I searched the forum and didnt see anything specifically regarding this. if there is something, please point me to it.
We implemented a networker solution with data domain 3 years or so ago. our SQL admin had been insistent on being in total control of the sql backups and we provided him a cifs share on a DD620 that was local to the sql box location.
It is unclear as to WHEN this started but im trying to get ahead of it. Apparently a job to backup a 170gb db is taking about 12 hours when he backs up directly to the dd620 cifs share. However if he backs up localy, its 20 min to backup and 40 min to xfer to teh same cifs share.
i dont know anything about sql. so i am hoping i can get some insight.
would the in line dedup of the dd620 increase backup time by 11 hrs?
Is there something in the sql backup that would keep "touching" the data or do something to slow this process down when it is backed up directly to a dd620's cifs share?
is this not a good practice to backup directly to the cifs share?
I don't know - depends on file content and de-dupe ratio. With backup application I can see those much easier than when doing it directly. If that dump represents same content as SQL dump, then question is why save from local disk versus save from app directly is so different. If not, ask DBAs to create new dump locally and then copy it and compare the speed.
if there is a better way to test the cifs share let me know. i just dumped a 40gb file into it from the same local location and it is running at about 25mbps using windows explorer. so that would have taken about 30 min or so.
Assuming CIFS performance is ok (you should test it) another thing which may influence it is the fact that SQL admins like to use compression which will slow down backup and make it less de-dupe efficient. Obviously, dump to local disk does not take advantage of de-dupe as it does on DD.
Liliana i appreciate the suggestion. This may end up becoming a situation where we change how sql is backed up. But as Hrvoje pointed out, at the moment we want to preserve our current backup practice. Just providing the DBA with a location on a cifs share to put his backups.
the surprising thing is the difference in backup times between doing a backup directly do the DD620 and doing a local backup then xfering to the dd620.
the purpose of this thread was to find out if that was "wrong" to do.
are there any tweaks that can be done the way SQL dump backups to a CIFS directory, somethign like write size, i/o size ..anything of that nature that could be adjusted ?
I still don't have a concrete resolution for this. im currently looking into our sans performance. many of these components have their luns on our san, its possible that the san has a high % utilization. I have a lot of data to mill through. I will post here if I ever find a resolution!
ble1
6 Operator
•
14354 Posts
•
56186 Points
1181
0
Posted January 5th, 2015 07:00
I don't know - depends on file content and de-dupe ratio. With backup application I can see those much easier than when doing it directly. If that dump represents same content as SQL dump, then question is why save from local disk versus save from app directly is so different. If not, ask DBAs to create new dump locally and then copy it and compare the speed.