Apache Hadoop

Apache Hadoop ( / h ə d u p / ) is an open source software framework used for distributed storage and processing of dataset of big data using the MapReduce programming model . It consists of computer clusters built from commodity hardware . All the modules in Hadoop are designed with a fundamental assumption that hardware failures are common occurrences and should be handled by the framework. [2]

The core of Apache Hadoop consists of a storage share, known as Hadoop Distributed File System (HDFS), and a processing part which is a MapReduce programming model. Hadoop splits files into large blocks and distributes them across nodes in a cluster. It then transfers packaged code to the data process. This approach takes advantage of data locality , [3]where they manipulate the data they have access to. This Allows the dataset to be processed faster and more Efficiently than it Would Be in a more conventional supercomputer architecture That was subsequently assembled parallel file system Where are distributed computation and data via high-speed networking. [4][5]

The Apache Hadoop framework is composed of the following modules:

  • Hadoop Common – contains libraries and utilities needed by other Hadoop modules;
  • Hadoop Distributed File System (HDFS) – a distributed file-system that stores data on commodity machines, providing very high aggregate bandwidth across the cluster;
  • Hadoop YARN – a platform responsible for managing computing resources in clusters and using them for scheduling users’ applications; [6] [7] and
  • Hadoop MapReduce – an implementation of the MapReduce programming model for large-scale data processing.

The term Hadoop has come to Refer not just to the aforementioned base modules and sub-modules, goal aussi to the ecosystem , [8] or collection of additional software packages That can be installed on top of gold Alongside Hadoop, Such As Apache Pig , Apache Hive , Apache HBase , Phoenix Apache , Apache Spark , Apache ZooKeeper , Cloudera Impala , Apache Flume , Apache Sqoop , Apache Oozie , and Apache Storm . [9]

Apache Hadoop ‘s MapReduce and HDFS components were inspired by Google papers on their MapReduce and Google File System . [10]

The Hadoop framework itself is mostly written in the Java programming language , with some native code in C and command line utilities written as shell scripts . MapReduce Java code is common, any programming language can be used with “Hadoop Streaming” to implement the “map” and “reduce” parts of the user’s program. [11] Other projects in the Hadoop ecosystem expose richer user interfaces.


According to its co-founders, Doug Cutting and Mike Cafarella , the genesis of Hadoop was the “Google File System” paper that was published in October 2003. [12] [13] This paper spawned another one from Google – “MapReduce: Simplified Data Processing on Large Clusters “. [14] Development started on the Apache Nutch project, but was moved to the new Hadoop subproject in January 2006. [15] Doug Cutting , who was working at Yahoo! at the time, named after his son’s toy elephant. [16] The initial code was made up of about 5,000 lines of code for HDFS and about 6,000 lines of code for MapReduce.

The first committer to the Hadoop project was Owen O’Malley (in March 2006); [17] Hadoop 0.1.0 was released in April 2006. [18] It continues to evolve through the many contributions that are made to the project. [19]


year month Event Ref.
2003 october Google File System paper released [20]
2004 december MapReduce : Simplified Data Processing on Large Clusters [21]
2006 january Hadoop subproject created with mailing lists, jira, and wiki [22]
2006 january Hadoop is born from Nutch 197 [23]
2006 february NDFS + MapReduce moved out of Apache Nutch to create Hadoop [24]
2006 february Owen O’Malley’s first patch goes into Hadoop [25]
2006 february Hadoop is named after Cutting son’s yellow plush toy [26]
2006 April Hadoop 0.1.0 released [27]
2006 April Hadoop spells 1.8 TB on 188 nodes in 47.9 hours [24]
2006 may Yahoo deploys 300 Hadoop cluster machine [24]
2006 october Yahoo Hadoop cluster reaches 600 machines [24]
2007 April Yahoo runs two clusters of 1,000 machines [24]
2007 june Only three companies on “Powered by Hadoop Page” [28]
2007 october First release of Hadoop that includes HBase [29]
2007 october Yahoo Labs creates Pig, and donates it to the ASF [30]
2008 january YARN JIRA opened Yarn Jira (Mapreduce 279)
2008 january 20 companies on “Powered by Hadoop Page” [28]
2008 february Yahoo moves its web index onto Hadoop [31]
2008 february Yahoo! production search index generated by a 10,000-core Hadoop cluster [24]
2008 march First Hadoop Summit [32]
2008 April Hadoop world record fastest system to a terabyte of data. Running on a 910-node cluster, Hadoop sorted on a terabyte in 209 seconds [24]
2008 may Hadoop Wins TeraByte Sort (World Record sortbenchmark.org) [33]
2008 july Hadoop wins Terabyte Sort Benchmark [34]
2008 october Loading 10 TB / day in Yahoo clusters [24]
2008 october Cloudera, Hadoop distributor is founded [35]
2008 november Google MapReduce implementation released in one terabyte in 68 seconds [24]
2009 march Yahoo runs 17 clusters with 24,000 machines [24]
2009 April Hadoop spells at petabyte [36]
2009 may Yahoo! used Hadoop to spell one terabyte in 62 seconds [24]
2009 june Second Hadoop Summit [37]
2009 july Hadoop Core is renamed Hadoop Common [38]
2009 july MapR, Hadoop Founded distributor [39]
2009 july HDFS now a separate subproject [38]
2009 july MapReduce now a separate subproject [38]
2010 january Kerberos support added to Hadoop [40]
2010 may Apache HBase Graduates [41]
2010 june Third Hadoop Summit [42]
2010 june Yahoo 4,000 nodes / 70 petabytes [43]
2010 june Facebook 2,300 clusters / 40 petabytes [43]
2010 september Apache Hive Graduates [44]
2010 september Apache Pig Graduates [45]
2011 january Apache Zookeeper Graduates [46]
2011 january Facebook, LinkedIn, eBay and IBM collectively contribute 200,000 lines of code [47]
2011 march Apache Hadoop takes top prize at Media Guardian Innovation Awards [48]
2011 june Rob Beardon and Eric Badleschieler spin out Hortonworks out of Yahoo. [49]
2011 june Yahoo has 42K Hadoop and hundreds of petabytes of storage [49]
2011 june Third Annual Hadoop Summit (1,700 expected) [50]
2011 october Debate over which company had contributed to Hadoop. [47]
2012 january Hadoop community moves to separate from MapReduce and replace with YARN [26]
2012 june San Jose Hadoop Summit (2,100 expected) [51]
2012 november Apache Hadoop 1.0 Available [38]
2013 march Hadoop Summit – Amsterdam (500 expected) [52]
2013 march YARN deployed in production at Yahoo [53]
2013 june San Jose Hadoop Summit (2,700 expected) [54]
2013 october Apache Hadoop 2.2 Available [38]
2014 february Apache Hadoop 2.3 Available [38]
2014 february Apache Spark Top Level Apache Project [55]
2014 April Hadoop summit Amsterdam (750 expected) [56]
2014 june Apache Hadoop 2.4 Available [38]
2014 june San Jose Hadoop Summit (3,200 expected) [57]
2014 august Apache Hadoop 2.5 Available [38]
2014 november Apache Hadoop 2.6 Available [38]
2015 April Hadoop Summit Europe [58]
2015 june Apache Hadoop 2.7 Available [38]
2017 march Apache Hadoop 2.8 Available [38]
2017 november Apache Hadoop 2.9 Available [38]
2017 december Apache Hadoop 3.0 Available [38]


See also: Hadoop Distributed File System , Apache HBase , and MapReduce

Hadoop consists of the Hadoop Common package, which provides an abstraction-based system, and a MapReduce engine (either MapReduce / MR1 or YARN / MR2) [59] and the Hadoop Distributed File System (HDFS). The Hadoop Common package contains the Java ARchive (JAR) files and scripts needed to start Hadoop.

For a successful scheduling of work, every Hadoop-compatible file system should provide location awareness – the name of the rack (or, more precisely, of the network switch) where a worker node is. Hadoop applications can use this information to execute code on the node where the data is, and, failing that, on the same rack / switch to reduce backbone traffic. HDFS uses this method when replicating data for data redundancy across multiple racks. This approach reduces the impact of a rack power outage or switch failure; if any of these hardware failures occurs, the data will remain available. [60]

A multi-node Hadoop cluster

A small Hadoop cluster includes a single master and multiple worker nodes. The master node consists of a Job Tracker, Task Tracker, NameNode, and DataNode. A slave or worker node acts as both a DataNode and TaskTracker, though it is possible to have data-only and compute-only worker nodes. These are normally used only in nonstandard applications. [61]

Hadoop requires Java Runtime Environment (JRE) 1.6 or higher. The Secure Shell (SSH) requires that the Secure Shell (SSH) be set up between nodes in the cluster. [62]

In a larger cluster, HDFS nodes are managed through a dedicated NameNode server to host the file system index, and a secondary NameNode that can generate snapshots of the namenode’s memory structures, thereby preventing file-system corruption and loss of data. Similarly, a standalone JobTracker server can manage job scheduling across nodes. When Hadoop MapReduce is used with an alternate file system, the NameNode, secondary NameNode, and DataNode architecture of HDFS are replaced by the file-system-specific equivalents.

File systems

Hadoop distributed file system

The HDFS is a distributed, scalable, and portable file system written in Java for the Hadoop framework. Some Consider It has to be INSTEAD data store due to ict Lack of POSIX compliance, [63] goal it does Provide shell commands and Java application programming interface (API) methods That are similar to other file systems. [64] A Hadoop cluster has nominally a single namenode plus a cluster of datanodes, although redundancy options are available for the namenode due to its criticality. Each datanode serves up blocks of data over the network using a specific protocol block to HDFS. The file system uses TCP / IP socketsfor communication. Clients use remote procedure calls (RPC) to communicate with each other.

HDFS stores large files (typically in the range of gigabytes to terabytes [65] ) across multiple machines. It achieves reliability by replicating the data across multiple hosts, and hence theoretically does not require redundant array of independent disks (RAID)storage on hosts (but to increase input-output (I / O) performance some RAID configurations are still useful). With the default replication value, 3 data stored on three nodes: two on the same rack, and one on a different rack. Data nodes can talk to each other to rebalance data, to move copies around, and to keep replication of data high. HDFS is not fully POSIX-compliant, because the requirements for a POSIX file-system differ from the target goals of a Hadoop application. The trade-off of not Having a fully POSIX-compliant file-system is Increased performance for data throughput and media for non-POSIX operations Such As Append. [66]

HDFS added the high-availability capabilities, as announced for version 2.0 in May 2012, [67] letting the metadata server (the NameNode) manually fail-over onto a backup. The project also began developing automatic fail-overs .

The HDFS file system includes a so-called secondary namenode, misleading term that some might incorrectly interpret a backup when the primary namenode goes offline. In fact, the secondary namenode regularly connects with the primary namenode and builds snapshots of the primary namenode’s directory information, which the system then saves to local or remote directories. These checkpointed images can be used to restart a file, then to edit the log file. Because it is a single point for storage and management of metadata, it can become a bottleneck for supporting a large number of files, especially a large number of small files. HDFS Federation, a new addition, to accommodate this problem to a certain extent by allowing multiple namespaces served by separate namenodes. Moreover, there are some issues in HDFS, such as small file issues, scalability problems, Single Point of Failure (SPoF), and bottlenecks in huge metadata requests. One advantage of using HDFS is between the job tracker and the job tracker. The job tracker schedules the map or reduce the jobs to the task tracker. For example: if node A contains data (x, y, z) and node B contains data (a, b, c), the job tracker schedules node B to perform a map or reduce tasks on (a, b, c) and node A would be scheduled to perform a map or reduce tasks on (x, y, z). This reduces the amount of traffic that goes beyond the network. When Hadoop is used with other file systems, this advantage is not always available. This can have a significant impact on job-completion times as demonstrated with data-intensive jobs.[68]

HDFS was conceived for mostly immutable files and may not be suitable for systems requiring concurrent write-operations. [66]

HDFS Can Be Mounted Directly With A Filesystem in Userspace (FUSE) virtual file system is Linux and Some Other Unix systems.

Java API, the Thrift API (C ++, Java, Python, PHP, Ruby, Erlang, Perl, Haskell, C #, Cocoa , Smalltalk, and OCaml ), the command-line interface , the HDFS-UI web application over HTTP , or via 3rd-party network client libraries. [69]

HDFS is designed for portability across various hardware platforms and for compatibility with a variety of underlying operating systems. The HDFS design introduces portability limitations that result in some performance bottlenecks, since the Java implementation can not use features that are exclusive to the platform on which HDFS is running. [70] Due to their widespread integration into an enterprise-level infrastructure, HDFS performance monitoring has become increasingly important. End-to-end performance monitoring requires tracking metrics from datanodes, namenodes, and the underlying operating system. [71] There are several HDFS performance monitoring platforms, including HortonWorks, Cloudera, and Datadog.

Other file systems

Hadoop works directly with any distributed file system that can be mounted by the underlying operating system by simply using a file://URL; however, this comes at a price – the loss of locality. Hadoop-specific file system bridges can provide.

In May 2011, the list of supported filesystems bundled with Apache Hadoop were:

  • HDFS: Hadoop’s own rack-aware file system. [72] This is designed to scale up the size of the operating systems .
  • FTP file system: This store has its data on remotely accessible FTP servers.
  • Amazon S3 (Simple Storage Service) file system: This is targeted at a cluster hosted on the Amazon Elastic Cloud Compute -server-on-demand infrastructure. There is no rack-awareness in this file system, as it is all remote.
  • Windows Azure Storage Blobs (WASB) file system: This is an extension of HDFS that allows the distribution of Hadoop to access data in Azure blob stores.

A number of third-party file system bridges have been written, none of which are currently in Hadoop distributions. However, some commercial distributions of IBM and MapR.

  • In 2009, IBM discussed running Hadoop over the IBM General Parallel File System . [73] The source code was published in October 2009. [74]
  • In April 2010, Parascale published the source code to run Hadoop against the Parascale file system. [75]
  • In April 2010, Appistry released a Hadoop file system driver for use with its own CloudIQ Storage product. [76]
  • In June 2010, HP discussed a location-aware IBRIX Fusion file system driver. [77]
  • In May 2011, MapR Technologies Inc. announced the availability of an alternative file system for Hadoop, MapR FS , which replaced the HDFS file system with a full random-access read / write file system.

JobTracker and TaskTracker: the MapReduce engine

Main article: MapReduce

Atop the file systems comes from the MapReduce Engine, which consists of JobTracker , to which client applications submit MapReduce jobs. The JobTracker pushes the task to the taskTracker nodes in the cluster, as much as possible. With a rack-aware file system, the JobTracker knows which node contains the data, and which other machines are nearby. If the work is not hosted on the actual node where the data resides, priority is given to nodes in the same rack. This reduces network traffic on the main backbone network. If a TaskTracker fails or times out, that part of the job is rescheduled. The TaskTracker on each node spawns a separate Java virtual machine(JVM) process to prevent the TaskTracker from failing if the running job crashes its JVM. A heartbeat is feeling from the TaskTracker to the JobTracker every few minutes to check its status. The Job Tracker and TaskTracker status and information is exposed by Jetty and can be viewed from a web browser.

Known limitations of this approach are:

  1. The allocation of work to TaskTrackers is very simple. Every TaskTracker has a number of available slots (such as “4 slots”). Every active map or task takes one. The Job Tracker allocates work to the nearest tracker with an available slot. There is no consideration of the current system load of the allocated machine, and hence its actual availability.
  2. If one TaskTracker is very slow, it can delay the entire MapReduce job – especially towards the end, when everything can end up waiting for the slowest task. With speculative execution enabled, however, a single task can be executed on multiple slave nodes.


By default Hadoop uses FIFO scheduling, and optionally 5 scheduling priorities to schedule jobs from a work queue. [78] In version 0.19 the job scheduler was refactored out of the JobTracker, while adding the ability to use an alternate scheduler (such as the Fair scheduler or the Capacity scheduler , described next). [79]

Fair scheduler

The fair scheduler was developed by Facebook . [80] The goal of the scheduler is fair to Provide fast response times for small jobs and Quality of Service (QoS) for producing jobs. The fair scheduler has three basic concepts. [81]

  1. Jobs are grouped into pools .
  2. Each pool is assigned a guaranteed minimum share.
  3. Excess capacity is split between jobs.

By default, jobs that are uncategorized go into a default pool. Pools have to specify the minimum number of slots, reduce slots, and more.

Capacity scheduler

The capacity scheduler was developed by Yahoo. The capacity scheduler supports several features that are similar to those of the fair scheduler. [82]

  1. Queues are allocated to the fraction of the total resource capacity.
  2. Free resources are allocated to queues beyond their total capacity.
  3. Within a queue, a job with a high level of priority.

There is no preemption once a job is running.

Difference between Hadoop 1 vs Hadoop 2 (YARN)

The biggest difference between Hadoop and Hadoop 2 is YARN technology. In the first version of Hadoop, the core components included Hadoop Common, HDFS, and MapReduce, but the second version of Hadoop came out with a new technology called YARN which was an acronym for Yet Another Resource Negotiator (YARN) .

It is an open source resource management technology which is deployed on a Hadoop cluster. YARN strives to allocate the resources to various applications effectively. It runs two daemons, which takes care of two different tasks: job tracking and progress monitoring .

These two daemons are called the resource manager and the application master respectively. The resource manager allocates resources to various applications, and the master application monitors the execution of the process.

Other applications

The HDFS is not restricted to MapReduce jobs. It can be used for other applications, many of which are under development at Apache. The list includes the HBase database, the Apache Mahout machine learning system, and the Apache Hive Data Warehouse system. Hadoop can, in theory, be used for any sort of work that is batch-oriented rather than real-time, is very data-intensive, and benefits from parallel processing of data. It can also be used as a real-time system, such as lambda architecture , Apache Storm, Flink and Spark Streaming. [83]

As of October 2009 , commercial applications of Hadoop [84] included: –

  • log and / or clickstream analysis of various kinds
  • marketing analytics
  • machine learning and / or sophisticated data mining
  • image processing
  • processing of XML messages
  • web crawling and / or text processing
  • general archiving, including of relational / tabular data, eg for compliance

Prominent use cases

On February 19, 2008, Yahoo! Inc. launched what they claimed was the world’s largest Hadoop production application. The Yahoo! Search Webmap is a Hadoop application that runs on a Linux cluster with more than 10,000 cores and produced data that was used in every Yahoo! web search query. [85] There are multiple Hadoop clusters at Yahoo! and no HDFS file systems or MapReduce jobs are split across multiple data centers. Every Hadoop cluster node bootstraps the Linux image, including the Hadoop distribution. Work that the clusters are known for the Yahoo! search engine. In June 2009, Yahoo! made the source code of its Hadoop version available at the open-source community. [86]

In 2010, Facebook claimed that they had the largest Hadoop cluster in the world with 21 PB of storage. [87] In June 2012, they announced that they had grown to 100 BP [88] and later that year they were growing up to a relatively low PB per day. [89]

As of 2013 , Hadoop adoption had more widespread: more than half of the Fortune 50 used Hadoop. [90]

Hadoop hosting in the cloud

Hadoop can be deployed in a traditional data center as well as in the cloud . [91] The cloud permits organizations to deploy Hadoop without the need to obtain hardware or specific setup expertise. [92] Vendors who currently have an offer for the cloud include Microsoft , Amazon , IBM , [93] Google, Oracle . [94] and CenturyLink Cloud [95]

On Microsoft Azure

Azure HDInsight [96] is a service that deploys Hadoop on Microsoft Azure . HDInsight uses Hortonworks HDP and is jointly developed for HDI with Hortonworks. HDI allows programming extensions with .NET (in addition to Java). HDInsight also supports the creation of Hadoop clusters using Linux with Ubuntu. [96] By deploying HDInsight in the cloud, organizations can spin up the number of nodes they want to use. [96] Hortonworks implementations can also move data from the data center to the cloud for backup, development / testing, and bursting scenarios. [96] It is also possible to run Cloudera or Hortonworks Hadoop clusters on Azure Virtual Machines.

On Amazon EC2 / S3 services

It is possible to run Hadoop on the Amazon Elastic Cloud Compute (EC2) and Amazon Simple Storage Service (S3). [97] As an example, The New York Times used 100 Amazon EC2 instances and a Hadoop application to process 4 TB of raw image TIFF data (stored in S3) in 11 million finished PDFs in the space of 24 hours at a computation cost of about $ 240 (not including bandwidth). [98]

There is support for the S3 object store in the Apache Hadoop releases, but this is what you expect from a traditional POSIX filesystem. Specifically, operations such as rename () and delete () are not atomic, and can take time proportional to the number of entries and the amount of data.

On Amazon Elastic MapReduce

Elastic MapReduce (EMR) [99] was introduced by Amazon.com in April 2009. Provisioning of the Hadoop cluster, running and terminating jobs, and handling data transfer between EC2 (VM) and S3 (Object Storage) are automated by Elastic MapReduce. Apache Hive, which is built on top of Hadoop for providing data warehouse services, is also offered in Elastic MapReduce. [100] Support for using Spot Instances [101] was later added in August 2011. [102]Elastic MapReduce is fault-tolerant for slave failures, [103] and it is recommended to only run the Task Instance Group on spot instances to take advantage of the lower cost [104]

On CenturyLink Cloud (CLC)

CenturyLink Cloud [105] offers Hadoop via both managed and un-managed models. [106] Cloudera Blueprints, the latest managed service in the CenturyLink Cloud, which also includes Cassandra and MongoDB solutions. [107]

On Google Cloud Platform

There are multiple ways to run the Hadoop ecosystem on Google Cloud platform from self-managed to Google-managed. [108]

  • Google Cloud Dataproc : managed Spark and Hadoop service [109]
  • command line tools (bdutil) : Spark and Hadoop clusters [110]
  • third party Hadoop distributions:
    • Cloudera – using the Cloudera Director Plugin for the Google Cloud Platform [111]
    • Hortonworks – using bdutil support for Hortonworks HDP [112]
    • MapR – using bdutil support for MapR [113]

Google also offers connectors for Google Cloud Platform products with Hadoop, such as Google Cloud Storage for Google Cloud Storage and Google BigQuery for Google BigQuery .

Commercial support

A number of companies offer commercial implementations or support for Hadoop. [114]


Apache Hadoop Project can be called Apache Hadoop or Apache Hadoop Distributions . [115] The naming of derivatives and the term “compatible” are somewhat controversial within the Hadoop developer community. [116]


Some papers influenced the birth and growth of Hadoop and big data processing. Some of these are:

  • Jeffrey Dean, Sanjay Ghemawat (2004) MapReduce: Simplified Data Processing on Large Clusters , Google. This paper inspired Doug Cutting to develop an open-source implementation of the Map-Reduce framework. He named it Hadoop, after his son’s toy elephant.
  • Michael Franklin, Alon Halevy, David Maier (2005) From Databases to Dataspaces: A New Abstraction for Information Management . The authors highlight the need for storage systems and the provision of data.
  • Fay Chang et al. (2006) Bigtable: A Distributed Storage System for Structured Data , Google.
  • Robert Kallman et al. (2008) H-store: a high-performance, distributed main memory transaction processing system