Backup performance via SAN transport - defragmentation

[Moltron83]

I am using Backup Exec 2010 R3 to backup my VMs using it's VMware agent and using SAN transport as opposed to network level backups. I've noticed that some of my VMs backed up nice and fast.  Nearly 4 GB/minute, while others limp along at 1GB/minute or less.

I was able to find a VM that backed up fast and one that was slow which used the same RAID groups on my SAN and about the nearly the same datastore.  After I migrated the disk of the slower server to the datastore where the backups ran fast for the other server, I noticed backup speeds similar to the fast server in this comarison.  So, then I moved the disk back to the origional location and it still backs up fast now.

On both servers their disks are thin disks, and are roughly the same size.  I'm really just trying to figure out what happened here.  Did I do some sort of defrag by migrating a thin disk to another datastore?  Do thin disks fragment and degrade over time within the VMFS due to the nature of how they expand?

If anyone knows of a good explaination or has experienced this themselves, I'm all ears.  In the mean time, I'm going to try this on a few other slow to backup servers and see if I can get the same result!

-Thanks

[chriswahl]

One thing that comes to mind is the use of CBT (changed block tracking). If the VMs that back up slower (1GB/min) have more changes occuring, this may account for the slower backup speed. For example, I had a pair of servers that had about 100GB of thin provisioned disk usage each. The active server would backup at 300-400MB per minute, whereas the relatively untouched server could push 1-2GB per minute using Veeam's CBT.

As far as fragmentation, the thin disk simply expands itself by the size of your storage block level (1,2,4,8MB) as space is required. I suppose fragmentation would depend on how your storage is configured (such as thin pools of disk). Outside of that, the guest fragmentation should not make a difference if your data is backed up at the block level (fragmentation affecting backups is a file level issue).

[Moltron83]

I tried this again on another system that backed up kinda slowly.  went from 1.3GB-2GB per minute to 2.6GB per minute.  This was a VM with a small disk.

I did some reading here:  http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=100681... about fragmentation, and it basically says not to worry about it.  Additionally in a performance whitepaper this was addressed and said there was no noticible performance impact between fragmented VMFS and VMDK  thin or thick.  They may have been taking about doing writes, not reads.

This again seems odd to me.  I think I'll continue migrating servers back and forth and viewing backup performance to see if this trend continues.