Hadoop Distributed File System (HDFS) Federation
The solution to expanding Hadoop clusters indefinitely is to federate the NameNode. Before Hadoop 2 entered the scene, Hadoop clusters had to live with the fact that NameNode placed limits on the degree to which they could scale. Few clusters were able to scale beyond 3,000 or 4,000 nodes.
NameNode’s need to maintain records for every block of data stored in the cluster turned out to be the most significant factor restricting greater cluster growth. When you have too many blocks, it becomes increasingly difficult for the NameNode to scale up as the Hadoop cluster scales out.
Specifically, you must set HDFS up so that you have multiple NameNode instances running on their own, dedicated master nodes and then making each NameNode responsible only for the file blocks in its own name space.
The figure shows replication patterns of data blocks in HDFS. You can see a Hadoop cluster with two NameNodes serving a single cluster. The slave nodes all contain blocks from both name spaces.