Seeing latency from the application, but not on the VMAX.
I've got a Win 2012 SQL server which is seeing pretty significant latency, up to 120ms or so. The disks are all bound to FC, and PowerPath is installed. My execution throttle is 65,535 on the QLogic HBA.
I did notice my Driver Port Down Retry Count, Link Down Timeout, and LUNS Per Target are all below EMC's recommendations.
Not sure if these would make any difference.
The SG's on the VMAX report a consistend 4-5ms response time. Front-end ports are 60% busy. Would changing those HBA settings net us any better latency from a SQL perspective? The fabric looks okay. No errors there.
Re: Seeing latency from the application, but not on the VMAX.
This is almost certainly going to be I/O queuing at the FE ports. The VMAX doesn't report the time spent in queue as part of what is reported for latency. Actual latency measured at the host is equal to time spent in the FE queue plus time spent processing the I/O plus a small amount of FC transport latency through your FC switches. Every time I've seen a significant difference between host reported latency and VMAX reported latency, it ends up being front end queuing.
Th check for queuing, look at the "Queue Depth Count Ranges" in the VMAX performance tool. You can find them under "All" metrics when looking at FE ports. There are 10 of them. If you see any I/O in the last range, you have significant queuing. Probably anything over 4 or 5 is worrisome.