Ideal DRS Setup with Different models of Servers

[BrianRTS]

Currently I have a mix of Dell Servers in our environment between r610's, 1950's, 2950's and r200's. I had one DRS cluster with all the servers in it which seemed to work well however I ran into cases where DRS shifted a VM off the r610's onto a 1950 and I wasnt able to vMotion it back because of CPUid mismatches between the two. In order to work around that, I think the best way is to create DRS clusters for each server type and let the VM's use vMotion/DRS within their like server type... the downside to that is I feel I am losing flexabilty because i now have to manage which DRS cluster gets the VM load.

Has anyone seen a problem like this and how have they faced it?

All servers are on vSphere 4.0 Update 1 or 2, soon to be upgraded/migrated to ESX4i 4.1 once my VC is upgraded.

Thanks for any help you can provide.

[RParker]

Has anyone seen a problem like this and how have they faced it?

Yes, but proper planning of ESX would have avoided this. ESX pretty much needs each of the servers to be identical. SOME servers may have chipsets which are compatible, but those servers are all basically different. The CPU is the most prevalent factor, but the reason why it's so important they be compatible, is because during vmotion the memory contents are transferred to a running ESX host, and the VM stays running. So a difference in CPU or chipset could cause the OS to glitch, because it could be in the middle of CPU operations, and all of a sudden features which aren't there will cause the VM to crash (in some cases).

So it's best to have the ESX host the same. Then in vCenter there is an EVC feature which will let you know (if you created the datacenter with this feature turned on) which hosts are compatible BEFORE they get added to the cluster.

So in the future, you should use the same hardware for ALL ESX deployments, especially if they are going to be in the same vCenter / Cluster. This way you can ensure compatibility throughout.

Now you have a problem, and only testing will let you know if there is a problem. If there are features on the R610 which aren't available on 1950, you may be able to move one way (because you can add those features, but once they are there, you can't remove them). You can always shut down a VM and force a restart on another host, but this get's complicated, because DRS is designed to be transparent. That's why it's best to keep the hardware the same.

The same goes with ESX hosts with different amounts of RAM, you will lost ESX hosts with higher RAM, because CPU is priority over memory when it comes to DRS. So basically those ESX hosts with more RAM, that "extra" RAM will never get used.. unless you force more VM's on the host, but that defeats having DRS.

For your case, I would use partial DRS, and let vCenter where it needs to place one when they power on, and later you can figure out which hosts will not let you migrate by testing with vmotion. You have a hodgepodge of servers, segregating them is the best option, but if you have only 1 or 2 or each, I see where you have a problem with resource, but in the future you should try to plan it out a little better so you know in advance what you need.