Slow restore performace

[philpe]

Hi,
I've been having a restore performance issue with Veritas Netbackup DataCenter. Here is the history.

I have a Sun Server with a 270Gig Raid 5 Partition (disksuite) with about 40 Gigs worth of data. I backed this up on Friday in about 3 hours over a 100 MB/s Full duplex network connection.
Then I killed the Raid 5 partition and recreated it exactly the same. Now I want to restore the data on the partition.

I started the restore on Monday at 10AM. The restore needed 2 Tapes, the image was not multiplexed, and the tapes consisted on one Cumulative Incremental backup from Friday, and one Full Backup tape from the Monday before.

The restore of the 40Gigs of data finished on Tuesday at 7:30 AM just over 21 hours later.

Now I know restores can take longer, but this is ridicules, It should not take 21 hours to restore 40 Gigs of data from 2 tapes not multiplexed. I checked the network connections, and they are running at 100 Full on the Sun box, and 1000/Full on the WindowsNT Netbackup Server. The tape drives are DLT8000 drives.

Any idea what else could be the problem here?
Thanks for your help.

Phil

 

[sdeb]

Could you try a partial restore of data directly on the LEGATO server to avoid any communication problem (Like auto-negociation ...) and to view local restore performances.

I agree with you, it should not be so long. Especially in a non-multiplexed configuration.

 

[philpe]

Thanks sdeb,

I found a document from Veritas about issues relating to restore performance with Mixed client/server. Basically, it says that you need to tune the tcp_deferred_ack_interval on the Sun box so it doesn't wait so long for responces from the server.

In my testing so far, I set it to 30 ms (from 100) and the performance jumped right up, restored a 1/2 gig in about 4 minutes. Thanks again for your responce, and if anyone is interested, the Veritas doc can be found at:

http://seer.support.veritas.com/docs/237215.htm

Phil