Question about VPLEX with RP Local and Remote replication

I think I may have answered my own question.  According to the RP and VPLEX best practices document the splitters have to have a preferred cluster in a VPLEX Metro, if that cluster fails the new writes at the other site will not be protected by RP.  So it only makes sense that if a system is moved to the non-preferred site then any write it makes will not be protected either.

Correct...if this is not MetroPoint then only one side of the DD volume will be protected. If that site that is protected by RP suffers a failure, you will not have any automated replication on the surviving leg through RP (usually an RP cluster won't even exist at this site sometimes in this topology). If you happen to have the topology where you have an RP cluster at both sides of a VPLEX Metro for protecting different DD volumes, then in a failure scenario you could reconfigure the losing side to push the surviving leg of the DD volume if needed (full sweep, reconfiguration, etc in the RP config is needed).

HTH

Thanks!

Thom,

As long as the VPLEX/RP infrastructure is fully operational (i.e. both VPLEX clusters are still operating normally), the splitter will pick up writes on both sides of a distributed volume. If a write arrives at the site with the RPAs, then the local VPLEX Cluster will split that write to the RPAs. If a write arrives at the site without the RPAs, then that site's VPLEX Cluster will mirror the write to the other site's VPLEX Cluster, which will in turn split it to the RPAs.

Hope that helps,

- Sean

Thanks Sean.  That makes sense.  So in the event that WAN link is down and the system is writing to the non-preferred leg of the distributed device then those writes would not get split.  Obviously if there's a witness involved and it is on the non-preferred cluster all writes would stop.

Thanks,

Thom

Thom,

If the only failure was a WAN partition, the VPLEX CG at the non-preferred site would be taken offline (both with and without witness)... so the apps would need to switch to writing to the preferred side, where the writes would be split to the local RPAs.

The witness would only keep the CGs at non-preferred site online if it determined that the preferred site was actually down. If you did NOT have witness and experienced that same failure (preferred site goes offline while apps are writing to the non-preferred site), then the non-preferred site would also be taken offline. Having the witness sit in a third failure domain allows it to make more intelligent decisions about when & where to take CGs offline in the event of a potential split brain.

Thanks,

- Sean