Hadoop Architecture Guide 101

By Niveditha

Last updated on Mar 7 2022

Hadoop Architecture Guide 101

Introduction to Hadoop

Hadoop was introduced on 1-April-2006, developed by Apache Software Foundation. Developers/authors of Hadoop are Michael j. Cafarella and Doug Cutting.

Hadoop is open-source software used for storing data and running applications, grouped to export hardware.

Hadoop is used effectively to store and process a wide range of datasets of size, from GB-PB. Hadoop is used by great organizations like Facebook, Apple, IBM, Google, Twitter, and hp. What makes Hadoop reliable? Let us learn in detail,


Features of Hadoop

Hadoop features are what makes it more impressive. They are as follows,

  • Open Source
  • Highly flexible cluster
  • Liberality in Fault
  • Easy approach
  • Cost Effective
  • Data Placement


Features of Hadoop make it adaptable and easy to use the software. 

Similarly, Hadoop architecture is used to understand certain software applications. Let's learn in detail about what makes Hadoop architecture an effective solution.


Learn more and get trained by the experts through Sprintzeal.


Hadoop Architecture Explained


New things are being introduced and developments are being done for a better future in technology, one among them is Hadoop. Hadoop has become an effective solution in today’s time. It carries an impressive set of goals across various hardware platforms. 

Hadoop's effective work performance regarding Hadoop architecture makes it essential to use open-source software. Let us understand the importance of Hadoop architecture in detail.

Hadoop Architecture acts as a topology, where one primary node contains multiple child nodes. Primary nodes assign several tasks for child nodes to manage resources. The child node or the sub-node does the actual work of computing. The child node stores the real data whereas the primary node holds the metadata meaning data is stored within data.

Let us understand the Diagram of Hadoop Architecture and its applications in detail.

The Diagram of Hadoop architecture contains three important layers. Those are as follows,

HDFS (Hadoop Distributed File System)

Map Reduce



Big data hadoop


About HDFS


Hadoop Distributed File System HDFS Architecture enables data storage. Enabled data is converted into small packets or units known as blocks. The converted data is then stored in a distributed way. It has two different domains running one for the primary node and one for the child node.


HDFS architecture in big data considers the Primary node as NameNode and other child nodes as a data node. Let's learn in detail, 


NameNode and DataNode

HDFS is a Primary-Slave architecture that runs the NameNode domain as the master server. Here it is in charge of namespace management and modulates the file access by the client.

DataNode runs on the child or the sub-nodes as it stores the required business data. Here the file gets split into several blocks and stored in the sub machines.

NameNode keeps track of the mapping of blocks to DataNodes.  Adjacent to that DataNodes read/write requests. Then forward it to the client’s system file. DataNodes also create, delete and repeat the demanding blocks of NameNode.


Block in HDFS

Blocks are small packets of storage in the system. The neighboring smallest storage value is allocated to a file.

Hadoop has a default block size of 128-256MB, for NameNode and DataNode.


Replication management

To provide one of the features of Hadoop, fault tolerance / Liberality of HDFs, Hadoop uses the replication technique. Where it copies the blocks and stores them on different DataNodes. Through replication, it’s decided how many copies of the blocks are to be stored. And three is the default value to configure data.


Rack Awareness

Rack awareness in HDFS is about maintaining the DataNode machines and producing several racks of data. HDFS in general follows an algorithm on rack awareness.



MapReduce architecture is a data processing layer of Hadoop Architecture. To process large amounts of data writing, applications are allowed through the Hadoop software framework.

MapReduce runs applications simultaneously on Hadoop cluster architecture for low-end machines. It is set to reduce tasks and each task works as per the mapped data.  A load of data is distributed across the group so the data can be filtered and transferred.

And there is stored data on HDFS as an input file of map-reduce which is split and processed to store as HDFS replications. Let’s go through the phases that occur and the process in detail.


Map Task

Map task contains phases mentioned below,


Record Reader

The record reader converts input split (logical split) into records. It claims only data into records but does not claim records in single. The Map function is provided with key-value pairs. Keys hold the positional information and value holds the data record.



The Mapper or the Map contains the subroutine of a key-value pair from the record reader. It contains zero or multiple midway key-value pairs.

The precise key-value pair is decided by the mapper function. Data that gets aggregated gets the final result from the reducer function.



A combiner is a localized reducer that helps group the data in the map phase.



It performs a set of modulus operations with the help of reducers in numbers. The partitioner is all about getting the intermediate key-value pair from the mapper.


Reduce Task


There are certain phases involved in reduce tasks which are as follows,


Shuffle and sort

Individual data pieces are sorted into large data lists. The data written by the Partitioner is downloaded to the machine where the reducer is running.

Shuffle and sort is about controlling and sorting the keys, in alignment. Where the tasks could be performed easily and a sorted object is given.



Reduce performs the function of reduction as per key grouping.

Reduced function gets the end key-value pairs to the output format, and gives zero as the reduced value. It is similar to the map function as changing from one task to another.



It is the final task where the key-value pair from the reducer is written by the record writer. Each key is separated to have a new record by a newline character. The final output is written to HDFS.



Yet another Resource Negotiator is a resource managing layer for Hadoop architecture. The main goal of yarn is to separate the resource and monitor functions into separate domains.

YARN architecture contains one global resource manager and an application master, for every single job.

The domain resource manager and application masterwork with the node functions will execute and complete the job. And the resource manager from Hadoop yarn architecture contains scheduler and application manager as two important components to negotiate.

The scheduler is the one that allocates resources to applications.

Application manager performs certain functions through application master and they are as follows,

  • Negotiates resources of the scheduler.
  • Keeps track of resource
  • Monitors the application to be in progress.


Features of Yarn


The yarn contains four major features in Hadoop, which are as follows,




It has access to various engines on the same Hadoop data set.


Cluster Utilization

Dynamic allocation of resources, where static reduces in map compared to previous versions of Hadoop with lesser utilization of groups.



The strength of data keeps on increasing as data is processed in petabytes PB.



Without any interruption, work can be completed using yarn. No disruption occurs as it acts as a map-reduce program for Hadoop.


Learn in detail about Hadoop with  Big Data Hadoop and Spark Developer Course


<iframe width="640" height="360" src="https://www.youtube.com/embed/eurcyTxpGNE" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe>



Hadoop is a very powerful open-source software developed for system work. Hadoop architecture in big data is effective with reliable software applications. With Hadoop, it’s made easy to interact with greater platforms. 

Learn about Hadoop to make a good start in your cloud computing career, enroll in Sprintzeal's Big Data Hadoop Training program

About the Author

Sprintzeal   Niveditha

Niveditha is a content writer at Sprintzeal. She enjoys creating fresh content pieces focused on the latest trends and updates in the E-learning domain.

Recommended Resources

IT Certifications List – Most Popular Certifications in 2022

IT Certifications List – Most Popular Certifications in 2022


CAPM Cheat Sheet 2022

CAPM Cheat Sheet 2022


Top DevOps Interview Questions and Answers 2022

Top DevOps Interview Questions and Answers 2022