What happens if there is contention and the reservation kicks in, but then within the resource pool these VMs compete for the 6GHz available to the resource pool?
Do they get assigned shares of the available 6GHz based on the number of vCPus they have (1000 shares per vcpu)?
If so I can't guarantee them each 2Ghz - but I need to avoid per-vm reservations.
I might not be right, but is a possible option is to have a 6Ghz resource pool at the top level in the cluster and then have 3 separate resource pools for each VM at 2Ghz underneath that?
Pool (6 Ghz) --> Pool 1 (2Ghz) --> Pool 2 (2Ghz) --> Pool 3 (2Ghz).
Is that feasible to you?
OK thanks for the idea - how much better is that than setting an individual VM reservation? It does allow me to view and manage things at the cluster level and keep track of it better - what are other comparison factors between that and individual VM reservations?
Well, the key things that I can think of are:
- VM reservations will affect your HA slot size if you are specifying numbers of host failures to tolerate.
- VM reservations take affect when the VM powers on but Resource Pool is active immediately/always.
My understanding is that CPU reservations are considered "friendly" where as memory reservations can be worse. I also think that Resource Pool is generally considered "friendly" where as per VM is not as much.
I'd highly suggest that if you want to know more, to go read some awesome articles by the legends that are Duncan Epping or Frank Denneman. They pretty much have the HA/DRS and resource market covered off to the hilt!