AHA-2940 Ultra/Ultra Wide

[desmond at ic.eecs.berkeley.edu]

Taras M. Dowhaluk wrote:

>  Dan Eischen wrote:
> > 
> > I'd also suggest using one of your disk drives to terminate the bus,
> > and not your JAZ.  Perhaps someone else can comment on whether your
> > QUANTUM FIREBALL ST6.4S 0F04 needs firmware upgrades?  Are you uising
> > tagged queueing?

> At the moment I have:
> 2940 - disk 0 - disk 1 - Jaz.
> Someone suggested that I turn the disk drives upside down such that 
> pin 1 faces the other way, that way I can run my scsi cable so that I 
> have 2940 - Jaz - disk 1 - disk 0, if you know what I mean.

I had horrible problems using a ZIP in my external SCSI chain.  I have
always used an active terminator, but the ZIP (as does the Jaz) uses
SCSI-1 cabling, and this seems to be the source of many problems, even
when transfers do not involve the ZIP.  I am using the stock 2.0.30
kernel.

Putting the ZIP at the end (even with an active terminator) does not
seem to terminate the bus properly -- I witnessed retries accessing
other devices.  Putting the ZIP in the middle seems to eliminate the
retries because at least the end device is next to an active
terminator.

However, I found a more sensitive test: whether my DAT streams
properly.  It will only stream properly when I completely remove the
ZIP from the chain.  I am not confident I am ever getting proper
termination with the ZIP involved because of the silly SCSI1 cable.
Another sign of this is that the auto-termination feature of the 2940
seems to be confused (it fully terminates) whenever the ZIP (and its
SCSI1 cable) are attached.

My setup is:
	
  internal     internal
ultra-wide ---  ultra    -- cd  -- 2940UW -- ZIP  -- DAT  -- active
  disk          disk                                         term

Internally, the narrow devices are on narrow->wide adaptors.
Externally, cabling is a mess: SCSI-III --> Centronics and then for
ZIP, Centronics to their funky SCSI1, back out to Centronics, with the
remainder Centronics-style connectors and active terminator.  But the
total length is under 3 meters.

This setup is stable (no retries appear in the logs), but tape backups
get corrupted (the DAT does not stream and you can hear it seeking a
lot).

For backups I therefore run:

 internal     internal
ultra-wide ---  ultra  -- cd  -- 2940UW ---- DAT  -- active
  disk          disk                                  term

I have concluded that while you may get no SCSI errors, the ZIP messes
up termination enough that it isn't worth having on the external chain
(I wish I got an IDE internal ZIP).  I don't think the Jaz is any
different in cabling or termination.

Desmond Kirkpatrick