SUMMARY 2: Was: Ultra 2/170 2GB root disk faulty?, Now: SCSI chain issues

From: Matt Reynolds (reynolmd@helios.aston.ac.uk)
Date: Tue Jan 26 1999 - 07:44:28 CST


  Hi again,

  Following my recent summary, I have received more information
regarding SCSI issues in general from a few people. I made four points
in my previous summary:

> 1) Does a wide-to-narrow SCSI adapter cable properly terminate the
> 'wide' pins which go no further? This was something I was worried
> about from the moment we fitted it, since I suspect not.

  Some do, and some don't. Roger Fujii pointed me to the following web
site which has some information regarding this issue.

  http://resource.simplenet.com/files/68_50_n.htm

  Basically it seems most wide-to-narrow SCSI cables DON'T terminate
the high 8 bytes. This is not a problem where you are ONLY using
narrow devices on the bus, but on a mixed bus this can cause trouble.
Consequently I'm surprised our system has run so happily for so
long...!

> 2) Someone suggested we synchronise the RPMs of the disk drives on
> the chain. Was this a serious suggestion?!

  This, as I suspected, is plain not true. (Drives may exist where
you can control the rotation speed, but I've never seen one...)

> 3) Someone said that the speed of the chain is totally dependent
> upon the speed of the slowest device on that chain. I don't think
> this is true, because the SCSI adapter negotiates transfer rates
> with each device entirely separately. (Unlike IDE.)

  My statement was correct, but the qualification that this only
applies to modern SCSI devices should be added to this. Much older
ones may well be unable to do this. Obviously this does not apply to
the SCSI card on an 18 month old Ultra 2 though...!

> 4) Someone suggested disabling tagged command queueing, but I don't
> think this would help. All the disks can handle tagged command
> queueing okay with the exception of the CDROM and the Tape drive.

  Again, this can be a problem for older drives which 'support' tagged
command queuing but don't necessarily get it right, but the kind of
drives we're using will not be worried by it. (The oldest is ~3 years
old.)

  John Weekley also commented that scsiping (which scsiinfo uses) can
cause data damage problems if used regularly, and high throughput can
excascerbate this problem. Perhaps 'prtconf -v' would be a better
source of information regarding SCSI transfer rates.

  Anyway, thanks again for useful information go to:

  Roger Fujii <rmf@unixbox.lookhere.com>
  David L. Markowitz <David.Markowitz@litronic.com>
  John Weekley <weekleyj@inlink.com>

-- 
#        -+- Matthew Reynolds, Contract Research Assistant -+-       #
#        -+-              Aston Space Geodesy              -+-       #
# Email: reynolmd @ sun.aston.ac.uk Web: http://www.sat.aston.ac.uk/ #
# Phone: +44 (0)121-359-3611 x4552  Fax: +44 (0)121-333-3389         #



This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:13:14 CDT