hadoop-1.0.4.tar.gz  | tar -xvf - 
  5. Create the Hadoop group:
    root@global_zone:~# groupadd hadoop
  6. Add the Hadoop user and set the user’s Hadoop password:
    root@global_zone:~# useradd -g hadoop hadoop
    root@global_zone:~# passwd hadoop
  7. Create the Hadoop user’s home directory:
    root@global_zone:~# mkdir -p /export/home/hadoop
    root@global_zone:~# chown hadoop:hadoop /export/home/hadoop/
  8. Rename the location of the Hadoop binaries and give ownership to the Hadoop user:
    root@global_zone:~# mv /usr/local/hadoop-1.0.4 /usr/local/hadoop
    root@global_zone:~# chown -R hadoop:hadoop /usr/local/hadoop*
  9. Edit the Hadoop configuration files, which are shown in Table 1:Table 1. Hadoop Configuration Files
    Filename Description
    hadoop-env.sh Specifies environment variable settings used by Hadoop.
    core-site.xml Specifies parameters relevant to all Hadoop daemons and clients.
    hdfs-site.xml Specifies parameters used by the HDFS daemons and clients.
    mapred-site.xml Specifies parameters used by the MapReduce daemons and clients.
    masters Contains a list of machines that run the Secondary NameNode.
    slaves Contains a list of machine names that run the DataNode and TaskTracker pair of daemons.

    To learn more about how the Hadoop framework is controlled by these configuration files, see http://hadoop.apache.org/docs/r1.0.4/api/org/apache/hadoop/conf/Configuration.html.

    1. Run the following command to change to the conf directory:
    root@global_zone:~# cd /usr/local/hadoop/conf
  10. Edit the hadoop-env.sh file to uncomment and change the export lines so they look like the following:
    export JAVA_HOME=/usr/java
    export HADOOP_LOG_DIR=/var/log/hadoop
  11. Edit the masters file so that it looks like the following:
  12. Edit the slaves file so that it looks like the following:
  13. Edit the core-site.xml file so it looks like the following:
  14. Edit the hdfs-site.xml file so it looks like Listing 1:

    Listing 1. hdfs-site.xml File

  15. Edit the mapred-site.xml file so it looks like this:

Configure the Network Time Protocol

We should ensure that the system clock on the Hadoop zones is synchronized by using the Network Time Protocol (NTP). In this example, the global zone is configured as an NTP server.

Note: It is best to select an NTP server that can be a dedicated time synchronization source so that other services are not negatively affected if the machine is brought down for planned maintenance.

The following example shows how to configure an NTP server.

  1. Edit the NTP server configuration file, as shown in Listing 2:
    root@global_zone:~# grep -v ^# /etc/inet/ntp.conf
    server prefer
    broadcast ttl 4
    enable auth monitor
    driftfile /var/ntp/ntp.drift
    statsdir /var/ntp/ntpstats/
    filegen peerstats file peerstats type day enable
    filegen loopstats file loopstats type day enable
    filegen clockstats file clockstats type day enable
    keys /etc/inet/ntp.keys
    trustedkey 0
    requestkey 0
    controlkey 0
    root@global_zone:~# touch /var/ntp/ntp.drift

    Listing 2. NTP Server Configuration File

  2. Enable the NTP server service:
    root@global_zone:~# svcadm enable ntp
  3. Verify that the NTP server is online by using the following command:
    root@global_zone:~# svcs -a | grep ntp
    online         16:04:15 svc:/network/ntp:default

Create the Virtual Network Interfaces

Oracle Solaris Zones on the same system can benefit from very high network I/O throughput (up to four times faster) with very low latency compared to systems with, say, 1 Gb physical network connections. For a Hadoop cluster, this means that the DataNodes can replicate the HDFS blocks much faster.

For more information about network virtualization benchmarks, see “How to Control Your Application’s Network Bandwidth.”

Create a series of virtual network interfaces (VNICs) for the different zones:

root@global_zone:~# dladm create-vnic -l net0 name_node1
root@global_zone:~# dladm create-vnic -l net0 secondary_name1
root@global_zone:~# dladm create-vnic -l net0 data_node1
root@global_zone:~# dladm create-vnic -l net0 data_node2
root@global_zone:~# dladm create-vnic -l net0 data_node3

Create the NameNode Zones

  1. If you don’t already have a file system for the NameNode and Secondary NameNode zones, run the following command:
    root@global_zone:~# zfs create -o mountpoint=/zones rpool/zones
  2. Create the name-node zone, as shown in Listing 3:
    root@global_zone:~# zonecfg -z name-node
    Use 'create' to begin configuring a new zone.
    Zonecfg:name-node> create
    create: Using system default template 'SYSdefault'
    zonecfg:name-node> set autoboot=true
    zonecfg:name-node> set limitpriv="default,sys_time"
    zonecfg:name-node> set zonepath=/zones/name-node
    zonecfg:name-node> add fs
    zonecfg:name-node:fs> set dir=/usr/local/hadoop
    zonecfg:name-node:fs> set special=/usr/local/hadoop
    zonecfg:name-node:fs> set type=lofs
    zonecfg:name-node:fs> set options=[ro,nodevices]
    zonecfg:name-node:fs> end
    zonecfg:name-node> add net
    zonecfg:name-node:net> set physical=name_node1
    zonecfg:name-node:net> end
    zonecfg:name-node> verify
    zonecfg:name-node> exit

    Listing 3. Creating the name-node Zone

  3. Create the sec-name-node zone, as shown in Listing 4:
    root@global_zone:~# zonecfg -z sec-name-node
    Use 'create' to begin configuring a new zone.
    Zonecfg:sec-name-node> create
    create: Using system default template 'SYSdefault'
    zonecfg:sec-name-node> set autoboot=true
    zonecfg:sec-name-node> set limitpriv="default,sys_time"
    zonecfg:sec-name-node> set zonepath=/zones/sec-name-node
    zonecfg:sec-name-node> add fs
    zonecfg:sec-name-node:fs> set dir=/usr/local/hadoop
    zonecfg:sec-name-node:fs> set special=/usr/local/hadoop
    zonecfg:sec-name-node:fs> set type=lofs
    zonecfg:sec-name-node:fs> set options=[ro,nodevices]
    zonecfg:sec-name-node:fs> end
    zonecfg:sec-name-node>  add net
    zonecfg:sec-name-node:net> set physical=secondary_name1
    zonecfg:sec-name-node:net> end
    zonecfg:sec-name-node> verify
    zonecfg:sec-name-node> exit

    Listing 4. Creating the sec-name-node Zone

Set Up the DataNode Zones

In this step, we can leverage the integration between Oracle Solaris Zones virtualization technology and the ZFS file system that is built into Oracle Solaris.

Hadoop best practice is to use a separate hard disk for each DataNode. Therefore, every DataNode zone will have its own hard disk in order to provide better I/O distribution, as shown in Figure 2.

Figure 2

Figure 2. Separate Disk for Each DataNode

Table 2 shows a summary of the Hadoop zones configuration we will create:

Table 2. Zone Summary

Function Zone Name ZFS Mount Point VNIC Name IP Address
NameNode name-node /zones/name-node name_node1
Secondary NameNode sec-name-node /zones/sec-name-node secondary_name1
DataNode data-node1 /zones/data-node1 data_node1
DataNode data-node2 /zones/data-node2 data_node2
DataNode data-node3 /zones/data-node3 data_node3
  1. Get the hard disk names using the following command:
    root@global_zone:~# format
    Searching for disks...done
           0. c7t0d0 <LSI-MR9261-8i-2.50-135.97GB>
           1. c7t1d0 <LSI-MR9261-8i-2.50-135.97GB>
           2. c7t2d0 <LSI-MR9261-8i-2.50-135.97GB>
  2. Create a separate ZFS file system for each zone in order to provide better disk I/O performance:
    root@global_zone:~# zpool create -O compression=on data-node1-pool c7t0d0
    root@global_zone:~# zpool create -O compression=on data-node2-pool c7t1d0
    root@global_zone:~# zpool create -O compression=on data-node3-pool c7t2d0
    root@global_zone:~# zfs create -o mountpoint=/zones/data-node1 data-node1-pool/data-node1
    root@global_zone:~# zfs create -o mountpoint=/zones/data-node2 data-node2-pool/data-node2
    root@global_zone:~# zfs create -o mountpoint=/zones/data-node3 data-node3-pool/data-node3
  3. Change the directories’ permissions in order to avoid warring messages during zone creation:
    root@global_zone:~# chmod 700 /zones/data-node1
    root@global_zone:~# chmod 700 /zones/data-node2
    root@global_zone:~# chmod 700 /zones/data-node3
  4. Point the zonecfg zonepath property to the ZFS file systems that you created, as shown in Listing 5.
    root@global_zone:~# zonecfg -z data-node1
    Use 'create' to begin configuring a new zone.
    zonecfg:data-node1> create
    create: Using system default template 'SYSdefault'
    zonecfg:data-node1> set autoboot=true
    zonecfg:data-node1> set limitpriv="default,sys_time"
    zonecfg:data-node1> set zonepath=/zones/data-node1 
    zonecfg:data-node1> add fs
    zonecfg:data-node1:fs> set dir=/usr/local/hadoop
    zonecfg:data-node1:fs> set special=/usr/local/hadoop
    zonecfg:data-node1:fs> set type=lofs
    zonecfg:data-node1:fs> set options=[ro,nodevices]
    zonecfg:data-node1:fs> end
    zonecfg:data-node1> add net
    zonecfg:data-node1:net> set physical=data_node1
    zonecfg:data-node1:net> end
    zonecfg:data-node1> verify
    zonecfg:data-node1> commit
    zonecfg:data-node1> exit
    root@global_zone:~# zonecfg -z data-node2
    Use 'create' to begin configuring a new zone.
    zonecfg:data-node2> create
    create: Using system default template 'SYSdefault'
    zonecfg:data-node2> set autoboot=true
    zonecfg:data-node2> set limitpriv="default,sys_time"
    zonecfg:data-node2> set zonepath=/zones/data-node2  
    zonecfg:data-node2> add fs
    zonecfg:data-node2:fs> set dir=/usr/local/hadoop
    zonecfg:data-node2:fs> set special=/usr/local/hadoop
    zonecfg:data-node2:fs> set type=lofs
    zonecfg:data-node2:fs> set options=[ro,nodevices]
    zonecfg:data-node2:fs> end
    zonecfg:data-node2> add net
    zonecfg:data-node2:net> set physical=data_node2
    zonecfg:data-node2:net> end
    zonecfg:data-node2> verify
    zonecfg:data-node2> commit
    zonecfg:data-node2> exit
    root@global_zone:~# zonecfg -z data-node3
    Use 'create' to begin configuring a new zone.
    zonecfg:data-node3> create
    create: Using system default template 'SYSdefault'
    zonecfg:data-node3> set autoboot=true
    zonecfg:data-node3> set limitpriv="default,sys_time"
    zonecfg:data-node3> set zonepath=/zones/data-node3
    zonecfg:data-node3> add fs
    zonecfg:data-node3:fs> set dir=/usr/local/hadoop
    zonecfg:data-node3:fs> set special=/usr/local/hadoop
    zonecfg:data-node3:fs> set type=lofs
    zonecfg:data-node3:fs> set options=[ro,nodevices]
    zonecfg:data-node3:fs> end
    zonecfg:data-node3> add net
    zonecfg:data-node3:net> set physical=data_node3
    zonecfg:data-node3:net> end
    zonecfg:data-node3> verify
    zonecfg:data-node3> commit
    zonecfg:data-node3> exit

    Listing 5. Setting the zonecfg zonepath Property

  5. Now, install the name-node zone; later we will clone it in order to accelerate zone creation time.
    root@global_zone:~# zoneadm -z name-node install
    The following ZFS file system(s) have been created:
    Progress being logged to /var/log/zones/zoneadm.20130106T134835Z.name-node.install
           Image: Preparing at /zones/name-node/root.
  6. Boot the name-node zone and check the status of the zones we’ve created, as shown in Listing 6:
    root@global_zone:~# zoneadm -z name-node boot
    root@global_zone:~# zoneadm list -cv
      ID NAME             STATUS     PATH                          BRAND    IP
       0 global           running    /                             solaris shared
       1 name-node        running    /zones/name-node              solaris  excl
       - sec-name-node    configured /zones/sec-name-node          solaris  excl
       - data-node1       configured /zones/data-node1             solaris  excl
       - data-node2       configured /zones/data-node2             solaris  excl
       - data-node3       configured /zones/data-node3             solaris  excl
    root@global_zone:~# zlogin -C name-node

    Listing 6. Booting the name-node Zone

  7. Provide the zone host information by using the following configuration for the name-node zone:
    1. For the host name, use name-node.
    2. Ensure the network interface name_node1 has an IP address of
    3. Ensure the name service is based on your network configuration.In this article, we will use /etc/hosts for name resolution, so we won’t set up DNS for host name resolution.
  8. After finishing the zone setup, log in to the zone.
  9. Developing for Hadoop requires a Java programming environment. You can install Java Development Kit (JDK) 6 using the following command:
    root@name-node:~# pkg install  jdk-6
  10. Verify the Java installation:
    root@name-node:~# which java
    root@name-node:~# java -version
    java version "1.6.0_35"
    Java(TM) SE Runtime Environment (build 1.6.0_35-b10)
    Java HotSpot(TM) Server VM (build 20.10-b01, mixed mode)
  11. Create a Hadoop user inside the name-node zone:
    root@name-node:~# groupadd hadoop
    root@name-node:~# useradd -g hadoop hadoop
    root@name-node:~# passwd hadoop
    root@name-node:~# mkdir -p /export/home/hadoop
    root@name-node:~# chown hadoop:hadoop /export/home/hadoop
  12. Configure an NTP client, as shown in the following example:
    1. Install the NTP package:
      root@name-node:~# pkg install ntp
    2. Edit the NTP client configuration file:
      root@name-node:~# grep -v ^# /etc/inet/ntp.conf
      server global_zone  prefer
      slewalways yes
      disable pll
    3. Enable the NTP service:
      root@name-node:~# svcadm enable ntp
  13. Add the Hadoop cluster members’ host names and IP addresses to /etc/hosts:
    root@name-node:~# cat /etc/hosts
    ::1             localhost       localhost loghost name-node sec-name-node data-node1 data-node2 data-node3

    Note: If you are using the global zone as an NTP server, you must also add its host name and IP address to /etc/hosts.

Set Up SSH

  1. Set up SSH key-based authentication for the Hadoop user on the name_node zone in order to enable password-less login to the Secondary DataNode and the DataNodes:
    root@name-node  # su - hadoop
    hadoop@name-node $ ssh-keygen -t dsa -P "" -f ~/.ssh/id_dsa
    hadoop@name-node $ cat ~/.ssh/id_dsa.pub >> ~/.ssh/authorized_keys
  2. Update $HOME/.profile:
    hadoop@name-node $ cat $HOME/.profile
    # Set JAVA_HOME 
    export JAVA_HOME=/usr/java
    # Add Hadoop bin/ directory to PATH
    export PATH=$PATH:/usr/local/hadoop/bin
  3. Check that Hadoop runs by typing the following command:
    hadoop@name-node $ hadoop version
    Hadoop 1.0.4
    Subversion https://svn.apache.org/repos/asf/hadoop/common/branches/branch-1.0 -r 1393290
    Compiled by hortonfo on Wed Oct  3 05:13:58 UTC 2012
    From source with checksum fe2baea87c4c81a2c505767f3f9b71f4
  4. Create a directory for the Hadoop log files:
    root@name-node:~# mkdir /var/log/hadoop
    root@name-node:~# chown hadoop:hadoop /var/log/hadoop
  5. From the global zone, run the following command to create the sec-name-node zone as a clone of name-node:
    root@global_zone:~# zoneadm -z name-node shutdown 
    root@global_zone:~# zoneadm -z sec-name-node clone name-node
  6. Boot the sec-name-node zone:
    root@global_zone:~# zoneadm -z sec-name-node boot
    root@global_zone:~# zlogin -C sec-name-node
  7. As we experienced previously, the system configuration tool is launched, so do the final configuration for sec-name-node zone:
    1. For the host name, use sec-name-node.
    2. For the network interface, use secondary_name1.
    3. Use an IP address of
    4. Ensure the name service is set to none.
  8. Perform similar steps for data-node1, data-node2, and data-node3:
    1. Do the following for data-node1:
      root@global_zone:~# zoneadm -z data-node1 clone name-node
      root@global_zone:~# zoneadm -z data-node1 boot
      root@global_zone:~# zlogin -C data-node1 
      • For the host name, use data-node1.
      • For the network interface, use data_node1.
      • Use an IP address of
      • Ensure the name service is set to none.
    2. Do the following for data-node2:
      root@global_zone:~# zoneadm -z data-node2 clone name-node
      root@global_zone:~# zoneadm -z data-node2 boot
      root@global_zone:~# zlogin -C data-node2
      • For the host name, use data-node2.
      • For the network interface, use data_node2.
      • Use an IP address of
      • Ensure the name service is set to none.
    3. Do the following for data-node3:
      root@global_zone:~# zoneadm -z data-node3  clone name-node
      root@global_zone:~# zoneadm -z data-node3 boot
      root@global_zone:~# zlogin -C data-node3
      • For the host name, use data-node3.
      • For the network interface, use data_node3.
      • Use an IP address of
      • Ensure the name service is set to none.
  9. Boot the name_node zone:
    root@global_zone:~#  zoneadm -z  name-node  boot
  10. Verify that all the zones are up and running:
    root@global_zone:~# zoneadm list -cv
      ID NAME             STATUS     PATH                          BRAND    IP
       0 global           running    /                             solaris shared
      10 sec-name-node    running    /zones/sec-name-node          solaris  excl
      12 data-node1       running    /zones/data-node1             solaris  excl
      14 data-node2       running    /zones/data-node2             solaris  excl
      16 data-node3       running    /zones/data-node3             solaris  excl
      17 name-node        running    /zones/name-node              solaris  excl

Verify the SSH Setup

To verify that you SSH access without using a password for the Hadoop user, do the following.

  1. From the name_node, log in via SSH into the name-node (that is, to itself):
    hadoop@name-node $ ssh name-node 
    The authenticity of host 'name-node (' can't be established.
    RSA key fingerprint is 04:93:a9:e0:b7:8c:d7:8b:51:b8:42:d7:9f:e1:80:ca.
    Are you sure you want to continue connecting (yes/no)? yes
    Warning: Permanently added 'name-node,' (RSA) to the list of known hosts.
  2. Now, try to log in to the sec-name-node and the DataNodes. When you try to log in with SSH the second time, you shouldn’t get any prompt to add the host to the known keys list.

Verify Name Resolution

Verify that all the Hadoop zones have the following host entries in /etc/hosts:

# cat /etc/hosts

::1             localhost       localhost loghost name-node sec-name-node data-node1 data-node2 data-node3

Note: If you are using the global zone as an NTP server, you must also add its host name and IP address to /etc/hosts.

Format the HDFS File System from the NameNode

  1. Run the commands shown in Listing 7:
    root@name-node:~# mkdir -p /hdfs/name
    root@name-node:~# chown -R hadoop:hadoop /hdfs/
    root@name-node:~# su - hadoop
    hadoop@name-node:$ /usr/local/hadoop/bin/hadoop namenode -format
    13/01/06 20:00:32 INFO namenode.NameNode: STARTUP_MSG:
    STARTUP_MSG: Starting NameNode
    STARTUP_MSG:   host = name-node/
    STARTUP_MSG:   args = [-format]
    STARTUP_MSG:   version = 1.0.4
    STARTUP_MSG:   build = https://svn.apache.org/repos/asf/hadoop/common/branches/branch-1.0 -r 1393290; 
    compiled by 'hortonfo' on Wed Oct  3 05:13:58 UTC 2012

    Listing 7. Formatting the HDFS File System

  2. On every DataNode (data-node1, data-node2, and data-node3), create a Hadoop data directory to store the HDFS blocks:
    root@data-node1:~# mkdir -p /hdfs/data
    root@data-node1:~# chown -R hadoop:hadoop /hdfs/
    root@data-node2:~# mkdir -p /hdfs/data
    root@data-node2:~# chown -R hadoop:hadoop /hdfs/
    root@data-node3:~# mkdir -p /hdfs/data
    root@data-node3:~# chown -R hadoop:hadoop /hdfs/

Start the Hadoop Cluster

Table 3 describes the startup scripts.

Table 3. Startup Scripts

Filename Description
start-dfs.sh Starts the Hadoop DFS daemons, the NameNode, and the DataNodes. Use this before start-mapred.sh.
stop-dfs.sh Stops the Hadoop DFS daemons.
start-mapred.sh Starts the Hadoop MapReduce daemons, the JobTracker, and TaskTrackers.
stop-mapred.sh Stops the Hadoop MapReduce daemons.
  1. From the name-node zone, start the Hadoop DFS daemons, the NameNode, and the DataNodes using the following command:
    hadoop@name-node:$ start-dfs.sh
    starting namenode, logging to /var/log/hadoop/hadoop--namenode-name-node.out
    data-node2: starting datanode, logging to /var/log/hadoop/hadoop-hadoop-datanode-data-node2.out
    data-node1: starting datanode, logging to /var/log/hadoop/hadoop-hadoop-datanode-data-node1.out
    data-node3: starting datanode, logging to /var/log/hadoop/hadoop-hadoop-datanode-data-node3.out
    sec-name-node: starting secondarynamenode, logging to 
  2. Start the Hadoop Map/Reduce daemons, the JobTracker, and TaskTrackers using the following command:
    hadoop@name-node:$ start-mapred.sh
    starting jobtracker, logging to /var/log/hadoop/hadoop--jobtracker-name-node.out
    data-node1: starting tasktracker, logging to /var/log/hadoop/hadoop-hadoop-tasktracker-data-node1.out
    data-node3: starting tasktracker, logging to /var/log/hadoop/hadoop-hadoop-tasktracker-data-node3.out
    data-node2: starting tasktracker, logging to /var/log/hadoop/hadoop-hadoop-tasktracker-data-node2.out
  3. To view a comprehensive status report, execute the command shown in Listing 8 to check the cluster status. The command will output basic statistics about the cluster health, such as NameNode details, the status of each DataNode, and disk capacity amounts.
    hadoop@name-node:$ hadoop dfsadmin -report
    Configured Capacity: 171455269888 (159.68 GB)
    Present Capacity: 169711053357 (158.06 GB)
    DFS Remaining: 169711028736 (158.06 GB)
    DFS Used: 24621 (24.04 KB)
    DFS Used%: 0%
    Under replicated blocks: 0
    Blocks with corrupt replicas: 0
    Missing blocks: 0
    Datanodes available: 3 (3 total, 0 dead)

    Listing 8. Checking the Cluster Status

    You can find the same information on the NameNode Web status page at http://<namenode IP address>:50070/dfshealth.jsp.

    Figure 3

    Figure 3. Cluster Summary

Run a MapReduce Job

MapReduce is a framework for processing parallelizable problems across huge data sets using a cluster of computers. For more information about MapReduce, see http://en.wikipedia.org/wiki/MapReduce.

We will use the WordCount example, which reads text files and counts how often words occur. The input and output consist of text files, each line of which contains a word and the number of times the word occurred, separated by a tab. For more information about WordCount, see http://wiki.apache.org/hadoop/WordCount.

  1. For the input file, download the following eBook from Project Gutenberg as a plain-text file with UTF-8 encoding, and store the file in a temporary directory of choice, for example /tmp/data: The Outline of Science, Vol. 1 (of 4) by J. Arthur Thomson.
  2. Copy the file to HDFS using the following command:
    hadoop@name-node:$ hadoop dfs -copyFromLocal /tmp/data/ /hdfs/data
  3. Verify that the file is located on HDFS:
    hadoop@name-node:$ hadoop dfs -ls /hdfs/data
    Found 1 items
    -rw-r--r--   3 hadoop supergroup     661664 2013-01-07 19:45 /hdfs/data/20417-8.txt
  4. Start the MapReduce job using the command shown in Listing 9:
    hadoop@name-node:$ hadoop jar /usr/local/hadoop/hadoop-examples-1.0.4.jar wordcount /hdfs/data /hdfs/data/data-output
    13/01/07 15:20:21 INFO input.FileInputFormat: Total input paths to process : 1
    13/01/07 15:20:21 WARN util.NativeCodeLoader: Unable to load native-hadoop library 
    for your platform... using builtin-java classes where applicable
    13/01/07 15:20:21 WARN snappy.LoadSnappy: Snappy native library not loaded
    13/01/07 15:20:22 INFO mapred.JobClient: Running job: job_201301071307_0006
    13/01/07 15:20:23 INFO mapred.JobClient:  map 0% reduce 0%
    13/01/07 15:20:38 INFO mapred.JobClient:  map 100% reduce 0%
    13/01/07 15:20:50 INFO mapred.JobClient:  map 100% reduce 100%
    13/01/07 15:20:55 INFO mapred.JobClient: Job complete: job_201301071307_0006
    13/01/07 15:20:55 INFO mapred.JobClient: Counters: 26
    13/01/07 15:20:55 INFO mapred.JobClient:   Job Counters

    Listing 9. Starting the MapReduce Job

  5. Verify the output data:
    hadoop@name-node:$ hadoop dfs -ls /hdfs/data/data-output
    Found 3 items
    -rw-r--r--   3 hadoop supergroup          0 2013-01-07 15:20 /hdfs/data/data-output/_SUCCESS
    drwxr-xr-x   - hadoop supergroup          0 2013-01-07 15:20 /hdfs/data/data-output/_logs
    -rw-r--r--   3 hadoop supergroup     196288 2013-01-07 15:20 /hdfs/data/data-output/part-r-00000
  6. The output data can contain sensitive information, so use ZFS encryption to protect the output data:
    1. Create the encrypted ZFS data set:
      root@name-node:~# zfs create -o encryption=on rpool/export/output
      Enter passphrase for 'rpool/export/output':
      Enter again:
    2. Change the ownership:
      root@name-node:~# chown hadoop:hadoop /export/output/
    3. Copy the output file from the Hadoop HDFS into ZFS:
      root@name-node:~# su - hadoop
      Oracle Corporation      SunOS 5.11      11.1    September 2012
      hadoop@name-node:$ hadoop dfs -getmerge /hdfs/data/data-output /export/output/
  7. Analyze the output text file. Each line contains a word and the number of times the word occurred, separated by a tab.
    hadoop@name-node:$ head /export/output/data-output
    "A      2
    "Alpha  1
    "Alpha,"        1
    "An     2
    "And    1
    "BOILING"       2
    "Batesian"      1
    "Beta   2
  8. Protect the output text file by unmounting the ZFS data set, and then unload the wrapping key for an encrypted data set using the following command:
    root@name-node:~# zfs key -u rpool/export/output

    If the command is successful, the data set is not accessible and it is unmounted.

    If you want to mount this ZFS file system, you need to provide the passphrase:

    root@name-node:~# zfs mount rpool/export/output
    Enter passphrase for 'rpool/export/output':

    By using a passphrase, you ensure that only those who know the passphrase can observe the output file. For more information about ZFS encryption, see “How to Manage ZFS Data Encryption.”