Hadoop Structure | Simplilearn

Hadoop is a framework allowing the storage of enormous volumes of knowledge on node programs. It helps course of information in parallel utilizing a number of elements:

  • Hadoop HDFS to retailer information throughout slave machines
  • Hadoop YARN for useful resource administration within the Hadoop cluster
  • Hadoop MapReduce to course of information in a distributed style
  • Zookeeper to make sure synchronization throughout a cluster

Wanting ahead to turning into a Hadoop Developer? Try the Big Data Hadoop Certification Training Course and get licensed immediately.

Hadoop HDFS

The Hadoop Distributed File System (HDFS) is Hadoop’s storage layer. Housed on a number of servers, information is split into blocks based mostly on file dimension. These blocks are then randomly distributed and saved throughout slave machines. 

HDFS divides giant information into completely different blocks. Replicated 3 times by default, every block incorporates 128 MB of knowledge. Replications function below two guidelines:

  1. Two equivalent blocks can’t be positioned on the identical DataNode
  2. When a cluster is rack conscious, all of the replicas of a block can’t be positioned on the identical rack

Contemplate the next picture:


On this instance, blocks A, B, C, and D are replicated 3 times and positioned on completely different racks. If DataNode 7 crashes, we nonetheless have two copies of block C information on DataNode 4 of Rack 1 and DataNode 9 of Rack 3.

There are three elements of the Hadoop Distributed File System:  

  1. NameNode (a.okay.a. masternode): Incorporates metadata in RAM and disk
  2. Secondary NameNode: Incorporates a duplicate of NameNode’s metadata on disk
  3. Slave Node: Incorporates the precise information within the type of blocks


NameNode is the grasp server. In a non-high availability cluster, there might be just one NameNode. In a excessive availability cluster, there’s a risk of two NameNodes, and if there are two NameNodes there isn’t a want for a secondary NameNode. 

NameNode holds metadata info on the varied DataNodes, their areas, the scale of every block, and many others. It additionally helps to execute file system namespace operations, similar to opening, closing, renaming information and directories.

Secondary NameNode

The secondary NameNode server is answerable for sustaining a duplicate of the metadata within the disk. The first goal of the secondary NameNode is to create a brand new NameNode in case of failure. 

In a excessive availability cluster, there are two NameNodes: energetic and standby. The secondary NameNode performs an analogous perform to the standby NameNode.

Hadoop Cluster – Rack Based mostly Structure

We all know that in a rack-aware cluster, nodes are positioned in racks, and every rack has its rack swap. Rack switches are linked to a core swap, which ensures a swap failure is not going to render a rack unavailable.

HDFS Learn and Write Mechanism

HDFS Learn and Write mechanisms are parallel actions. To learn or write a file in HDFS, a shopper should work together with the namenode. The namenode checks the privileges of the shopper and offers permission to learn or write on the info blocks.


Datanodes retailer and preserve the blocks. Whereas there is just one namenode, there might be a number of datanodes, that are answerable for retrieving the blocks when requested by the namenode. Datanodes ship the block reviews to the namenode each 10 seconds; on this means, the namenode receives details about the datanodes saved in its RAM and disk.

Huge Knowledge Hadoop Certification Coaching Course

Grasp Huge Knowledge and Hadoop EcosystemExplore Course

Hadoop YARN

Hadoop YARN (But One other Useful resource Negotiator) is the cluster useful resource administration layer of Hadoop and is answerable for useful resource allocation and job scheduling. Launched within the Hadoop 2.zero model, YARN is the center layer between HDFS and MapReduce

The weather of YARN embody:

  • ResourceManager (one per cluster)
  • ApplicationMaster (one per utility)
  • NodeManagers (one per node)

Useful resource Supervisor

Useful resource Supervisor manages the useful resource allocation within the cluster and is answerable for monitoring what number of assets can be found within the cluster and every node supervisor’s contribution. It has two major elements:

  1. Scheduler: Allocating assets to varied operating purposes and scheduling assets based mostly on the necessities of the applying; it doesn’t monitor or monitor the standing of the purposes
  2. Software Supervisor: Accepting job submissions from the shopper or monitoring and restarting utility masters in case of failure

Software Grasp

Software Grasp manages the useful resource wants of particular person purposes and interacts with the scheduler to accumulate the required assets. It connects with the node supervisor to execute and monitor duties.

Node Supervisor

Node Supervisor tracks operating jobs and sends alerts (or heartbeats) to the useful resource supervisor to relay the standing of a node. It additionally displays every container’s useful resource utilization.


Container homes a set of assets like RAM, CPU, and community bandwidth. Allocations are based mostly on what YARN has calculated for the assets. A container supplies the rights to an utility to make use of particular useful resource quantities.

Steps to Working an Software in YARN

  1. Consumer submits an utility to the ResourceManager
  2. ResourceManager allocates a container
  3. ApplicationMaster contacts the associated NodeManager as a result of it wants to make use of the containers
  4. NodeManager launches the container 
  5. Container executes the ApplicationMaster


MapReduce is a framework conducting distributed and parallel processing of enormous volumes of knowledge. Written utilizing a number of programming languages, it has two major phases: Map Section and Cut back Section.

Map Section 

Map Section shops information within the type of blocks. Knowledge is learn, processed, and given a key-value pair on this part. It’s answerable for operating a selected job on one or a number of splits or inputs.

Cut back Section

Cut back Section receives the key-value pair from the map part. The important thing-value pair is then aggregated into smaller units, and an output is produced. Processes similar to shuffling and sorting happen within the cut back part.

The mapper perform handles the enter information and runs a perform on each enter cut up (generally known as map duties). There might be one or a number of map duties based mostly on the scale of the file and the configuration setup. Knowledge is then sorted, shuffled, and moved to the cut back part, the place a cut back perform aggregates the info and supplies the output.

Are you expert sufficient for a Huge Knowledge profession? Attempt answering these Big Data and Hadoop Developer Test Questions and discover out now!

MapReduce Job Execution

  • The enter information is saved within the HDFS and browse utilizing an enter format. 
  • The file is cut up into a number of chunks based mostly on the scale of the file and the enter format. 
  • The default chunk dimension is 128 MB however might be personalized. 
  • The file reader reads the info from the enter splits and forwards this info to the mapper. 
  • The mapper breaks the data in each chunk into a listing of knowledge components (or key-value pairs). 
  • The combiner works on the intermediate information created by the map duties and acts as a mini reducer to cut back the info. 
  • The partitioner decides what number of cut back duties will probably be required to mixture the info. 
  • The info is then sorted and shuffled based mostly on their key-value pairs and despatched to the cut back perform. 
  • Based mostly on the output format determined by the cut back perform, the output information is then saved on the HDFS.

Get an in-depth understanding of the structure of Apache Hadoop from the next video tutorial –

Grasp the Ideas of the Hadoop Framework

Companies are actually able to making higher choices by gaining actionable insights by large information analytics. The Hadoop Structure is a serious, however one side of all the Hadoop ecosystem. 

Be taught extra about different points of Huge Knowledge with Simplilearn’s Big Data Hadoop Certification Training Course. Aside from gaining hands-on expertise with instruments like HDFS, YARN, MapReduce, Hive, Impala, Pig, and HBase, you too can begin your journey in the direction of reaching Cloudera’s CCA175 Huge Knowledge certification.

About the Author

Leave a Reply

Your email address will not be published. Required fields are marked *