I normally setup larger clusters than a 2-node cluster with shared storage, but we have built a cluster specifically for hardware based GPU's for a few VMs and thus new hardware and just a 2-node cluster with backend shared SAN storage.

The 2 hosts have 384GB RAM and dual 10 core procs.

Total VM count = 12 with total vCPU assigned equaling 36 Total VM cpu utilization is around 5% on each, and 35% ram on each.

My question is:

Is there a preference or best practice on the host admission policy? Right now I have it set to "cluster tolerates 1 host failure" but typically I would set cpu/ram percentages with larger clusters. However, when I do this regardless of setting (25/25, 15/15, 50/50) I end up with VM's showing "unprotected". Setting it back to a single host failure sets them all back to "protected".

Are there certain settings for HA and DRS to take into consideration when only dealing with a 2 host cluster?

The cluster is currently running on 5.1 but I don't think the options have changed since.


See: Should VMware HA Admission Control be enabled for small (2/3-host) clusters?

Don't enable admission control for a 2-host environment. Unless you're dealing with well-calculated or custom slot sizes or VMs or resource pools with reservations, there's no benefit to doing this.

  • Thank you. I don't know how I missed that one in the search, but after reading your discussion there it makes sense. I'll go with that! – TheCleaner Jan 25 '16 at 19:00
  • I think VMware provides bad information about this stuff. The defaults are terrible and understanding resource math is tough. – ewwhite Jan 25 '16 at 19:01
  • @TheCleaner: You can view the "calculated" slot size by clicking Advanced Runtime Info on the cluster summary page. As an aside, Hyper-V uses only allocated RAM as the slot size determinant. – joeqwerty Jan 25 '16 at 21:07

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.