1

Can anyone give some insight on the performance impact of enabling Shadow Copy on a file server? We have a volume that has about 700 GB of data on it, used by 100 users. There isn't a really high load for changing files, but many reads.

Shadow Copies can only be enabled on the same volume due to size, and lack of alternative drives. Does having the source and storage area on the same drive severly impact performance? I know the documentation says it will, but I'm looking for real world examples.

Eddie
  • 11,332
  • 8
  • 36
  • 48
Jeff Miles
  • 2,020
  • 2
  • 19
  • 26

2 Answers2

2

I've got a similar situation (more data less uses - around 60) and the benefits of having shadow copies available to undo mistakes is invaluable. That being said your question was about performance.

Microsoft doesn't say much about it because it's quite subjective. With mostly reads you probably wont notice anything. NTFS uses copy on write to achieve snapshots. (akin to zfs if you've heard their marketing) Which means for each write to an existing file it has to write a new block and make a note about the old block being old. I've got most of my servers on a raid 5 or 6 setup so I don't notice the extra writes. On a single spindle you might.

I also use the snapshots for backup purposes, I don't need to worry about open files etc. A lot of different backup utilities (ghost, true image, and possibly Microsoft backup) leverage this.

There are so many other variables to network share performance (network, cpu, application) snapshots weren't noticed by any of our staff. Until they did did something stupid with a file that is.

It is possible to turn snapshotting on and then turn it off again later if there's a problem. So it's worth a try.

reconbot
  • 2,435
  • 3
  • 25
  • 30
  • Thanks for the specific example. We're not on a single spindle either (RAID5) so I'm not too worried anymore. I realize the question is subjective, so I won't leave it unanswered. – Jeff Miles May 06 '09 at 20:36
1

I believe the performance hit depends a lot on how much data is being changed between snapshots, and when/how often your snapshots are scheduled to run.

Shadow-copy generally doesn't do anything to your read-performance instead what happens is that whenever you write to a file that is considered to be part of a snapshot a copy will be made and your write will be made to the copy.

On one system I was responsible for I enabled shadow copy on our server, given your description I may have a roughly equivalent situation to yours and it didn't really seem cause much of a performance hit at all. I have about 100 users and around 600gb of data that is most doc, pdfs, xls and so on. I do 2 snapshots a day.

I wish I could offer you something more concrete, but since it is so dependent on your specific usage patterns, you may have to warn your users and simply try it. You probably would want to start with a conservative one snapshot a day schedule, and then increase the frequency as needed/desired.

It will be important to get some performance data on the server before you make any changes so you will have some data see what impact the change had.

Zoredache
  • 128,755
  • 40
  • 271
  • 413