VMs not in Write-Back mode because of Replication Network issues
Caching policy of ‘Write-Back 1 Replica’ is assigned to VMs or Datastores, but the operational state remains ‘Write-Through’. The reasons could be one or more listed below.
- The VirtuCache replication network is the vmotion network and the vmotion network might be using the vmotion stack that precludes all other traffic from running on it.
- If a separate write replication network has been created for VirtuCache, then make sure it has an MTU set to 9000. Its highly recommended that we use Jumbo Frames end to end, on the vSwitch, vmknic, and physical switch ports. All three need to be on MTU 9000.
- This is a peculiar situation – if a separate network is created for VirtuCache replication and it is under the same vSwitch that runs other networks, then make sure that the VirtuCache replication network has the same MTU size as the vSwitch. Setting the MTU size on the VirtuCache network to 9000 when the MTU size on the vSwitch is different, increases latencies on the VirtuCache replication network considerably.
- If possible turn on LACP and LLDP on the VirtuCache replication network. For some high throughput workloads, and this is rare, not having LACP and/or LLDP increases latencies on the VirtuCache replication network.
- Ensure that all hosts have connectivity to the Datastore(s) that hosts the VMs that are assigned the ‘Write-Back 1 Replica’ policy. Say you have a 3-host cluster and 2 of the hosts don’t have connectivity to the Datastore(s) that hosts the VMs that are assigned the ‘Write-Back 1 Replica’ policy, then the VMs will be in Write-Through operational state because VirtuCache will prevent write replicas from being created on hosts that don’t have connectivity to these Datastores since VirtuCache won’t be able to sync the backend storage if there is no connectivity from the hosts to the storage.