Write cache replication to protect against data loss in case of host failure
Write cache replication (mirroring) is only applicable to VirtuCache’s Write-Back caching policy (caches reads and writes). It doesn’t apply to VirtuCache’s Write-Through caching policy (caches only reads).
Using the VirtuCache GUI, the administrator can assign Write-Back caching policy at the Datastore or VM level. There are 3 Write-Back caching policies in VirtuCache. ‘Write-Back with No Replica’, ‘Write-Back with 1 Replica’, and ‘Write-Back with 2 Replicas’. These policies specify the number (0, 1, or 2) of copies of write cache that VirtuCache will keep on separate hosts in the ESXi cluster for the local write cache of each VM assigned the write-back caching policy. We call this a ‘write replica’. The number (0, 1, or 2) of write replicas indicates the maximum number of node failures that can be sustained before there is data loss in the cluster. If an administrator chooses to keep, say, one write replica, VirtuCache automatically replicates the writes to cache media on one additional VMware host in the same cluster. In the event of a host failure, VirtuCache syncs the backend storage appliance with write replicas (of all VMs on the failed host) that are kept on other hosts. In this way, VirtuCache protects against data loss in case of host failures. Reads are not replicated since the backend storage appliance is always in-sync as far as reads go.
Setting up the VirtuCache Replication Network
- Before you start using ‘Write-Back with 1 Replica’, and ‘Write-Back with 2 Replicas’ policies, you need to select a VMware network for VirtuCache write replication. You do this by using the Hosts tab in VirtuCache GUI. By default, this network is the vMotion network, but you can select a different VMware network.
- Since VM writes will be transmitted over this network, it should be configured like a vmkernel storage network. In other words, you must use the ‘Default’ TCP/IP stack and not the ‘vMotion’ TCP/IP stack on the vmkernel adapters (this is a requirement). For low throughput environments (< 10MBps per host), 10gbps network and Jumbo Frames are preferred, but not required. For high throughput environments, both 10gps and Jumbo Frames are required, and in some cases of very high or random throughput, LACP or LLDP protocols might be needed as well. VirtuCache charts will let you know if you need 10gbps, Jumbo Frames, LACP, and LLDP.
NB 1: ‘Write-Back 2 Replicas’ policy should be used only if you have 20gbps networks AND for VMware clusters of more than 16 hosts.
NB 2: ‘Write-Back No Replica’ policy is usually not recommended since it will result in data loss if a host fails. However, it is popular where data loss can be tolerated like non-persistent VDI, Terminal Server / RDS VMs, and Dev/Ops VMs.