Situations where writes might not be accelerated even with Write Back policy

Reason 1 : If SAN is slow and the write throughput is consistently high, then the cache media might not be able to flush the writes to SAN fast enough, resulting in writes not being accelerated.

We keep a maximum of 2 minutes of writes in the cache. We do this so that if a host fails, we can quickly sync the backend SAN appliance with write replicas (that are kept on other hosts) of all the VMs from the failed host. If we didn’t have the 2 minute cap, then the process of syncing write replicas to the backend SAN appliance might choke the network and take too long, preventing the VMs on the failed host to restart immediately on other hosts as would happen with VMware HA. The 2 minutes are measured in terms of your SAN speed. So if the SAN does 100MBps, the 2 minute works out to 1.2GB. Now if you have sustained high write throughput, and if your SAN is slow, then the writes are not getting flushed to the SAN appliance fast enough, and the 2 minute cap is always full, effectively causing the write speeds to the cache media to be the same as the SAN.

Reason 2: High network latencies

Write replication happens over your IP network, so if your network is slow (say 1gbps), then write replication will show high latencies and so writes might not be accelerated.