Enabling EVC and combining two loaded clusters?

[nyjz1298]

Background:  vCenter 4.1 and ESX 4.0 (no update for all hosts).

Cluster 1 (EVC Not enabled) 2 hosts with E7320 Intel procs - 50% memory usage

Cluster 2 (EVC Not enabled) 5 hosts with E5540 Intel procs - 25% memory usage

I'd like to eliminate Cluster 1 by vMotioning all VMs to Cluster 2 causing no downtime.  Not even a reboot / reregister.  Can it be done with all the VMs running?

Thanks!

Joe

[Troy_Clavell]

in theory yes.  However, when you attempt to vMotion for a non EVC Cluster to an EVC cluster, there is a chance the guests may not live migrate.  If this is the case, you will have to shutdown the guest(s) and do a cold migration.

[nyjz1298]

I know I can enable EVC on both clusters using "Intel Xeon Core 2" but from my understanding it won't really matter.  EVC is adaptive and will use the lowest common denominator out of all hosts in the EVC cluster when deciding what attributes of the CPU's to mask.  If this is true, both EVC clusters would be different and result in the same error.  This is a financial company and it's Production farm so I can't really expariment.  I can't get approval to test this because our seperate test environment doesn't have mismatched hardware.

[Troy_Clavell]

I somewhat misunderstood your question.  You want to dissolve one Cluster, and make a new EVC Cluster, correct.  So, for a period of time you'll have 3 clusters.  The 3rd being EVC enabled.

Same thing applies.  You cannot move an ESX(i) Host  into an EVC cluster with running VM's.  You will have to vMotion around your existing guests, free up a host, disconnect from the Non EVC Cluster, and then drag the host to the new EVC Cluster and connect.

Below is a good how to, even if you don't have vCenter running as a VM.  However, as stated in step 20, you  may have to incur some downtime of your guests.

http://kb.vmware.com/kb/1013111

[DSTAVERT]

How many hosts in each cluster?

[nyjz1298]

2 in cluster 1 and 5 in cluster 2.  I'd like to trash the hosts in Cluster 1 since they're old and non-blades and vmotion all the VM's to the newer hosts in cluster 2.  The simplest way to do this is to just shutdown the guests and turn them on, on the other hosts.  I'd rather avoid that because there are 30+ VM's and this is a financial company who doesn't like any downtime.  If it's not *possible* then I can look into getting downtime.  I remember with some previous experence, I had to create a new cluster because I was unable to enable EVC on a cluster with running VM's, but now with vSphere 4 it looks like I'll be able to do it.  From my understanding, the evc masking is placed on the VM's at boot time.  Since they're all running I don't know if EVC will help at all.

[DSTAVERT]

I can't verify that a guest running on a lower capable CPU would migrate better to a higher capable than a higher to lower and perhaps someone could comment. If migration went smoothly from the 73xx hosts to the 55xx hosts then it could almost be a cheap upgrade depending on how costly any downtime was. You could try at least a simple dummy guest OS in both directions. This might be the ideal time to push for upgrading to new hardware. The question will be do you move to 56XX processors or 55xx.

[nyjz1298]

I've tried doing a vmotion just to see if it would validate... and it wouldn't.  I got the usual incompatible message.  Going from 55 to 56 will be easy.  I can enable evc then add the 56 hosts... No problems there as there wouldn't be VMs already on the 56xx hosts.

[DSTAVERT]

So if you need to have downtime then getting everything done and prepared for moving on could still be a good deal.

[nyjz1298]

Come on now... I'm looking for a way to NOT have down time.  I thought someone would have a clear answer on this but it seems like we're all guessing.  I'm also having a hard time finding exact details from vmwares site.

[Troy_Clavell]

The clear answer is defined in the KB article posted.  As it stands, based on your testing, you need downtime.

[DSTAVERT]

Depending on the number of guests on the 73xx based hosts it may take less downtime with 2 replacement hosts that could be added to your 55xx cluster. ????

[idle-jam]

there is the closest you can get for no down time. http://malaysiavm.com/blog/how-to-enable-evc-on-esx-35/