Subject: Re: ncr hangs system
To: None <firstname.lastname@example.org, email@example.com>
From: Ross Harvey <firstname.lastname@example.org>
Date: 03/18/1999 15:05:46
: From: Wilko Bulte <email@example.com>
: As Matthew Jacob scribbled...
> Come on, that was 1994 when this was written. I've had lengthy discussions
> with one of the fathers of Ultra SCSI and it seems to make a he* of a
> difference how well you can design and produce your silicon. And control
> your bus impedance, which gets us back to the cable issue.
> See www.symbios.com for the bus isolator chips. We can now do
> 20 meters _single ended ultra scsi_ (though point to point only)
> using these things. (we == DEC ^H^H^H Compaq storage division).
Elaborate contortions to make a bad interface work, membership on a spec
committee, necessary legacy compatibility, and examples of dubious
accomplishments have nothing to do with the fact that SE interfaces are
now and will always be poor interfaces spec'ed by bad engineers or hostages
to poorly chosen world standards. (Yes, I know, you guys don't really have
a choice, just like Auspex didn't really have freedom to choose.)
Signal integrity just wasn't well understood in the small computer world when
SCSI was thought up.
And in fact, after we went to the 68-pin connector for wide drives,
differential wasn't even intrinsically more expensive, the cost difference
was purely from volume issues, the world having chosen the wrong standard.
> Take a look at the Storageworks UltraSCSI shelves. The trick used is
> to have a isolator chip between the host-shelve cable and the
> actual shelf backplane that houses the disks. The disk / backplane
> part is approx 90 cm's long. But the isolator chip ensures you
> keep sound signals, even with the electrically undesirable bunch of
> stubs (the disk to backplane cabling) at the end of the bus.
There are fundamental theoretical reasons why SE doesn't work well.
If someone doesn't know what they are, and would like to know, I would be
happy to fill them in off the list. :-) :-)
Note that the NONE of the new low voltage interfaces even have a SE
spec: they are ALWAYS differential.