I have a cluster of HP BL685c G6 running ESXi5.1 with Enterprise Plus licenses connected to a vcenter server 5.1 environment.
DRS is enabled and set to fully automated.
The cluster runs close to 100 VMs. Nobody is screaming that performance is bad, but in my daily checks of the environment I constantly notice that the resource distribution chart shows seemingly unhealthy CPU resource delivery. yellows all over the place.
In addition to this, vcops shows an alarming risk assessment of the cluster stating that I'm over provisioned by dozens of VMs!
yet, when I investigate each host individually in vcops they tell a much different story where some are so under utilized as to allow for over 100 more VMs per host!
What could be causing this and what might help?
For comparison sake, I also have another cluster with HP BL685c G7 hosts similarly licensed in the same vcenter server environment running over 100 VMs and every one of them is green and being supplied with nearly 100% of the requested resources. DRS is set exactly the same.
I would think that if my hosts were not up to the task of supplying the necessary CPU resources the VM objects in this chart would be much larger showing that they are consuming more resources than the host can handle, but as you can see from the above chart they are all very small and total utilization per host is far below even 25%.
I had a similar issue on a Dell Chassis.
I found if you turned off the power management/profile in the BIOS (set it to performance as opposed to performance/watt) I no longer had this problem.
Not sure you actually need to do this...some research on the bios/server power management (in the optimize for perf per watt mode it was in) say that it should not effect performance significantly as it should feed power to the CPU as it needs it. But it does seem to lag enough that the VMware chart is effected.
Found the answer here:
In my experience this can be caused by problems with your vSphere HA configuration. Make sure all your hosts are connected to the HA master. If a host is not connected to the HA master, or isolated, it's resources will not be counted in the VROPS workload calculation and will likely trigger workload alarms.