I've been reviewing our HA configuration on one of our clusters, & the slot size is being set at default for CPU @ 256mhz. This is because we don't have any CPU reservations set on any of our VMs, otherwise HA would select the highest reservation as a "worst case scenario" and use that as the slot size.
I'm concerned because I don't think HA is going to be able to accurately calculate for a failover scenario in our cluster because the slot size is far to small to reflect a real world VM in our cluster. So I'd like to increase the slot size to something more conservative.. say 600-800mhz.
My question is, firstly is this a good idea? And if so is the best way to do it by just picking a VM and setting a CPU reservation on it of say 600mhz, or should I use the built in option das.slotCpuInMHz to set this?
Thanks in advance
Note: This discussion was moved from thecommunity to the community.
please keep in mind that one VM can span more than one HA slot. You can find a very good Article on this topic here www.yellow-bricks.com under the HA Deep Dive. I thinck after reading that it should be more cleare to you.
The das.slotCpuInMHz advanced option lets you set a _cap_ on the slot size, not an exact value. It is used to ignore outlier vms (with larger reservations than the rest) so that won't help you here. If you want to increase the default slot size, you can set the advanced option das.vmcpuminmhz (see page 27 in the Availability Guide - http://www.vmware.com/pdf/vsphere4/r41/vsp_41_availability.pdf).
Thanks for that.
I've had a read through the Yellow Bricks HA article, but it still leaves a few questions for me.
Basically I'm concerned that our HA cluster is reporting that in a 5 node cluster we can handle 3 hosts going offline, it gives a very unrealistic impression that our cluster is under utilised.
These hosts are each currently running at around 80% CPU, so there is no way we can lost 3 hosts and run all VMs on just 2 hosts. So I believe the HA calculations are incorrect.
It is also reporting that we have 280 total slots in the cluster, and 143 available.. Again there's no way we could add an additional 143 VMs to this cluster without things going pear shaped.
Increasing the slot size seem to me like a valid way of making the above figures more closely reflect whats in our environment.
Does that sound reasonable?