HP Storage Works DLT VS80 slow performance

[pramodm]

We are using a HP Storage Works DLT VS80 for DATA back up on hp rp 2430 Server with OS HP-UX 11.11. The DLT Catridge has a storage space of up to 80 GB compressed and can back up 6 MB/s compressed (assumes 2 to 1 compression) as per the specification of the device. Comparing this at our site the backup is taking much more time. More than 36 Hours for 80GB backup and so that we cannot complete it till now.
I checked with changing the SCSI cables and terminators, but no results. Here SAP 4.7 with ORacle9.2 DB is running on the system. The command using for the backup is cpio in SAP std backup script.
Please help to solve this issue.

Pramod.M

[Steven E. Protter]

Shalom Pramod,

First, don't expect the system to work at its rated speed. That is a theoretical speed and in real life, nothing goes at theoretical speeds.

Still we do seem to be moving a bit slow.

First, check your system, it may have OS related I/O issues that prevents full speed.

http://www.hpux.ws/system.perf.sh

Second, make sure the SCSI card is supported for use with this device. IF not, it could explain the problem.

Third, look for conflicts with the scsi id.

Forth, see that the sytem is properly patched in general.

Last on my list, an rp2430 is not the quickest system around any more and it may be past end of life. It may not be capable of using this deive at anywhere near rated capacity.

SEP

[David Nixon]

Some comparable figures from an rp5400
using a DLT-1 and 'vxdump' : 60-70GB is max
cartridge capacity achieved: data xfer rates
vary fom 6 - 16 GB/hour; depending on the
speed of the external RAID device being
dumped.

Possibly you have an I/O bottleneck. Suggest you ensure that the VS80 is on
separate IO bus to the disc drives

Also consider using 'fbackup' instead of 'cpio'.

DavidN

[A. Clay Stephenson]

cpio is singly-threaded and will have a hard time keeping a DLT streaming. If the drive is not supplied data ata a suffiicient rate to keep it streaming, performance degrades by a factor of 100x or so as the drive has to stop, rewind, and start again. Moreover, if you are using the default cpio blocksize (512) or even the 'B' option (5KiB), you are really crippling the throughtput. You can improve it somewhat by putting dd in the middle.
find . -print | cpio -ocBv | dd ibs=5k obs=250k of=/dev/rmt/xxxx
so that data are written in 250KiB chunks.

but you will be far better served to change to fbackup and use 5 reader processes and a large blocksize.

[pramodm]

The backupscript is attached. The cpio command flag is as below
# flags for cpio command
# default: -ovB
cpio_flags = -ovB
cpio_in_flags = -iduvB

also pls reffer the attachment.
We are having the hp rp 2470 server with same OS and using Hp Sure store Ultrium 230 ext drive(LTO1). The drive is capable of taking 84GB datae in a 3 Hrs time. The script is same so far.
Can give any advice....
thanks in advance...

Pramod

[perumal_2]

Hi

It looks like I/O bottle neck. Is the tape drive connected to the seperate scsi bus, were there any scsi reset after you had connected your tape drive. Look at your syslog to get any additional details
There is a known issue in rp3440 & rp4440 that there is no extra scsi card to connect the tape drive though there is an external scsi port, after connecting the tape drive in this system it creates a scsi reset and your whole disk i/o will be reset to tape speed than the disks speed. But still the confusing part is 32hrs for 80GB, why dont you check the tape drive with any other machine to isolate any issues in the drive.

TQ
Perumal