path_latency_threshold: Why only in Seconds? Anyone using it effectively?

Question

I was excited to learn about this powerpath feature only to see that's it not very useful. I'm hoping someone is finding it useful. If a system is normally running at 10ms or less for all I/O, it would be nice to know when it goes to 15 or more and, try to resolve it proactively. The path_latency_threshold seems to work only 1 sec increments.  Setting it would only log a notification in the system log when the I/O goes to 1000ms or more. This does not seem very useful in reality. We get calls on applications that go from 4ms to 6ms, let alone going to 1 sec. It would be nice if this worked with milliseconds instead of seconds. Anyone finding this parameter useful in their env?

SKT2 · Answer

how r u checking this value?

Brion2 · Answer

Hi. Please keep in mind that SCSI timeouts are in seconds (with the SCSI driver having a 60 second timeout). PowerPath sits above the SCSI driver. So, an I/O that was retried at the SCSI driver could have a rather large latency value.

You do have a way of using this to discover the latency in this setup.

1. powermt set path_latency_monitor=on

2. powermt set path_latency_threshold = 0

This is called discovery mode and will cause a threshold cross every time a new high watermark is discovered.

This high watermark is also visible from the "powermt display latency".

I hope this helps.

SKT2 · Answer

is there a possible performance impact by enabling the path_latency_monitor? Did u mean high watermark  is the 'latency max' displayed from `powermt display latency`

Brion2 · Answer

Hello.  No performance impact has been documented to date.  Yes, I am describing max latency.

PowerPath

Was this post helpful?