backups taking too long

[Andrew Griffin]

Hello,
On one of my N-Class servers I just upgraded from an old DL/15 DLT7000 library to a newer SureStore 2/20 LTO library with 2 Ultrium drives. The library is directly connected to the server via fiber connection - the fc card in the server is an HP Tachyon 1Gb. I have all the latest and correct patches for the library (this was the first thing I tried), and the server is up-to-date with the September 2003 Quality Pack.
The server has two SureStore 12H AutoRAID arrays with about 270Gb worth of data on them. The arrays are connected in sequence, i.e. - two SCSI cards in the server connected to array A, array A connected by SCSI to array B.
My problem is that full backups are taking about 12 hours to complete, and 12 hours for 270Gb of data is roughly only 20Gb an hour. I had expected much more than that, especially with 2 drives working in tandem. The drives are Ultrium 230s, rated at (up to) 54Gb/hour each. I was assuming I would get at least 60Gb/hour between the two of them.
Like I've said I've already searched for and applied any patches that may be relevant, but there is one thing that might be a concern - The server and the library are in 2 different racks, so I wasn't able to use the shorter 4ft. fibre cable that the library came with. Instead I had to use a 30ft one (the shortest we have available here) to connect the server to the library. But this should have any affect on the transfer speed, should it?
I'm kind of running out of ideas here. I'd like to get these backups completed in under 6 hours. Any suggestions?

[Andrew Griffin]

oops, I forgot - we're using OmniBack II version 4.00

[Joshua Scott]

The problem may be in the fact that the 12H arrays are HVD (20MB/s) while the combination of the 2 ultrium drives are 30MB/s native. the 12H can only transfer out 72GB/h. if you have both drives pulling backups off the arrays, then they are expecting input at 108GB/h.

My reasoning may be flawed, but i know that if you don't feed enough data to the drives, then they have to stop, rewind and read to find the end of the data on the tape, then start writing again. this is known as the "shoeshine effect" and could drastically slow down backup time.

If you only used (1) of the drives to do the backup, you should be able to back everything up in 5-6 hours MAX.

[Leif Halvarsson_2]

Hi,
A common problem with Ultrium and SDLT drives is that the disk system becomes a bottleneck, particulary with older disk systems or if the filesystem contains a large number of small files. You can do some performance tests if you want:

- Create a null device in OmniBack (file device fo /dev/null) and try a backup.This backup eliminates possible performance problems with the tape drives. You can't expect better performance when backing up to tape then when backing up to null device.

- Do a rawdisk backup of one of the Raid systems to one of the Ultrium drives. Rawdisk backup bypasses the filesystem overhead and you will get the raw performance of the array. You can not expect better performance with a filesystem backup then with a rawdisk.

[Alzhy · Accepted solution]

Your problem is your AutoRAIDs are just too slow (~ 5-8Mb/s ). No matter how fast your drives are, your arrays will just be able to feed the tape drive(s) it's measly maximum throughput...
.
Imagine pulling a 50 ton rig with a 575HP Maranello! No matter how fast your puller - Maranello(the tape drives), the rig (your arrays) can only go as fast...
.
So no matter what you will do, your full backups will still happen in 12-13 hours (270GB @ 20 GB/hour)