Reply to Message

View discussion in a popup

Replying to:
sxnxr
Commander
Commander

I will try to answer some of this but bare in mind i am making some assumptions and could be wrong

But enabling admission control will also block any new start of VM if ressources are not available. So even if we are proactive with vrops, we can be block by admission control.

This is a problem with your admission control. I am amusing you have it disabled (dont worry it is the same for you and about 90% of the world) The point of that message is so HA has enough resources to power on the VMs if there is an HA event. Simply ignoring this and disabling admission control will only lead to problems later. Setting an appropriate admission control is a must in my mind. Setting memory reservations will effect HA the most by reducing the HA slots available.

But, how can you explain me that all my cluster are fine (1 yr remaining) and for some clusters i have CPU contention (7%) and vrops doesn't warm me. My clusters densification is too high and the CPU overcommit too (1:8.5)

You are mixing capacity models here. Over commit is allocation based capacity model. If you are using this i would expect Contention to be all over the map depending on the VM workloads or lack of

CPU contention is not used in the Allocation model but in the demand based model. If using a demand based model i would expect the allocation ratios to be all over the place as well depending on the workload of the VMs.

You could have low contention and still have 10 or 15 to on consolidation ratios for example i have 70:1 with 2% contention in one cluster and 10 to 1 with 30% contention due to the hardware we are running on and the power policy the servers are set to.

This is why we are moving to a performance based capacity model and away from both of these

Reply
0 Kudos