We just upgraded to 5876.229.145 today on our VMAX 20K and since the upgrade i'm seeing higher Back End DA utilization. It's about 15% increase with the same workload so now we are averaging in the 80% range with spikes to high 90's. I know in different 5876 releases they changed how much CPU was used for FAST VP. Did they bump it back up to what it was with the origional 5876?
What you observe is expected at this level of code. Additional tasks are now counted which can account for a 10-20% increase in reported DA CPU busy. Disk response times should not have degraded as a result (maybe even improved) of this upgrade. Also at this code version, the FAST relocation rate may be a little more aggressive, contributing to DA busy. You could try changing your FRR value if you are actually seeing performance degradation as a result of the code upgrade.
The fix that increased the DA busy in some cases was 66045, and was introduced in Arno. It wasn't written up in the release notes, but should have been. For many workloads it made the DA % busy more accurate. In some cases it also counted some background iVTOC and WP scan activity in the % busy.
There is a fix (67571) that is in the Q3SR that will move that iVTOC scan and some of the WP scan activity to the % LP busy.
Quincy - Are these release notes you are referring to available on the support site or are the release notes only available to EMC employees? The notes on the support site that I saw didn't seem to point any of this out.
Also any details of the fix you are referring to would be great. Definitely seems to be a fix that we would benefit from.
PedalHarder
4 Apprentice
•
465 Posts
252
1
Posted September 4th, 2013 21:00
What you observe is expected at this level of code. Additional tasks are now counted which can account for a 10-20% increase in reported DA CPU busy. Disk response times should not have degraded as a result (maybe even improved) of this upgrade. Also at this code version, the FAST relocation rate may be a little more aggressive, contributing to DA busy. You could try changing your FRR value if you are actually seeing performance degradation as a result of the code upgrade.