Comment 4 for bug 1371360

Using Cinder for HDFS has performance issues that are not well understood.
Once you lose data locality, computation will no longer happen near the data and jobs will make heavy use of the network. Many of the assumptions that Hadoop/Spark make are no longer valid when HDFS is backed by network volumes.

While Sahara supporting Cinder is a good thing, I would be wary of even suggesting such a solution without a well-thought document of the pros and cons that people can read and understand.