We help IT Professionals succeed at work.

SAN Snapshot DB performance

anushahanna
anushahanna asked
on
When you do a SAN Snapshot of a database, using VSS technology, (through any vendor),

*is it always only a refresh of the newly written data since the last snapshot?
*Is there any vendor who provides a complete snapshot that is 100% new copy of the source database, so that there is no impact on performance and interdependence between source and new snapshot database?
Comment
Watch Question

Senior IT Advisor
Top Expert 2008
Commented:
Take a look at Netapp.  Their snapshots are point in time and the controller just gets a location where the blocks are.  It is not a copy on write as with most SANs.  It has tie ins to SQL, Exchange, Oracle, etc..  There is no impact on performace because the data is not being written

http://media.netapp.com/documents/snapshot-datasheet.pdf

*is it always only a refresh of the newly written data since the last snapshot?
Yes and no - depends exactly how you do it. It also depends on the snapshot technology in use.

*Is there any vendor who provides a complete snapshot that is 100% new copy of the source database, so that there is no impact on performance and interdependence between source and new snapshot database?
Yes - all SAN vendors do this. A 100% copy is usually called a clone, while a snapshot is a copy on write or changed block copy. Clones will always have zero performance impact on the source (except during the clone creation), while a snapshot will always have a performance impact, even if its minor.
Paul SolovyovskySenior IT Advisor
Top Expert 2008

Commented:
@meyersd: I agree on everything you stated but on the Netapp there is no copy on write or change block copy.  The controller just freezes the original blocks and and new blocks are written as they would be in anyway during snapshot.  This is why the snapshots take almost no space unless you delete data and don't delete teh snaphshots that have the blocks frozen.

Took me awhile to figure it out but it's cool stuff.
Agreed - it's well groovy technology- but despite NetApp's claims, the reading from the snap has a performance impact on the source - it's unavoidable. The only way to isolate any impact on the source LUN is to make a clone on separate spindles.

Author

Commented:
Thanks for confirming that.
Thanks! Glad I could help.