Hadoop Servers

Client Server

• Client is neither a master or a slave. Its role here is to store data on hadoop cluster and submit MR jobs with instructions saying how data has to be processed. Also Client can retrieve & view the result after completion of job.
• Data Storage:  Client accesses the file system on behalf of users by communicating with NameNode & DataNode.
• Data Processing: To begin the data processing job, client calls the Job Tracker (Master Daemon), Job Tracker then partitions the work & assigns different Map & Reduce tasks to each Task Tracker in the Hadoop Cluster.

Master Servers (Machines)

• Key Role of masters is to store large amount of data on HDFS & perform parallel computation on this data using Map Reduce programs (Algorithms). 
• NameNode, Secondary NameNode & Job Tracker are the masters servers in Hadoop.
• Master Servers are CPU/Memory intensive.
• Master Servers always assign & monitor the task, they never work on data directly.
• In a hadoop cluster there will be only 1 NameNode, 1 Secondary NameNode & 1 Job Tracker.
• NameNode is also called as Brain of Hadoop as it stores the Metadata.
• Secondary NameNode is also called as CheckPoint Node & it does all housekeeping job. Secondary NameNode is not the backup for Primary NameNode during failure state but with some manual intervention, cluster can be reconfigured to use Secondary NameNode as a Primary NameNode.
• Job Tracker overseas & coordinates the parallel processing of data using MapReduce algorithms.

Slave Servers (Machines) 

• Key role of slaves is to do all dirty job, they are also called as work horses in Hadoop Cluster.
• Slaves always work on the task assigned by masters, they dont act/work by there own.
• Slave Servers are Storage intensive. So data node always have lots of storage
• DataNode & Task Tracker are daemons run on every slave node (Machine).
• Data will be actually stored on DataNodes in equal size chunks/blocks.
• Task Tracker monitors task running on individual slave machine and coordinates with JobTRacker.
• Each slave machine has Task Tracker daemon.
• DataNode is slave to NameNode & they primarily act while storing data to HDFS.
• Task Tracker is slave to Job Tracker and they primarily act while running the parallel computation using MapReduce.
• DataNode to DataNode communication in Hadoop Cluster is removed except during data replication.

Master Slave Relation in Hadoop Cluster - (Data Storage & Data Processing)