I would like to know what success / recommendations people have had with different backup strategies for vCOps.
Problem: In a large environment your vCOps VM gets very big (>2TB of virtual disk required). In addition, if you use the "add a virtual disk as you need it" strategy, then you may have added 3 or 4 virtual disks to your vCOps Analytics VM since it was first created. Then you end up with say 4 virtual disks 600GB each. Then your backup software goes and takes a snapshot of your Analytics VM attempting to back it up. Each virtual disk then has a snapshot disk created for it. Then your backup software has to remove the snapshot when complete.
We've seen our backup software (Avamar) not handle this very well and the situation not work due to the large size of the VM, multiple virtual disks, and the need to re-consolidate the snapshots for these big virtual disks afterward.
I'm wondering what backup solutions work for you for vCOps given the following considerations / questions:
- Ideally image-level and not in-guest backup is preferable to facilitate automation and manageability
- If more virtual disks create issues, should we be creating fewer, bigger disks at at the outset to reduce complications.
- Which scenario is more error-prone, a snapshot of a huge vmdk and reconsolidation afterward, or a snapshot of a large number of smaller disks, with each associated delta disk needing to be reconsolidated into its parent disk?
- How much overhead should be reserved in the datastore of the analytics VM to faciliate image level backup with it's process of taking and consolidating snapshots?
- Since the Analytics VM can take a whole 2 TB datastore to itself, array-level snapshots could be done on the LUN. However, there is no utility in vCOps to insure array-consistent snapshots so that the vCOps database has flushed its info to disk and the VM data is consistent with the array-level snapshot. (For example, a tool like NetApp SnapManager for Exchange) What impact / problems lay in restoring vCOps by restoring the LUN from an array-based snapshot? Has anyone made this work?
Thanks for your input