Enable EVC without downtime

[RemcoD]

I need some help with following situation.

We have an ESX 5.5 cluster (no EVC configured) with several hosts with Intel Xeon E5-2680 v2 cpu's. We wanted to extend the cluster with 4 new hosts. The problem is that the new hosts got Intel Xeon E5-2680 v3 cpu's which are the newer version of the same cpu. I've found several topics where people mention that it isn't a good idea to mix v2 and v3 in 1 cluster without enabling EVC. I want to enable EVC but of course without rebooting the active vm's. I've created a new cluster with EVC enabled (Sandy bridge EVC mode), added the 4 new hosts no problem so far. But how can I add the other hosts where active vm's are running to this new EVC enabled cluster without rebooting those vm's. Is this even possible? I've tried disconnecting the host in vCenter moving it to the EVC enabled cluster and connecting it again but even that failed.

Anyone who can give me some advice?

Thanks in advance.

[narenotpp]

since you have your existing hosts with v2 CPU and new hosts with v3 CPU.

Now do a little test, try to vMotion migrate your one of the existing VM from old cluster to new cluster that you have created.

1) if your VM migration works, than do this,

Empty one of the host from your old cluster by doing vMotion migration of all the VMs from old cluster to new one, Enter that host in maintenance mode and move it from old to new cluster and exit the maintenance mode.

carry on doing this for rest of the VMs and hosts, that way you don't have to face any downtime.

2) But if your vMotion migration of VM from old cluster to new fails. then you are left with only one option, suspend or power off your VMs on existing cluster host, put the host in maintancence mode and move it from old cluster to new. in this case there's downtime associated with it.

[RemcoD]

I've done some test prior to enabling EVC on the new clusters and I could do a vmotion without issues between the different hosts. The reason why we changed to an EVC enabled cluster is because we saw topics of people mentioning performance issues on guests when they moved it from a host with a v2 cpu to a host with v3 cpu or back. I hesitated to enable EVC as I could vmotion the vm guests between the hosts without errors. The problem is that I couldn't proof if there are any performance issues when moving the vm guests between the old and new hosts.

So yes option 1 was possible but wasn't convinced due to the performance issue.

Enabled EVC on the existing cluster with the v2 cpu's isn't possible due to active vm's.

[sneddo]

Enabled EVC on the existing cluster with the v2 cpu's isn't possible due to active vm's.

You're on 5.5, right? Then you can do it live as long as you are setting it to the matching level for your CPU. The doco is a bit confusing on this point, but technically supported since either 4.1 or 5.0 from memory.

Power off all virtual machines in the cluster that are running on hosts with a feature set greater than the EVC mode that you intend to enable, or migrate out of the cluster.

Source: Enable EVC on an Existing Cluster in the vSphere Web Client

Since you are effectively enabling EVC at the same level as the VMs are currently running, you are able to do this. I don't have a test cluster to do this on, but I'm sure I've done this in the past. Similarly, you can raise the EVC mode without a reboot, but the VMs will not use the additional features until the are power cycled.

[RemcoD]

Ok problem fixed.

You can indeed enabled EVC on a cluster if you select the mode that is the same as your current cpu's or higher, i knew that.

I've used the vsphere windows client (I know, should use webclient) and I could only select Sandy Bridge mode as highest mode  which is lower then the currect v2 cpu's. So that's the reason why I couldn't enable EVC on the existing cluster. If I enabled EVC via the webclient I could select Ivy Bridge generation and that's the same as the current cpu's in our cluster with active vm's. And knowing that I realized that Sandy Bridge wasn't the correct mode for the cpu's we got... So I enabled EVC with Ivy bridge generation on our cluster and now I can add the new hosts to it.

Thanks for all the help guys but this shows again that I really really really need to learn to use the webclient more :smileyhappy:

[sneddo]

Do it the other way around- enable EVC on the existing hosts (which you can do without any reboot), then add the new hosts. As long as you have no powered on VMs on the new hosts which are using CPU features above your EVC baseline, you will be fine.

[mtremblay]

I had the same issue with one of my hosts when i moved them from a 5.0 vCenter to a 5.5.

I did the following with no downtime and no VM's rebooted. Life is good!:smileyhappy:

1- Disconnect each host

     a. Right-Click host disconnect

2- Remove each host form vCenter

     a. Right-Click host remove

3- Enable EVC for the cluster

     a. Right-Click Cluster and select Edit Settings

     b. Select "Chanage EVC"

     c. Select "Enable EVC for Intel Hosts -> Sandy Bridge". Mine was an Intel Sandy Bridge generation so yours may be different.

4- Reattach the hosts

Boom!

Cheers,
Marc

[ozarkscokeitman]

So were you moving your EVC up or down from your processor capability when you did this?  I have a situation where we have a Cisco Hyperflex cluster (6.5) that we are trying to get VMs from a UCS/VNX cluster (5.5) migrated to.   The Hyperflex cluster has existing production VMs running on 3 of the 4 nodes and of course EVC is currently disabled.   This makes it impossible to migrate VMs from the UCS/VNX cluster with enabled EVC to the Hyperflex.   The EVC mode of the UCS/VNX cluster is Intel Westmeyer but can go all the way to Intel Ivy Bridge.   Could I use your procedure to enable EVC on the hyperflex and specify Ivy Bridge after moving the UCS/VNX cluster up to Ivy Bridge?

[mtremblay]

I was moving 1 of my hosts down to the processor capabilities of my other 2 hosts.

You should be able to disconnect the hosts, make the EVC change, then reattach after with no problems. What I did was attach the host then try to vMotion over a VM, it then told me it needed to be at Ivy Bridge. I then disconnected all the hosts from vCenter, made the change, tjhen reattached.

You should be fine but I obviously can't say 100% so do at your own risk. I can say I've disconnected hosts from various other companies with no issues at all as this does not shutdown any VM's residing on the hosts just the connection to vCenter and its features.

Cheers

[ozarkscokeitman]

Thanks so much for the quick response...  I'm going to give it a whirl this weekend...  Does it matter that it's the whole cluster that I need to move down?   I will have to remove ALL of the hosts from this cluster..  enable EVC at the level of the other cluster (after I raise it to the max) then re-add the hosts to the now EVC enabled cluster.  Man I hope this works.   Getting the downtime for all the VMs will be difficult. 

[mtremblay]

Nah should be fine. You are really just lowering down the EVC to the lowest supported CPU type and disconnecting/Reconnecting the hosts from the vCenter server so it should have no impact. Make sure you remember the hypervisor passwords and I always disconnected any mounted ISO's from my VM's :smileyhappy:

Best of luck

[cgcit]

Did it work for you?