I'm aware of the other question on this topic, but it doesn't quite scratch my itch.
Here's my - probably quite simplistic - model for how CPU work gets done. On the host we have a pool of logical CPUs. On the VMs we have some number of virtual CPUs. Each vCPU is periodically given stuff to do. The hypervisor chooses a spare logical CPU from the pool to actually do the stuff. When the stuff is done, the logical CPU returns to the pool.
On this model, 'CPU Wait Time per Dispatch' represents the time between the vCPU having stuff to do and the hypervisor being able to find a spare logical CPU. So - like any other equivalent setup that makes use of a pool of resources - the metric that measures the wait for a resource to be free should remain effectively constant until there is a point at which the pool is empty and the hypervisor has to wait for a logical CPU to be returned to the pool.
Suppose, further, that we have fewer vCPUs than logical CPUs (such that the management overhead of the hypervisor on the logical CPUs can be discounted). It seems to me that in this case we could never exhaust the pool of logical CPUs, so the CPU Wait Time per Dispatch value would have to remain effectively constant. (This does assume that the work of one vCPU can only ever be handled at any time by a single logical CPU, but I don't see how that can be wrong.)
Does this seem correct, or am I missing something important? I have a colleague at work who deals with the hyper-v hosts, who tells me I am wrong, but is struggling to communicate how (or at least I am struggling to understand him).
