DLT8000 Tape backup time differences

[Joel Shank]

Hi All

I hope someone will be able to give me a place to look to for this.

I use fbackup to dump user data to DLT8000 tape drives. One server (K580) dumps about 27GB in about 2 1/2 hours (running HPUX 10.20 using GSC F/W SCSI adapter). The other server (V2200) dumps about 6GB in 4+ hours (running HPUX 11.00 using C875 F/W SCSI adapter).

Can anyone suggest where/what to look at to see why my V is taking so much longer to dump so much less data? I am stumped. It doesn't make sense to me.

Thanks,
JLS

[Ken Scharpell]

That does indeed look strange. Are both using the same fbackup options in their config files ?

Are both the tape drives on a dedicated bus of their own - they dont share it with some disks ? if so on the V class this could be why the backup is running so slowly, anything else on the same SCSI controller as the tape drive could seriously slow down the tape drive. On the V class each controller slot is also shared by another slot, so its possible you think your drive is on a dedicated bus when in fact it is being shared.

[John Palmer]

Hi,

As Ken stated, the fbackup options supplied in fbackup's configuration file (-c option) can make a vast difference to performance.

This previous thread may be of interest:-

http://my1.itrc.hp.com/cm/QuestionAnswer/1,1150,0x629a0559ff7cd4118fef0090279cd0f9,00.html

Regards,
John

[Joel Shank]

I do not use a config file, so I am taking the fbackup defaults. I do not believe I am sharing the DLT with disks. I did an ioscan:
ext_bus 1/2/0 c720 CLAIMED INTERFACE SCSI C875 Fast Wide Differential
tape 1/2/0.2.0 stape QUANTUM DLT7000
tape 1/2/0.3.0 stape QUANTUM DLT8000
ctl 1/2/0.7.0 sctl initiator

The DLT7000 is not being used, but the DLT8000 is daisy-chained off it. Could that be causing the slow-down? Should I use a config file, and if so, what values should I use to get better performance?

Thanks for your help...
JLS

[Chris Moore]

The extra tape device on the bus won't affect your performance if it's not being used.

Check the minor numbers of the device files you're using - are you using the "BEST" device file on both?

Also, if you can, watch the lights on the drives while the backup is running. See what density is getting picked and verify that the compression light is on in both cases.

[Joel Shank]

Thanks for your comments and suggestions. I think I found my problem. I was comparing apples to oranges. My K server is attached to EMC disk drives, my V is using fiber disks. Although the fiber is faster than SCSI, it isn't nearly as fast as sequentially reading from EMC disks with 2GB of cache.

I will be increasing the number of fbackup readers in an attempt to speed the disk IO up.

Thanks again
JLS

[paul courry]

Please also check to make sure you are not using DLT III tapes. These for the I/O rate to 1.5Mb a sec.

[Franz Sigl_1]

according to your posting you are running a DLT8000 tape drive on HP-UX 10.20.
How did you get it working, which patches/drivers are required?
How did you create the device files?

# ioscan -funCtape
Class I H/W Path Driver S/W State H/W Type Description
===============================================================
tape 1 56/52.3.0 tape2 CLAIMED DEVICE TANDBERGDLT8000


Thanks
Franz

[Darrel Louis · Accepted solution]

Next time you experience such things you can use the following to tackle where the bottleneck resides:

time dd if=/dev/dsk/c0t5d0 of=/dev/null bs=64k count=5000
time dd if=/dev/dsk/c0t5d0 of=/dev/rmt/0m bs=64k count=5000

With the outcome you can calculate the troughput of your disk and tape.