Migrating VM from non-EVC cluster to EVC cluster

[anupampanta]

Hello there, 

I understand this question is directly linked and already answered in the post: here, but as this was answered almost 13 years back, is there any way I can do this now??

 

My Setup: 

I have a production cluster of 3 hosts with the same CPU: "Intel(R) Xeon(R) Gold 6148 CPU @ 2.40GHz" with almost 100 running VMs. We need to migrate them to a new cluster having CPU: "Intel(R) Xeon(R) Gold 5317 CPU @ 3.00GHz" where we have enabled EVC mode of "Intel® "Haswell" Generation". As per my research, I am convinced that there are 2 options now: 

1. Power Down the VMs from the Old cluster, and enable EVC on it then perform vmotion
2. Perform offline migration (not applicable at all)

 

I can also see a third option using third-party tools like Veeam to perform a replication. But I found that veeam also performs a "Cold Migration" wherever there is a mismatch in the CPU architecture. 

 

Is there any other approach that may reduce my downtime. 

 

Cheers, 

Anupam

 

[kastlr]

Hi,

 

may I ask why you use an EVC configuration for Haswell while your new cluster is running on Ice Lake CPUs?

Wouldn't it be easier to lift the EVC setting to match Ice Lake?

When I remember correctly lifting the EVC to a higher level could be done online without impacting already running VMs.
The other way round would be an offline event, but as I didn't expect that you plan to expand your new cluster with nodes driven by Haswell your current EVC setting should be adjusted.

If you would lift your EVC Setting to (a least) Skylake you would be able to perform an online migration.

[anupampanta]

Hi  

Really appreciated your reply. 

One thing I did not mention was we have multiple clusters, to match our other cluster we did set EVC to "Haswell". But as you mentioned I lifted the EVC to a higher level i.e. Skylake and tried a live migration. It worked only after enabling 'per-VM EVC' on a test VM. 

Since to enable EVC on 'per-VM' I need to power down and enable it, I will take this as a last resort (still more reliable than Veeam migration). All thanks to you. 

 

But this still does not answer my question. Is there any way I could perform VM migration from Non-EVC Cluster to EVC? 

 

Regards, 

 

 

 

[kastlr]

Hi,

 

in your special case it's only possible to perform an offline migration.

As you started your VM on your Skylake Cluster vMotion wouldn't allow an Online Migration to any cluster/node below that Level, regardless if that cluster is using EVC or not.

What surprises me is that you run into issues even after lifting the EVC settings on your target cluster to Skylake (or Ice Lake).

Which CPU feature(s) did prevent the live migration of the original test VM, as prior using "per-VM EVC" setting it did run on a Skylake architecture and therefor it should also run on an Ice Lake cluster using a Skylake EVC setting.

 

What I've seen in the past that some server vendors did disable CPU features at the BIOS level, so might be a good idea to check this out.
Even if I don't expect that such a configuration is responsible for the symptoms you saw during your tests it still could cause problems with future migrations if source and target nodes (running same EVC mode) might use different CPU BIOS settings.

[a_p_]

Just as an additional information. There was an issue with vMotion from Syklake to newer CPUs due to a CPU feature that was removed by Intel. For details, see https://kb.vmware.com/s/article/76155

André

[anupampanta]

Hi, 

As mentioned by  , memory Protection Extensions (MPX) feature is disabled in my new cluster. I am getting the same MPX feature not available in Intel(R) Xeon(R) Gold 5317 CPU (new cluster).

Seems like my only option now is to enable per-VM EVC on each VM by powering down one by one

 

Anyways, thank you  and  for your effort. Learned a few more things. 

Cheers, 

 

[a_p_]

>>> Seems like my only option now is to enable per-VM EVC on each VM by powering down one by one

Per-VM EVC is not necessarily required, i.e. it depend on your ESXi versions. However a power-cycle is unfortunately required in any case for VMs which still have this feature exposed.

From the KB article: "Starting with ESXi 6.7 P02 and ESXi 7.0 GA, in order to minimize disruptions during future upgrades, VMware will no longer expose MPX by default to VMs at power-on. ..."

André