[SunRescue] Proper use of VME scsi controllers?????

[Greg A. Woods]

[ On Wednesday, February 9, 2000 at 10:32:14 (+0100), Peter Koch wrote: ]
> Subject: Re: [SunRescue] Proper use of VME scsi controllers?????
>
> The Sun3 only has SCSI-I.

I'm fairly certain the command set used is definitely SCSI-II.  The
connectors are within spec. for SCSI-II as well as "alternatives", but
they are obviously not the "new" high-density connectors first
introduced for SCSI-II.  So, strictly speaking it is SCSI-II, just not
the most modern physical options of SCSI-II.  Indeed you can safely
attach almost any SCSI-II compatible drive to a sun3 with the only
potential for problems being non-conformance to the command set and the
need for quirk handling in the device drivers.

> This means you can have at
> most 12 meter cable (every piece inside a case counts!).

Ah, don't you mean "feet"?  :-)  12m is never mentioned in the standards
documents I have until you get to talking about the most modern
low-voltage differential interfaces.

The length limit depends on the speed the bus is running at, and of
course the quality of the cable and connectors (lesson learned -- always
check back at http://www.scsifaq.org/)

At 5MHz SCSI-1 synchronous negotiation the bus may be 6 meters long iff
the cable is of high quality and of the correct impedance.

If anything runs at SCSI-2 "FAST" you have to go to 3 meters max.  Note
though that SCSI-2 FAST is *NOT* recommended for single-ended cable *AT*
*ALL*.  (It's a miracle it works at all and this issue is the source of
over half the problems with SCSI in PCs since all try to push this
option on unsuspecting users.)

> This is enough for connecting four disks, two tapes
> and a cdrom. This would be the maximum configuration
> of seven devices on the SCSI bus.
> Termination is a real problem on the Sun3. I found
> some configurations work WITHOUT terminator that failed
> WITH terminator. Funny!

The problem isn't necessarily termination but rather poor termination
power.  If you have a disk that you know supplies good termination power
then put it at the end of the chain furthest from the host adapter.
Just make sure your power supplies are all running in spec. and that you
have an active terminator (one with the little light to prove that the
terminator power is really on).
 
Proper use of an active terminator on both ends of the bus is absolutely
critical if you have any devices running at faster than 5MHz.  I don't
think the sun3 "si" controller will negotiate faster than 5MHz, but I'm
not sure any more.  (there's no mention of such things in the NetBSD
code that I can find on a quick skimming)

> Important: ALL VME-Bus SCSI controllers are TERMINATED.
> This severely limits the possibility to hung internal
> AND external devices to the bus.

It's trivial to pull the terminators on the board.  I took an internal
board set and an external frame and after putting them together and
pulling the terminators had a perfectly working internal/external
combination.  Like I said it ran reliabily with all targets full for
very long periods of time (total of maybe two years on a very busy
server with no SCSI problems whatsoever).  You can hack one together
without having to joint the parts of two boards but you do have to have
the right 6U-9U adapter with the feed-through and the header for the
external connector.

The internal terminators could even be part of the problem because they
are just passive and though they are near the source of termination
power they cannot do anything active to correct the signal itself and
prevent echos and such.

> From my experience, more than 4 devices on the Sun
> VME SCSI bus do not work properly. I try to minimize
> as much as possible, because this avoids hassle.

That's quite possible.....  Most of the 12-slot backplane models I've
seen are very good and of high-quality manufacture but there may be some
limits being pushed.  I'm almost certain they were the cause of the
problems I had with trying to run my 3/260 with 128MB RAM.  It worked,
but often crashed as much as two or three times daily.

-- 
							Greg A. Woods

+1 416 218-0098      VE3TCP      <gwoods at acm.org>      <robohack!woods>
Planix, Inc. <woods at planix.com>; Secrets of the Weird <woods at weird.com>