VMWare compute cluster EVC mode
Home > VMWare platform > VMWare vCenter > VMWare compute cluster EVC mode
It is very important to enable EVC mode on any VMWare compute cluster, even if at present all processors are of same model. By enabling EVC base-line there would be scope of adding more ESXi hosts to the existing cluster without requiring poweroff or migration of all VMs. This can be very useful if vCenter appliance is running on top of same cluster (converged management + VM cluster).
For understanding different types of EVC baselines refer: https://kb.vmware.com/s/article/1003212
Note that for EVC mode cluster must contains hosts with CPUs from only one vendor, either Intel or AMD. Also note that this is an issue till vCenter/vSphere 6.5. Starting vSphere/vCenter 6.7 per-VM EVC mode options are available.
To decide which EVC baseline should be used for a particular cluster use following steps in:
- Login into ESXi host or vcenter and determine processor model
- Search for technical specs for corresponding processor and look for keywords such as: Skywell, Broadwell, Haswell, Ivy Bridge, Sandy Bridge, etc. in its description in case of Intel based processor
- For example Intel Xeon processor E5-2683 belongs to Haswell generation. Refer https://ark.intel.com/products/81055/Intel-Xeon-Processor-E5-2683-v3-35M-Cache-2_00-GHz
- Similarly, Intel Xeon Gold 6130 processor belongs to Skylake generation. Refer https://ark.intel.com/products/120492/Intel-Xeon-Gold-6130-Processor-22M-Cache-2_10-GHz
- Repeat the same exercise for all existing hosts in cluster.
- The EVC mode to be used is the oldest Intel architecture supported among processors in use. Note that even if all the current processors are of Haswell generation, setting EVC mode to Haswell may show compatibility warning.
To configure EVC mode when there are running VMs leading to compatibility warning use: This wont work in case the compute cluster is also configured for VSAN and compute nodes are using VSAN datastore. Steps assume some other NFS/iSCSi/FC based network shared data-store
- In case of converged management + workload cluster where vCenter VM is also running on same cluster refer https://kb.vmware.com/s/article/1013111
- Put one of the hosts in maintenance mode so that VMs vMotion on other hosts.
- Create a new EVC cluster with only this host
- Migrate a few running VMs to this host. If error is shown during vMotion then poweroff of the VM in current cluster and power on of the VM once it is moved to EVC cluster is necessary.
- Once enough VMs are migrated to a single host in new cluster, put another host in older non evc-cluster in maintenance mode.
- Again migrate VMs to EVC cluster.
- Once all VMs are migrated, migrate all remaining hosts to new EVC cluster
- Remove old cluster after ensuring all other configuration such as DRS, etc. same as old cluster is also configured on new cluster
In case of VSAN cluster putting hosts in maintenance mode and creating a new cluster is not possible as VSAN datastore wont move. In this case powering of all VMs (except vCenter) might be necessary. Try your luck by powering of all VMs except vcenter and try enabling EVC mode. If running vCenter also causes compatibility warning then try rebooting vcenter once. After reboot try enabling EVC mode again. If after reboot also EVC mode option is not coming then use following risky steps:
- We need one host in another vcenter cluster for running vcenter (vcsa-live). If another host is not available outside existing VSAN cluster then put any one of existing VSAN nodes into maintenance mode with "Evacuate all data" option. This might take long time. Create a new vcenter cluster and move the host to new cluster
- Clone running vcenter (vcsa-live) appliance as one more VM (vcsa-clone1)
- Migrate both compute and storage of vcenter appliance clone (vcsa-clone1) to another cluster
- Shutdown existing vcenter (vcsa-live) after noting its ESXi host and ESXi host where vcenter clone is migrated. Also these ESXi hosts should have lockdown mode disabled so that they can be managed directly through ESXi console.
- Start vcenter clone (vcsa-clone1) on the new cluster
- Verify that new clone (vcsa-clone1) is working properly. In one scenario it was observed that in vcenter clone even administrator@vsphere.local did not have enough permissions on the cluster to do basic operations. In such cases we need to poweroff clone and abort. We cannot continue further with a non-working clone.
- From the vcenter clone enable EVC mode on cluster. Wait for some time and then create a new clone of this vcenter (vcsa-clone1 to vcsa-clone2) from which EVC mode was enabled.
- Migrate new vcenter clone-clone (vcsa-clone2) to the older cluster
- Poweroff running vcenter (vcsa-clone1) and poweron latest vcenter clone (vcsa-clone2).
- Verify that EVC mode is still enabled and that there no issue with vCenter.
Note that multiple vcenter VM cloning without shutdown as suggested above is risky. Perhaps shutting down vcenter and using ESXi file browser to copy vcenter VM (cloning) might be better.
Refer:
- https://kb.vmware.com/s/article/1013111 (Not applicable for VSAN nodes)
Home > VMWare platform > VMWare vCenter > VMWare compute cluster EVC mode