in vSphere documentations it's noted that the CPU Usage MHz of a VM is estimated from the host's point of view, not the guest operating system view.
which means the sum of the CPU MHz usages of all the VMs in a host should be equal or correlated to Host's CPU MHz usage.
however, I exported the usages, did a comparison between the sum of VM usages and the host usage and found the following:
The previous plot shows the host CPU Usage in MHz (in blue) and the CPU Usage in MHz of two VMs running random CPU load simulation (in green and orange)
however, when we take the sum of the VMs usages and compare it to the host I found the following:
When the sever is running at full capacity, sometimes the vms CPU usages is over estimated and hence the sum (in red) goes over the server CPU Cap.
I would like to know how this behavior can be explained?