HA by datastore mirror

[Linog]

Hi! I have a group of some ESX hosts and two SAN. All ESX locate on one SAN storage. Is it possible to do mirror between two SAN, so if one of them fail, ESX continue to work on other SAN WITHOUT shutdown? I know that Mirrorview (EMC) or Continue Access (HP) just do a snapshots of luns, but if one SAN fail, I have to represent mirroring luns by hands, and of cource, ESX and all VM stopped. Does anybody have a best practict to make a HA with SAN?

[MR-T]

You should take a look at Site Recovery Manager (SRM). Although it's not an automatic failover, you'll be able to switch your environment from one SAN to another in the event of disaster.

I should point out, this process will involve all machines being switched off during a failover.

Unless you're implementing some kind of long-distance cluster, you'll always have this problem.

[Linog]

Ok, I have a host-based replication of my SAN to another site. What another application should I have to restore ESX and all VM on backup site automaticly? Does SRM intefrated with host-based replication (MirrorView, Continue Access, etc)?

[MR-T]

The storage vendor (in your case EMC) provides a plug-in known as a storage replication adapter (SRA). This allows the SRM server to talk directly with the storage and control failover.

It's a new product, so there isn't a great deal of information knocking about at the moment, but I do know there's an SRA available for MirrorView.

The thing is, this is not an automatic failover tool. It requires someone to make a decision and press a button to start the process. It's not normal to have automatic failover for DR purposes, this should always be a human decision as it's very involved and requires a total outage whilst switching across.

[Linog]

I shold have:

1) Snapshot/clone of SAN on array-based level (Does HP (EVA) supported by SRM??)

2) SRM

Is it correct?

[MR-T · Accepted solution]

HP haven't released their agent for SRM just yet. I expect it will be available soon.

To use SRM, you need a VirtualCenter at either site, a few ESX hosts at either site. An SRM server at each site (although this software can be loaded on the VC).

You install the storage agent on the SRM server so communication is available to the SAN and then run through a number of setup wizards.

Once SRM is properly configured, you can simulate failover to alternative sites without actually powering off any virtual machines. This allows you to test the recovey plan. Imagine youi've got multiple LUNS and hundreds of VM's and network. It would be nice to see this all switch from one site to another without actually taking down the production environment. It achieves this by using a snapshot of the production LUN and presenting this to the DR site. The VM's are then powered up, but only connected to a closed-off 'bubble' network. You can then logon to these VM's and ensure any changes you've requested have actually taken place (such as IP addresses).