Diagram: This is how HDFS on Isilon works ...

Before, I was under the impression Isilon just took care of the Datanode portion of the Hadoop stack, but it seems it also runs the Namenode portion, which makes sense as to why they are able to handle ingesting data in using standard protocols like CIFS and NFS as they are handling the namespace portion of the Namenode.  It seems to be take care Namenode High Availability and NameNode Federation.

The only thing that concerns me is the fact that if there are issues with performance, this gives you less flexibility to diagnose and resolve the issue. You are also dependent on EMC’s flavor of HDFS and their roadmap … as new features are released and bugs fixed, how do they handle that?

Here’s the Hadoop stack with Isilon in the mix …