Which value of MPS the hypervisor use for scheduling?
I would like to ask a different question about the host scheduler.
More or less, the host scheduler uses the MhzPerShares(MPS) metric to schedule VMs. This means that, according to the VM shares, the scheduler selects the VM with the lowest MPS to be scheduled, thereby the VM with the lowest utilization of its resources.
However, considering that a VM can be scheduled with fewer resources than provisioned (a 8-vCPU VM may be scheduled with 2 vCPUs initially), it raises the question what number of shares the host scheduler uses to calculate MPS. The shares that are being used by the VM the current moment(thus 2000) or the nominal number of shares the VM has been provisioned (thus 8000 shares)?