DLT drive oddity on HP-UX 11.11

[Stan Sieler]

One of our clients is having a problem reading data from a DLT drive. (First noticed
with our product, but apparently confirmed
with "dd", seemingly moving the problem
outside our software :)

HP-UX 11i (11.11) on PA-RISC.

It's a new DLT drive, and it sure looks like for every read() call
we get what we should get and then a spurious EOF.

I.e., if we expect: data, data, EOF, data, data
(from 5 calls to read())
then we see: data, EOF, data, EOF, EOF, EOF, data, EOF, data, EOF
(from 10 calls to read())

Taking a single tape into consideration...

We know the tape has hundreds of files ... (read it on another computer)
(a "file" here means one or more data records followed by an EOF)

But, we "see" only one file (because our software stops at the first
double EOF, and even "dd" with -files stops early)

> sys42:/ # dd if=/dev/rmt/c45t5d0BEST of=curtis files=10
> 4+2 records in
> 4+1 records out

Unfortunately, we know that the first file has about 4 blocks,
so the above tells us that dd only processed one file, not the 10 we asked for.


Any suggestions appreciated!


Various data (mt status, lssf, ioscan, ioscan -fnC) follows...

mt status: (after the "dd")

sys42:/ # mt -t /dev/rmt/c45t5d0BEST status
Drive: QUANTUM DLT8000
Format:
Status: [41118985] BOT online compression immediate-report-mode
File: 0
Block: 0

An lssf shows the device is claimed:

stape card instance 45 SCSI target 5 SCSI LUN 0 at&t best density
available at address 1/0/10/1/0.5.0 /dev/rmt/c45t5d0BEST
stape card instance 45 SCSI target 5 SCSI LUN 0 berkeley best density
available at address 1/0/10/1/0.5.0 /dev/rmt/c45t5d0BESTb
stape card instance 45 SCSI target 5 SCSI LUN 0 at&t no rewind best
density available at address 1/0/10/1/0.5.0 /dev/rmt/c45t5d0BESTn
stape card instance 45 SCSI target 5 SCSI LUN 0 berkeley no rewind best
density available at address 1/0/10/1/0.5.0 /dev/rmt/c45t5d0BESTnb

(it's the first one above)


sys42:/ # ioscan (subset)

H/W Path Class Description
====================================================================
root
1 cell
1/0 ioa System Bus Adapter (805)
1/0/10 ba Local PCI-X Bus Adapter (783)
1/0/10/1/0 ext_bus SCSI Ultra320 A6961-60011
1/0/10/1/0.5 target
1/0/10/1/0.5.0 tape QUANTUM DLT8000


sys42:/ # ls -l /dev/rmt (subset)
crw-rw-rw- 2 bin bin 205 0x023000 Jan 6 2006 18m
crw-rw-rw- 1 bin bin 99 0x1b0000 Mar 4 09:34 1m
crw-rw-rw- 1 bin bin 205 0x2b9000 Feb 20 14:23 22m
crw-rw-rw- 2 bin bin 205 0x023000 Jan 6 2006 c2t3d0BEST
crw-rw-rw- 1 bin bin 205 0x023001 Jan 6 2006 c2t3d0DDS
crw-rw-rw- 1 bin bin 99 0x1e0000 Mar 4 09:34 c30t0d0BEST
crw-rw-rw- 1 bin bin 99 0x1f0000 Mar 4 09:34 c31t0d0BEST
crw-rw-rw- 1 bin bin 99 0x200000 Mar 4 09:34 c32t0d0BEST
crw-rw-rw- 1 bin bin 99 0x22f000 Mar 4 09:34 c34t15d0BEST
crw-rw-rw- 1 bin bin 99 0x230000 Mar 4 09:34 c35t0d0BEST
crw-rw-rw- 1 bin bin 205 0x2d5000 Mar 4 09:57 c45t5d0BEST

The last one above is the DLT drive in question.

sys42:/ # ioscan -fnC tape (subset)
Class I H/W Path Driver S/W State H/W Type
Description
========================================================================================================================
tape 18 1/0/1/1/0/1/0.3.0 stape CLAIMED DEVICE HP C5683A
/dev/rmt/18m /dev/rmt/18mnb

tape 8 1/0/6/1/0.1.6.255.0.0.0 atdd CLAIMED DEVICE IBM ULT3580-TD2
/dev/rmt/8m /dev/rmt/8mn

tape 22 1/0/10/1/0.5.0 stape CLAIMED DEVICE QUANTUM DLT8000
/dev/rmt/c45t5d0BEST /dev/rmt/c45t5d0BESTb /dev/rmt/c45t5d0BESTn /dev/rmt/c45t5d0BESTnb
========================================================================


thanks,
Stan
sieler@allegro.com

[A. Clay Stephenson]

1) Make absolutely certain that this SCSI bus is terminated in exactly 2 places -- on the physical ends of the bus. Related to this, make certain that at least 1 device on the bus is supplying termination power. Poor/incorrect termination can cause a device to almost work well --- the worst possible kind of problem.

2) Try your dd again but specify a blocksize; I rather doubt that the tape was written in 512-byte (the default for dd) blocks.

[Stan Sieler]

hi,

I'll double check the SCSI termination, thanks!

As for the block size with dd...
(1) the tape has variable size records ...
I could say -bs=128kb, since no record
will be larger than that; but (2) testing
with dd on a local DLT and a local DDS
shows that's not needed to simply verify
that the tape can be read past the first
real EOF.

thanks,
Stan