Disaster in progress: disk is dying, tape doesn't work :-(

[Randy Gobbel]

It appears that, indeed, the medium errors I've been seing are not a problem
with the aic7xxx driver, the disk really is failing.  Judging from the part of
the grown defect table that I am able to print out (the table is too big for
scsiinfo to show the whole thing), one head, i.e. one surface, may be failing
completely, since all the sectors that I can see in the list have the same head.

I just blew my entire weekend in a futile attempt to back up my system before
it becomes unuseable.  I tried running dump with every possible combination of
block size setting, and sync/async, enable/disable disconnect, using both a
2.0.35 and a 2.1.122 kernel.  It would be nice if the aic7xxx driver has some
options for more verbosity--even with verbose==0xffff it never said that the
tape drive was *not* synchronous, or that disconnect was disabled.

In 2.0.35, dumps would almost always complete without any error messages, but
I was unable to get "restore -C" (i.e., verify) to get very far before
complaining about massive numbers of errors.  I'm not sure whether the tape is
really unreadable, or the software is broken.  In 2.1.122, things are much
worse: dump gets all the way to the end, then hangs trying to close the tape.

All dumps were run in single-user mode.  The tape drive is a Conner (i.e.,
Seagate) CTT8000-S Travan TR4 drive, firmware 1.22.  The disk is a Quantum
Atlas I, firmware L915, and the adapter is a 2940UW with firmware 1.23.  I'm
running driver 5.1.0pre12.

I could really use some hints about what SCSI parameters are most likely to be
right for a tape drive, re: disconnect and sync/async.  It appears that
disabling disconnect doesn't really work at all, because it causes the disk to
get tons of SCSI timeouts.  Async/sync I'm not sure about--I really don't
understand the pros and cons of each mode.

If anyone out there has *ANY* suggestions about what to do here, I would
really appreciate hearing from you.  I'm getting desperate at this point, and
the number of grown defects is outrageous and growing rapidly: 963 when I
looked just now.  If this is not an emergency I don't know what would be....

-Randy

To Unsubscribe: send mail to majordomo at FreeBSD.org
with "unsubscribe freebsd-aic7xxx" in the body of the message