we have HDFS cluster based on Hortonworks HDP 2.6.5 version
cluster include 328 data node + node-manager machines and 2 name-node machines ( when name-nodes are high availability)
we runs spark application on node-manager machines , and from our testing we are feeling that reading / writing to HDFS data is slow then what was in the past
HDFS configuration include many parameters and maybe by the right HDFS parameters tuning we can get better performance