Originally posted by: lylklb

This might be a historical and recessive problem in AIX(5/6/7) nMPIO , that is, when one of the multipath links broken/failed, all the other good links' I/O would also together hang up for a short time (15~30s)！ So how to resolve such problem？

Notes: This should be nothing to do with the attributes of fscsi/hdisk because we have already set up for dyntrk=yes & fc_err_recov=fast_fail with fscsi and round-robin with nMPIO.

Originally posted by: Chris Gibson

Could be a few things. Read this: https://www.ibm.com/developerworks/aix/library/au-aix-mpio/index.html

And this: https://www.ibm.com/developerworks/aix/library/au-aix-multipath-io-mpio/index.html

Originally posted by: lylklb

I don't think that the two following doc reference could explain why the I/O stream of all the other good links would hang up for a short time(15~30s) when one path was down！

If a path fails or is disabled, it is no longer used for sending I/O. The priority of the remaining paths is then recalculated to determine the percentage of I/O that should be sent down each path. If all paths have the same path_priority value, the PCM attempts to equally distribute I/O across all enabled paths.

All the MPIO paths that use the failed component indicate a Failed state, and error log entries are created against that component, and the MPIO software ceases using the failed paths for user I/O.

Originally posted by: Chris Gibson

It sounds like the behaviour you are experiencing is not expected or correct. I would recommend you open a case with IBM support for investigation.

Originally posted by: lylklb

In fact,  this is a common and underlying problem,  not a special issue！

Referring to the following similar case links:
https://community.emc.com/thread/117936?start=0