> From: woods@most.weird.com (Greg A. Woods)
> [ On Mon, June 29, 1998 at 16:18:01 (-0600), Michael K. Sanders wrote: ]
> > Subject: Differential SCSI?
> >
> > Does anyone have any opinions on using Differential SCSI?  I realize
> > that the controllers are more expensive (~$350 for the 2944).
>
> In my opinion the "common" single-ended SCSI bus electrical design is a
> piece of crap and just as per the IEEE specs say -- it's a bloody wonder
> it works at all at any speed.
>
> The differential SCSI bus is indeed far superior, with better electrical
> noise immunity, far longer maximum cable lengths, and fewer termination
> problems.
>
> I'm not an electrical engineer by trade, but I'm not entirely fresh off
> the turnip truck either.  I've never had problems with busses using
> differential signaling (eg. SMD drives with 50' of cable!), but I've had
> no end of problems with single-ended SCSI.

I _am_ "an electrical engineer by trade", and I agree 100% with everything
Greg said. The only reason single-ended SCSI is so popular is because almost
all drives are internal, and there it works well enough, usually.  The
SCSI-2 spec flatly states (Section 5.1) that using single-ended signalling
in the fast synch mode is "not recommended", and IIRC Ultra SCSI limits it
to 3 feet or 18 inches depending on the number of devices and speed. If I
had bet a dollar every time someone tried to tell me "SCSI can be 6m"...

A good compromise between common practice and good practice is to use
single-ended internally (one or two drives only) and differential externally
or for long chains.

--Ross Harvey

> I'll also note that all of the decent SCSI-based RAID sub-systems I've
> researched in the past few weeks come only with differential host
> interfaces.