(off topic) surprising results with unterminated tape drives

[James E. Leinweber]

short version
--------------------
Can the combination of active negation by SCSI-fast devices plus
SCSI-3 controllers backing down from unsustainably high transfer
speeds tend to hide termination problems?

long version
--------------------
I recently discovered, to my shock and horror, that a small cabal of
student-age hardware techies at our shop had been putting tape drives
into high end PC systems without bothering to terminate them.  A
typical scenario which I just fixed this afternoon would be:

  Adaptec 29160U2W SCSI card <-50 pin cable->Quantum DLT 4000 tape

They were blithely ignoring boot time warnings about undertermination
on the grounds that the tape backups "were working OK".

I'm unpersuaded; _all_ of said systems have a history of mysterious
failures of nightly tape backups, with much pointless troubleshooting
of everything - drives, cables, and PCI slots - except the missing
terminators :-)

But, they were getting a surprising high percentage of backups which
where the data on tape apparently do compare favorably with the
original data on disk.  Being a software guy who originally met up
with SCSI in the SCSI-1 Centronics connector days, when lack of
termination was a certain kiss of death, I'm a bit puzzled as to how
they got away with it.

I'm entertaining the hypothesis that the combination of:
  - fast SCSI SE devices being required to do active negation
  - SCSI-3 controllers _testing_ synchronous transfer rates and picking
    slower ones if they see data corruption at higher ones
  - the general slowness of tape drives compared to disks
  - short cable runs with only 1 device at each end of the bus

was resulting in low-speed transfers with a lot of retranmissions and
occasional errors, instead of the higher speed transfers with gobs
of errors I was naively imagining.

I did notice that comparing backup times on terminated versus
unterminated tape drives, the terminated ones are doing about 3X
faster, and the unterminated ones aren't running at anywhere near the
expected speed for streaming tape motion.

Has anyone else seen this?  Or got a better hypothesis?

-- James E. Leinweber
Information Systems   -  State Laboratory of Hygiene - University of Wisconsin
<jiml at slh.wisc.edu> 465 Henry Mall, Madison WI 53706-1578 USA  +1 608 262 0736

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