We have recently come across a discrepancy between CPU utilization on a webserver using Vcenter and W2k8. I understand, from training, that Vcenter is considered the most accurate indicator for performance monitoring. Recently we reduced the # of CPUs assigned to the server from 3 to 2 based on the utilization shown in Vcenter; we are trying to minimize CPU ready time by rightsizing our servers. Shortly after making the change we started getting calls about performance. The problem seems to be that with 2 processors W2K8 shows processor utilization at 100% for several seconds when one of the processes spikes. Vcenter at the same time never goes above 50% utilization! We have increased the # of CPU back to 3 and the problem resolves itself. The piece I am struggling with is that even with 3 CPU, W2K8 CPU utilization spikes to 85% but Vcenter shows spikes to only 35%. It almost seems that W2K8 does not know about all of the processor that it has to work with.
This leaves me with a couple of questions:
1. Why are we be seeing such a discrepancy?
2. Is the direction we received in training, to use vCenter to monitor performance, incorrect?
3. Is there a better way to evaluate the correct number of CPU to assign to a server?
Thanks in advance.
Willie
