I was just browsing through our documentation and stumbled on something which has got some cool potential.
This describes how to setup the vSphere side of things for a mirrored disk within a Windows 2008 Guest. Just imagine doing in-guest mirroring of your data while doing FT on the outside. This means you would be able to span multiple sites without the need for a replication mechanism.
I asked around and unfortunately this scenario is not supported today for vSphere FT, but it definitely has potential… Another way to solve this problem would be if we could somehow leverage the Mirror Mode driver that is used by Storage vMotion today. Once again, this is not available today and I don’t even know if people are working on it… just something that popped up and something that has great potential and seems like a small step.
I wondered how something like this could be used with a ZFS storage pool running inside a guest? Out-of-the box, ZFS is not able to be easily set up into a failover cluster configuration. If we could utilize FT and/or the mirroring driver, this could open up a whole new world of possibilities for file servers in vSphere…
There’s a nice post about working with those two configuration parameters by Erik Zandboer: http://www.vmdamentals.com/?p=919
While this is ‘neat’ and could be useful for some situations, I think a much more powerful idea would be to have ESXi be able to mirror VMFS datastores. This way, you don’t have individual VM configuration, but could then get ‘replication’ with lower tiers of disk and auto-transparent failover (no SRM or array based replication needed). Also, SAN replacement could be much easier not having to manage many VMotions. Just break the Datastore Mirror, and resync with the new array.
The Virtual Storage Appliance from VMware available in vSphere 5 does basically mirror VMFS datastores (well, NFS) across the local disks of up to 3 vSphere 5 hosts.
In general this is a cool idea and I think the only logical and cost effective place to do this would be for a FT VM or a DR scenario if you are using IP storage and cannot afford replication software. At the very least in a DR you’d have a disk that has consistent data up to the point of failure.
Interesting but you still need to replicate vm files like vmx too