Subject: Re: pc164/ncr scsi problems
To: None <mouse@Rodents.Montreal.QC.CA,,>
From: Ross Harvey <>
List: port-alpha
Date: 08/29/1999 21:34:52
> From: der Mouse  <mouse@Rodents.Montreal.QC.CA>
> >>> For reference, this drive was working, synchronous, under SunOS
> >>> 4.1.4 in a Sparc LX (Fast SCSI, just not Ultra) at 10Mbit/s.
> >> Also for reference, my disks were working fine _for_years_ [...]
> > I hate to sound like a broken record, but 4 drives is really too many
> > for a singled-ended chain, especially if any of them are external.
> This is not my experience, unless you mean the comment to apply
> specifically to an alpha, or to that scsi card, or some such.
> I have a SS1+ at home - the machine I'm typing on right now, in fact -
> that I booted with six drives on, all external:
> - Micro-SCSI to DB50 cable (approx 1 metre).
> - Enclosure with two disk drives and a tape drive (approx .5 metre
>    internal ribbon cable).
> - DB50 to DB50 cable (approx 1 metre).
> - Two drives sitting loose on top of the enclosure, ribbon cable and
>    connectors from a dead enclosure, powered by a pc power supply
>    outside a case (approx .5 metre ribbon cable).
> - Custom cable, DB50 to DB25 - interestingly, many of the DB50 pins are
>    not present (presumably grounds).  This cable was made by giving the
>    cable shop a pasted-together assembly of adapters and cables and
>    telling them to build a single cable equivalent to it.  (Approx 2
>    metres of cable.)
> - SCSI-only zip drive.
> - Termination done with the switch on the zip drive.
> Worked fine.
> And there was the setup I had on a Sun-3 with seven disks mounted
> between two pieces of sheet metal, connected with a piece of ribbon
> cable with seven Bergs on it...terminated by resistor packs on the last
> drive.  That too worked Just Fine, though admittedly the cabling was
> less questionable.

Your one data point is easily explained, but please, do I _really_ have to
explain the difference between anecdotal evidence and signal integrity

Let's put this in more easily visible terms: let's suppose we have a bridge
design rated for 3 tons. One day, you drive a 10 ton truck over one such
bridge, maybe even for a year .. does that mean every such bridge is really
safe for 10 ton trucks?

But secondly, Brett's case was 10 MHz, yours is only 5 MHz. FOR YOUR ANCIENT
SCSI THE RULES ARE DIFFERENT. Geez. And I'm assuming you negotiated sync,
if not, the difference is more like 500%, instead of merely 200%.  So your
test is completely inapplicable to his case or almost any other non-
prehistoric setup. (Different bridge design, or different truck. :-)

Thirdly, you may have an electrically quiet environment.

Fourthly, the really old equipment like yours, besides being half the bus
clock rate, also had considerably slower rise and fall times due to (1)
older, bigger, slower transisters in the output drivers, and (2) they are
probably external bipolar IC's anyway. You don't see that much, anymore.

And no, I did not mean to apply it specifically to alpha or a particular
scsi card. What exactly did _you_ mean? That all of the world's knowledge on
signal integrity is invalid, because you saw an old system once work fine
with a pile of targets? 

For the record, the SCSI-2 Standard specifically recommends against using
any fast sync mode with single ended SCSI. *I* didn't even go that far,
because almost every SCSI-2 system violates that recommendation. But
engineers who understand transmission lines know how marginal it is.

Let me know if you want to debate issues of signal integrity any further. :-)