Moving ESXi host from lower EVC mode cluster to higher EVC mode cluster

[bernz]

Hi

I have two clusters with both EVC enabled and planning to move hosts from old cluster to new cluster.

Old cluster (EVC mode baseline Intel "Merom" Generation (Intel Xeon Core 2) )

6 x IBM x3850 X5

New cluster  with new hardware (EVC mode baseline Intel "Nehalem" Generation  or Intel "Westmere" Generation )

6 x IBM x3850 X5

What approach is recommended, so I can move old hosts from old cluster with lower EVC baseline to new cluster with higher EVC baseline, with ZERO virtual machine downtime?

After moving all old hosts to new cluster, I'm planning to vmotion all VMs to new ESXi hosts and disconnect/remove old hosts from the cluster.

[sneddo]

If you are adding hosts with lower EVC mode to a cluster of higher EVC mode, your VMs will be running at this level until they are powered off and on again (note: a restart operation is not sufficient).

So if you add your 'Merom' hosts to your "Nehalem/Westmere" cluster, you will notice your VMs list "Merom" as their EVC mode- they just won't have access to the additional CPU features of the higher EVC mode.

The only time you have to have outages for migrations is if you are going to a lower EVC mode, as the CPU features the VMs are running are being reduced. Of course the only caveat to this is the power cycle requirement to go up a level, but your VMs will still run.

[JohnsVCP5]

Hi ,

Where is two cluster you mentioned only one cluster (Intel "Merom") - 6 x IBM x3850 X5 , other one is EVC enabling New cluster "Intel "Nehalem"

Whenever you enable the EVC mode it's require downtime because apply the same CPU family on cluster level

Please let me know the CPU speed of 6 servers

[bernz]

Hi John

Old Prod Cluster (EVC Mode Baseline: Intel "Merom")

3 x IBM System x3850 X5 (Intel Xeon E7-4800 Series)

2 x IBM System x3850 X5 (Intel Xeon 75xx Series)

New Cluster (EVC Mode Baseline: Intel "Nehalem" or Westmere")

6 x IBM System x3850 X5 (Intel Xeon E7-4800 Series)

Both clusters are presented with same storage group & network.

My objective is to move all PROD vm running in Old Cluster to New Cluster with ZERO downtime due to DC movement.

[bernz]

Hello Sneddo

Can you verify my steps please:

1. Disconnect host running from Old Cluster (EVC Mode: Merom).

2. Remove host from vCenter Server Inventory.

3. Add host to New Cluster (EVC Mode: Nehalem/Westmere).

4. Enter host maintenance mode (old host) to  vmotion VMs to other host within New Cluster (EVC Mode: Nehalem/Westmere).

ZERO downtime should be expected in this process?

New CPU features will only available to VMs upon power cycle.

thanks.

[JohnsVCP5]

Hi bernz,

Defiantly you require downtime for old EVC mode Merom cluster while adding the hosts to New EVC mode Nehalem/Westmere cluster

because both are different EVC mode

Shutdown all the VMs which is in Merom cluster

Disconnected & Remove the hosts from Merom cluster

add the hosts to Nehalem/Westmere cluster

Power on the VMs

[npadmani]

No downtime, Tested it in my lab. Basically we are trying to go from lower EVC mode to Higher so only VMs from old EVC continue to see old EVC mode and that needs complete power off and on again.

Should be all Ok to go with the steps you have mentioned.

Note: tested it in vSphere 5.5 environment & vSphere 6 environment both, using web client.

VMware KB - 1013111 states following note.

Note: With vCenter Server 4.1 and 5.0, it is possible to add a host with running virtual machines to an EVC cluster as long as none of the virtual machines on the host are running instructions not supported by the EVC baseline. If any virtual machine is running an instruction not supported by the baseline, a warning displays. For more information, see CPU Compatibility and EVC in the vSphere Datacenter Administration Guide for 4.1 and vCenter Server and Host Management Guide‌ for 5.0.

[JohnsVCP5]

Hi Npadmani,

Downtime require for VMs because old cluster EVC and New Cluster EVC mode are different CPU family it will not reflected from old CPU family to new CPU, if the VMs are migrate to new cluster without downtime.

[Madmax01]

Hi theire,

i don't think that theire is a downtime necessary.  i have   in my environment  for each Architecture an own Cluster to provide all CPU Features correctly.

Westmere,Sandy,Ivy,Haswell

And i'am able to migrate from Westmere Cluster to each of the Higher CPU Clusters.

i'am only not able to migrate Back a Haswell VM online to Westmere once it's having the CPUID.

But from Lower to Higher Architecture it should work everytime.

So just try to migrate online from old cluster to new Cluster and see if you get error or not.   after you cleared first Host from Old cluster, you could disconnect and connect to the higher CPU Architecture.

in the Past i everytime was succesfully with that. 

Best regards

Max

[bernz]

That might be a problem, since the old cluster are all production VM and does not require downtime.

Any other suggestion please.

[JohnsVCP5]

Hi bernzbernz
Please get the downtime the from bussiness and plan it accordingly.

Shutdown all the VMs from old Cluster

Disconnect & remove the hosts from old cluster

Add the hosts to new clusters

Power on Old Cluster VMs

[sneddo]

Hello Sneddo

Can you verify my steps please:

1. Disconnect host running from Old Cluster (EVC Mode: Merom).

2. Remove host from vCenter Server Inventory.

3. Add host to New Cluster (EVC Mode: Nehalem/Westmere).

4. Enter host maintenance mode (old host) to  vmotion VMs to other host within New Cluster (EVC Mode: Nehalem/Westmere).

ZERO downtime should be expected in this process?

New CPU features will only available to VMs upon power cycle.

thanks.

Sorry for the delay, been offline for the weekend.

You don't need to do these steps since you have shared storage accross both clusters, you can just vmotion to the new cluster

@JohnsVCP5 Do some research, you DO NOT require downtime for this operation.Only if you are migrating 'down' a level of EVC do you require downtime.

[JohnsVCP5]

Am Waiting for the outcome and will see which one is right

[sneddo]

Well, you have three people in this thread saying the same thing...and I've personally done this several times. And the doco agrees

If virtual machines are running on hosts that have feature sets greater than the EVC mode you intend to enable, ensure that the cluster has no powered-on virtual machines.

Enable EVC on an Existing Cluster

[bernz]

Sounds great sneddo...

I might save some time here...

I will let you know guys once the activity took place.

[sneddo]

What you could do if you are worried about downtime is spin up a test VM and test the migration using that, before doing your production VMs. Generally a good idea anyway, to make sure you don't have misnamed port groups or anything else that might cause you pain when you migrate.