01-05-2015 01:32 PM
We are using Enterprise Media server to back up big local database dumps around 3.0TB daily. It takes around 5 hours for two LTO5 tape drives with around 90MB/sec speed. We are trying to find a way to reduce the backup windows, instead of adding more tapes, is there any way such as SAN client can make better performance? Any one has idea about SAN client, since we are backing up locally on media server, not sure if SAN client will still help. Please share with your ideas.
we are on 7.5.0.6, media server is on AIX 6.1.
Solved! Go to Solution.
01-09-2015 01:57 PM
I understand from this response that the backup server referenced here is the media server. You assumption that the SAN client will not help you is correct because essentially the SAN client is a very stripped down media server.
Without multiplexing or adding drives I cannot see how you can cut back the time. Even if you did add drives you would need to figure out how to split up the data into multiple streams. What kind of database is it?
01-07-2015 06:22 AM
First question is what is the data flow? Is it DB dump to local disk on DB server, NBU client backs up dump over LAN to media server, Media server spins to tape? Or is it DB server dumps over LAN to NFS share on media server, Media server backs up local file system to tape?
If it is the first scenario then yes the SAN client can help with throughput however, you would probably need to increase the amount of drives or stagger the backup windows so you do not have drive contention.
If you are using the media server as a NFS server and backing up locally you are getting the best performance possible with the amount of drives you have unless you are not multiplexing. If you are not then enable multiplexing and set streams to 4 and see if there is any improvement.
01-08-2015 01:26 AM
I agree with Andrew.
You first of all need to find out where exactly the bottleneck is.
Start by testing read speed from disk on media server. See:
Measuring disk performance with bpbkar
http://www.symantec.com/docs/HOWTO99824
Also wondering how long it takes to write these local database dumps?
If you really want to save on backup window, consider using NBU agents instead of database dumps.
01-09-2015 09:30 AM
Yes, it is the Cache database dumps on backup server. EMC VMAX SAN disk based filesystems hold on these dumps. It is synced from production server. After synced completed, the morr will be splited from prduction for backup on backup server.
Agent is not allowed on production server. We use our backup server to hold the dump copy for daily backup. we can use more tape drives to reduce backup windows but too many tapes will be needed for off site rotation... the backup performance is decent now but we have been asked to reduce the backup windows from 5 hours to 2. Wondering is SAN client operation can help in my situation.
Backup server is only used for backing up this dump.
No muplexing is being used since we also have to guaratte the restore windows to 2 or 3 hours.
01-09-2015 11:38 AM
Felix,
If you cannot install the client on the production server then a SAN client will not work on the production machine. When you say the mirror is mounted on the backup server do you mean a server that is going to perform the backup? If so does the scenario meet what is happening:
If this is the case then yes using a SAN client on the backup server would work PROVIDED the backup sever is not already a master or a media server. A few configuration considerations:
01-09-2015 01:25 PM
Thanks Adrew. Yes, this is our case. And we do use a decidated media server to back up the mirror locally. will the SAN client option still help us? Since the mirror is already assigned from storage to our media server, i am not sure if SAN client will still apply to us.
01-09-2015 01:57 PM
I understand from this response that the backup server referenced here is the media server. You assumption that the SAN client will not help you is correct because essentially the SAN client is a very stripped down media server.
Without multiplexing or adding drives I cannot see how you can cut back the time. Even if you did add drives you would need to figure out how to split up the data into multiple streams. What kind of database is it?