Jaa


Host-based Backup of CSV Volumes

Happy New Year!

I’ve received several questions and comment about backing up VMs that reside on CSV volumes.  In short, you can do it, but it’s tricky.

There’s a great post on the DPM Insider blog about the challenges (Here).  Host-based backup of CSV-base VMs works with DPM 2010 and other tools.   The trick is to tie the CSV volume to the physical cluster node where a VM is executing for the duration of a VSS snapshot and backup.  As noted in Asim’s post, using a hardware provider makes the most sense (saves time!).  I’ve already heard moans about the suggestion of using a hardware VSS provider.

What strikes me about all the questions is how greedy we’ve gotten with our expectations around the benefits of virtualization!  If you’re using CSV (and Live Migration or even VMotion for that matter) you have already invested in shared storage.  What’s the big deal with using a hardware provider?  Yes it may be more expense, but if you are serious about host-base backup of highly available VMs you need to do it.  Yes, I’m being a hypocrite here – my primary storage subsystem does not support hardware integrated snapshots!  I know they aren’t always an option.  You can still use software snapshots (with System Center DPM 2010), just be aware that it will take longer to complete, and that IO will be pinned to the cluster node being backed up for the duration of the backup.  Remember that you don’t need CSV for Live Migration, it just makes storage provisioning easier.  You can setup each VM on an individual LUN and not have a CSV backup challenge…it’s a trade off.

Don’t forget that you can still backup using a process or agent IN the VM!   It may be somewhat old school, but nothing should stop you from deploying an agent inside your critical VMs and executing a backup.  Depending on your application workload, this may be the best solution REGUARDLESS of your use of CSV / Live Migration. 

What if, for example your VM is running a workload that just doesn’t work right with VSS?  The backup coordination from the host (via the “Backup” Integration Services) is meaningless and must be disabled.  If it is disabled, then the VM would be “saved” during a host-based backup – interrupting processing.  The best alternative in a situation like this would be to use an application aware backup agent inside the VM.

-John