DLT8000 slow backup times on K580

[Geoff Gowers_1]

Hi,
I have had an interesting day looking at an issue on a K580 running HPUX 11.0.0 patched to March 04.

2 DLT8000 HVD hot swap drives have recently been installed and claimed on 8/0.3.0 and 10/8.3.0.
stape drivers claimed by default.

When running fbackup against the drives it was noticable (under controlled tests) that one drive was approx 4 times slower than the other

Scsi bus and cables were eliminated during test, and the outcome was that all 3 drives we had available for test marked as 'original' C6379-63001 units worked normally, whilst the two drives we had marked HP exchange part C6379-69201 TH8AF-nn revison A02 ran slowly.

The ones that worked OK were on firmware 20,44 and 80, product revision 114,226,250 respectively; whilst the slow ones were on firmware 80.

Having looked at other threads, we suspected firmware above 59, and possible patch updates required, but I don't believe this is shown up
in testing.

(We also have a live (mirror)system backing up to 2 dlt8000 successfully both at f/w 80.)

We did also note that of the good ones, fbackup test times, are fastest for the lowest firmware revisions- but not significantly so.

Can any one shed a light or pointer to this apparrant problem with the C6379-69201 A02 devices ??

We have no errors logged other than fbackup timeout that caused the challenge to be highlighted !

Thanks.

[Rick Garland]

Are the minor numbers of all tape devices the same?

Are you trying to do SW compression while HW compression is already turned on (via the minor number)?

[Geoff Gowers_1]

Thanks Rick,
May have to do another check, but as far as I understand , the minor device numbers were unchanged, problem with the specific drive type could be generated on both bus by hotswapping the drives in the enclosure.

The same script was used to test all devices, but cannot confirm what compression was used till next week.

Geoff

[Geoff Gowers_1]

Perhaps I was too quick to malight the tape device. Further overnight test showed on the 'good' devices one normal at 1hr and one long at 5hr.
More testing, looking at writing rates showed that backing up to the two devices was ok in parallell for up to an hour at about 10mb/sec , then the device dev/rmt/1m on the core io board scsi 10/8.3.0 lost speed and slowed right down. The test was repeated sequentially, with less data, both jobs ran fine on the 'good' bus approx 20 min each, and on the 'bad' bus was abandoned after 45minutes with much less data written.
We are currently changing the scsi cards, and if no improvement there, then the cable/terminator and retest.

[Geoff Gowers_1]

Well we are happy that backups are performing well now, having swapped out two replacement dlt8000 that did not perform with this configuration, it happens that they are both at fw level 80. Our working drives now are 2 at fw80, 1 at 20, 1 at 44.

We are awaiting LTT to be available to do a bit more testing and firmware upgrade on the spare devices to position ourselves with a decent spare device in the event of future failures.

[Geoff Gowers_1]

See previous comment