I'm running VMWare and have an ESX host that is running at 100% CPU. I also have two Exchange 2010 CAS servers that are running at 100% CPU,
Here are some additiona observations:
The sum of the Processes tab is only a fraction of the total CPU. It appears kernel time is not included when calculating process time. (Top process was RPC client service at 40%)
When I VMotion one server to a different host, the first VM immediately goes to 20% (normal load)
When I VMotion the second server stays at 100% for a period of time, then goes to normal load after 2 minutes.
Image below
Question
What Perfmon metric would be most useful in determining contention within the CPU in this scenario?