We're in the middle of a POC to demonstrate to a customer the impact to performance during a total director failure. Running a simple workload and system is running about 80K IOPS with sub 1ms read and write response time. Performance is steady and not a problem.
When we fail the director, Unipshere RealTime statistics shows a spike in performance up to 95+K IOPS and stays there until the director is recoved using Symmwin scripts, at which point, the Host IOPS drops back down to steady 80K IOPS.
What's confusing is that Inlines and IOMeter both show a decrease in IOPS, down to 65K - 70K IOPS, whereas Unisphere shows an uptick to 95-100K IOPS... Why the differance?
Problem only seems to be with RealTime. When we go back and look at Historical during the same timeframe, the charts are accurate.
High level config:
VMAX450F 2-engine, 8 paths to hosts (4 per engine), (2) ESX servers with 2 HBA paths each, powerpath installed.
Unisphere 188.8.131.52 (Also tried 184.108.40.206) with same results.
If you haven't done so, contact your local EMC Sales Team and get in touch with the account's SPEED Guru. They would be your fastest path to a concrete explanation of your condition, and possibly indicate why Unisphere is inaccurate (Inlines is possibly doing something funky, but explainable and "missing" data from the down engine, misrepresenting what's actually going).