AnsweredAssumed Answered

MFS Slow disk detection mechanism

Question asked by akarjp on Jan 31, 2013
Latest reply on Feb 4, 2013 by akarjp
We have a 16 node MapR (M3 cluster running on a mix of Dell PowerEdge 410
and PowerEdge 415 servers. Each one has three 3TB disks in a single storage pool for mfs.

Recently we started running test jobs on the cluster and we started to experience several
instances of MapR reporting disk failures (I/O timeout, Slow Disk). Informal hdparm tests
seem to indicate the drives are suitably zippy (140 MB/s).

We are aware that the mfs timeout parameter can be modified to mitigate this condition,
but we are curious about the mechanism or process that is used to detect underperforming
disks? Also, would there be any detrimental affect of setting the timeout too high (we
currently have it set it to 300 secs.)?