I recently noticed something when trying to resolve a stability issue with my nested lab setup (ESXi 5.1 inside Workstation 10). The "Package C-states" setting in my motherboard BIOS has an impact on the idle CPU utilization of my ESX hosts. The nested ESX hosts were hovering around 15% CPU for no apparent reason. ESXTOP reflected the 15% CPU usage but all running processes were under 1-2% total. Odd. Now my ESX installs are at 1% (with no nested VMs running).
Worth a look if you have a nested lab setup. I also had to disable "report C6" and disable "enhanced C1 (C1E)" to make everything stable.
Has anyone else seen this kind of behavior?
Yep, I've run into a lot of issues in production environments (not nested) as a result of power state options. The fix is usually setting everything to high performance.
When the system is switching between states there is latency introduced, additionally if you have a host with 8 physical core's available, and there's a VM with 1 proc hitting 100% the board might think that technically the physical CPU isn't working very hard and adjust c states which will impact performance on that VM.
Here's an interesting link on it: VMware KB: Poor virtual machine application performance may be caused by processor power management ...
I just found another error from WHEA-Logger (Event ID 19). I disabled C7... We'll see if that helps.
So far my CPU operating temp hasn't changed. It will definitely ramp up if I disable ALL power mgmt settings; the CPU will idle around 65C instead of the usual 42C. Speedstep is an important one from a power savings perspective; hopefully I can keep it.