Do server hosts in a FT or HA scenario need to be identical?

[LePivert]

I have an IBM x3350 server running ESXi 4.1

I would like to practice FT and HA.

Yes, I'll have to set up a number of other things as well (shared storage, etc.), but I'd like to concentrate on the subject line for the time being.

So if I add another server to the first, for practicing FT or HA operations, does it need to be identical?

The other server would be an IBM x3250. Similar... but not identical.

Yes, both would be running ESXi 4.1.

Thank you in advance!

[weinstein5]

HA has more flexibility in the servers since it is a restart of the VM - in other words you can have an AMD and Intel server in an HA cluster - vmotion would noy work between the two of them -

For FT the requirements are much tight in regards to CPU compaability in that I believe they need to be identical -

If you find this or any other answer useful please consider awarding points - 

[vmroyale]

Hello.

There is some compatibility information about FT in kb 1008027.

Good Luck!

[bulletprooffool]

HA requires identical instruction sets, but this does not necesaroily mean that the hosting ESX servers need to be 100% identical.

You could for example use EVC mode to get 2 non-identical servers working in the same Cluster (and thus both servicing HA requests)

EVC basically masks portions of the CPU instruction set that are not available on both hosts.

http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=100576...

Unfortunately however, once you enable EVC, you're probably going to lose the ability to FT (not all servers can handle FT just yet and I am not aware of any instances where they retain FT after EVC has been enabled)

so in short - HA, you can use EVC, FT - probably not

[weinstein5]

I think you meant FT requires identical instruction sets - HA does not

[bulletprooffool]

I think you meant FT requires identical instruction sets - HA does not

Correct - my mistake. (long day)

DRS requires EVC to reduce the CPUs sets on ESX hosts to match (by masking), so yes, the CPUs do not need to be identical, but the instruction sets that are presented by the ESX host does.

And yes, FT requires instruction sets (for now)

HA does not require consistent CPU to the same level as it does not use vMotion. the VM is effectively restarted on a new host from scratch, so the CPU match requirement is no longer there,

but thanks weinstein5 - perhaps I could have clarified that a little more carefully and stated it a little less ambiguously - worth making the point to avoid confusion for future readers

[weinstein5]

EVC (Enhanced Vmotion Compatability) as I understand comes into play for DRS not HA - allowing vmotion to occur across a broader set of CPUs - since HA does not use vmotion EVC will not be a factor.

Sent from my iPhone

[LePivert]

Thanks to all. It looks like I'd do best to have identical material then.

All posts were useful - not sure if I can grant more than 2 (?) useful comment points?

[depping]

EVC is not a requirement for HA as VMs are restarted and thus CPU capabilities are irrelevant.

Duncan (VCDX)

Available now on Amazon: vSphere 4.1 HA and DRS technical deepdive

[DSeaman]

And we have EVC enabled on our clusters and FT works just fine, so they are not mutually exclusive in my experience.