Peer Persistence utilises HP 3PAR StoreServ Remote Copy to provide automatic, transparent failover of storage from one HP 3PAR StoreServ storage array to another. Lee Morris, a Senior Storage Consultant within OCSL’s Infrastructure team, shares his views on the HP 3PAR StorServ.
At OCSL we work with a huge amount of customers who are making the most of HP 3PAR technology, regardless of the market sector they are in.
Peer Persistence is an excellent way of minimising downtime in an ESX environment and another great addition to the HP 3PAR StoreServ advanced software portfolio. We have implemented Peer Persistence with vMSC uniform configuration many times and it works really well. The diagram shows a typical Uniform vSphere Metro Storage Cluster.
The way it works is quite simple. When adding the source Remote Copy volumes (to be used as VMFS datastores) on each array to a Remote Copy group, the corresponding destination Remote Copy volume is configured with the same WWN. Then both the source and destination volumes are presented to all ESX hosts in the cluster with the same LUN number.
The ESX hosts then see the paths to the source volumes as active and the paths to destination volumes as standby. In this scenario replication can be switched over on the arrays so the source volumes that were on Array 1 become destination volumes and vice versa.
All that happens on the ESX side is the active and standby paths switch over with no disruption at all to VMs.
The third part of the puzzle is the Quorum Witness, a Linux VM (in OVF format) that needs to be installed on an ESX host in a third site or location. The Quorum Witness acts as an arbitrator in case of a failure of the remote copy links and /or arrays. The QW allows a
Peer Persistence configuration to handle different failure scenarios without affecting VMs. It must be able to communicate over TCP/IP to both arrays management ports with a RTT of < 150ms.
During post installation testing, we have replicated all failure scenarios achieving the expected results and always recovered from the enforced failures leaving the arrays in their normal state.