I'm wondering if anyone out there has any tips on tuning VSS paramaters on winblows clients? We aren't getting the best performance from our new Exchange 2010 environment (all netbackup 7), however, non-VSS file backups from the same exchange servers produce 90meg/sec backups. I appears that as soon as we start using VSS for either exchange or file level backups, we get significantly less performance from those machines.
I guess you could check that the relevant hotfixes are applied, and whether VSS tracing is enabled. Tracing may incurr a performance penality, and should only be used for troubleshooting. don't know if VSS tracing setting is the same in Windows 2008 as it was in 2003, but have a look at http://support.microsoft.com/kb/887013/
What is the status of your VSS providers and writers? vssadmin list writers; vssadmin list providers.
Are you using the system provider, or a software or hardware provider?
You could also try to allocate more space for VSS diff area.
All the writers report as healthy. Using the system and exchange writers? We don't have any aftermarket writers involved, just the ones supplied by MS.
VSS is currently setup to just place the shadow copy into the root of the volume where the snap is being taken, ie if we are snapping databases on G: then the shadow gets placed into the root of the G:. For an exchange backup, would the snap be a "clone" or "copy-on-write" shadow copy?
Besides moving where the shadow copy is getting placed during the backup, is there anything else that can be changed to affect the performance of VSS?
For example, i'm able to get a non-VSS file backup off these machines at approx 90meg/sec, however, as soon as VSS is involved the most I see is 25meg/sec.
AFAIK the VSS system provider is Copy-On-Write only. If you use SW or HW providers you can choose in their configuration whether it should be space-efficient (COW), fully allocated snapshots, or full clones.
COW performance impact primarily happens if you have a lot of activity in the application, in this case Exchange. If you have a system that constantly update the information stores, indexes, etc, the VSS provider must spend a lot of time keeping track of the block changes. But I must admit, have yet to see a performance degradation of the scale you experience.