Subject: Re: ncr hangs system
To: Ross Harvey <>
From: Matthew Jacob <>
List: port-alpha
Date: 03/18/1999 13:17:10
> I kind of agree with you here, but my predicate was not only correct, but
> the actual statement in the spec is even stronger. Since SCSI-2 only had
> FAST-10, that's what they had to be referring to.
> 	From ANSI X3.131-1994 section 5.1 page 8:
> 		NOTE 2 Use of single-ended drivers and receivers with the
> 		fast synchronous data transfer option is not recommended.
> 	and in 6.8 (pp 37) `fast ...' is defined as any negotiated data
> 	transfer period less than 200 nS. (Actually, `ns', SIC.)
> That is, they specifically discourage INTERNAL singled-ended SCSI. Talk
> about an almost universally ignored recommendation! Think of all the trouble

*SMACK* Am *I* enlightened now. That note was something that escaped my
notice... But that's also from the same committee that recommended 100ohm
or greater cables years before cable manufacturers could make them...

> that would have been avoided if everyone had just listened and gone
> differential.

I argued for this at Auspex and was told that a >=200$ per drive
additional cost made this a 'poor idea'.

> I also agree: use hi-Z cables of one type, stagger the lengths but keep
> them short, and I would add: avoid PVC, use polyolefin or teflon cables.


> Of course, if you just go differential then all your SCSI problems will
> just vanish, but everyone always wants to use the built-in SCSI and the
> cheap drives. This is why I generally buy bare motherboards and add my own
> differential controllers and drives. Sometimes you can get great deals on
> these things from sellers who don't realize their value.

Actually, at this point, I'd recommend getting LVD controllers. You can,
at least for some of them, run SingleEnded, Differential, or LVD/Ultra2
SCSI chains. Talk about maximum flexibility..

> And I run my Multias diskless...

Aw, that's cheating.....