Problems with restore throughput

David Greenhow

Problems with restore throughput

Hi we have the following setup
1x EVA 3000
1x MSL6030 with 2x LTO Ultrium drives and NSR
2x RP4440s

When we backup an area, it does so quite quickly but the problems start when restoring the data.
We started restoring a backup this morning at about 9am from the MLS to a directory on the EVA and so far it has restored about 12% of a 70gig backup. Thats 12% in around 9 hours.

Does anyone have any idea why its taking so long for the restore? (The data that is being restored is an Oracle environment and is crucial that it be back up and running as soon as possible)


Steven E. Protter
Exalted Contributor

Re: Problems with restore throughput

Those figures are pretty bad.

Note that if it takes an hour to create the backup, it will take longer to restore. This is because the backup is a read on disk, write to tape and the restore is a read on tape(slower) and write to disk(which can be interfered with by system activity, especially in a raid 5 environment).

Are you getting any progress at all?

Can you restore to an idle or raid 1 filesystem?

Steven E Protter
Owner of ISN Corporation
David Greenhow

Re: Problems with restore throughput

I see what you are saying but if the backup takes 4 hours, what is the expected recovery time? as an estimate. and what would the throughput normally be (these are all connected via fibre)

At present its done 14% in a 12 hour period.

Should the block size on the logical volme be the same on the tape or does that not matter?
Leif Halvarsson_2
Honored Contributor

Re: Problems with restore throughput

You tell that "it is an Oracle enviroment". Dors it mean that it is a RMAN backup/restore. If so, it may be an RMAN problem.

If it is a filesystem backup/restore your backup performance (70GB/4h) seems slow to me, a rp4440/EVA/MSL6030 should perform better. Is there a very large number of small files ?

Which backup tool are you using, if Data Protector, you should not change any default settings (e.g. bloksize).
Sheriff Andy
Trusted Contributor

Re: Problems with restore throughput


Sometimes there is a penalty for faster backups.

Are you using Legato? If so, there are two things to look at. One is the server parallelism (# of streams that can be processed by a client) and the other is the target sessions (# streams each tape drive can process).

If your client is backing up to multiple tapes, then you have to restore by multiple tapes as well. Also if you have it set so multiple clients are writing simultaneously to tape, then it takes longer for the system to browse for the files that are needed to restore because the tape is quite fragmented .

I hope that this helps some.
David Greenhow

Re: Problems with restore throughput

Not using RMAN, just a simple frecover. It takes us 4 hours to backup, how long as an estimate should it take? Also does the block size of the logical volume have to match up with the block size (or equiv) on the tape.

The oracle environment does contain lots of small files, but surely this still should have finished by now - 1am and still at work :( with 14% complete
David Greenhow

Re: Problems with restore throughput

Sorry for not posting the fix for this

It turned out to be a faulty Gbic on the fibre switch