[SunRescue] NetBSD 1.4 miniroot problem fixed

[Greg A. Woods]

[ On Wednesday, December 8, 1999 at 18:23:32 (+0000), Martin Frost wrote: ]
> Subject: Re: [SunRescue] NetBSD 1.4 miniroot problem fixed
>
> IMHO, the reason SCSI gets a bad reputation is that it's been
> the only bus available for external disk connections for some
> time. As many people have noted in this thread already,
> strange reflections and interference cause all sorts of
> problems with fast buses (and remember that modern SCSI
> is clocked at up to 40MHz: that's really quite fast for a
> bus that can be several feet long). All the other buses just
> give up and support only one or two devices on a much shorter
> bus, which is a lot easier.

Oddly enough most of the percieved problems with SCSI are actually due
to the fact that two options were made available in the standard for
SCSI buses:  single-ended, and differential.  Unfortunately despite the
rather grave warnings in the standard itself about the extremely
unreliable and persnickety nature of single-ended buses, many
manufacturers chose to avoid the cost of a couple more line driver chips
(which would have been relatively low in the long run) and went with it
anyway.

Differential buses are of course much more reliable and can withstand
more noise, go longer distances, and so on.  If differential had have
been the only "option", SCSI would have been much more reliable, right
from day one (and only slightly more expensive, especially in contrast
to its already inflated price over some of the other common commodity
devices such as IDE/EIDE disks in the PC world).

I would most strongly recommend that anyone even considering playing
with SCSI (i.e. either an already working system, or building something
out of scrounged parts that were not explicily validated as compatible
by the OEM, esp. with Sun gear) should read, and re-read, and read
again, the relevant bits of the SCSI FAQ:  <URL:http://www.scsifaq.org/>

Of course if you stick to one vendors gear, and always only use their
components (cable, connectors, terminators, etc.), and always only in
OEM configurations (i.e. never do anything unless their manuals say you
can), then you probably won't encounter too many problems.  This
definitely kills your "plug'n'play" ability though!  ;-)

> Who was it who said that `SCSI is the second worst IO bus
> ever invented, but all others tie for first place'?

SMD, which used a differential bus design, was OK.  IPI was used
successfully for some time too (though I don't know much about its
design).  ST-506, and its decendant ESDI, was fine for internal disks,
but it really couldn't go external without a lot of hassle (AT&T built
some nice but expensive external ESDI disk subsystems), and of course it
wasn't really a "bus" (just the control lines were shared from disk to
disk).  IDE (and EIDE) really are not much more than extensions of the
ISA bus itself and again it is effectively useless for external
devices.

>From what I've seen so far the high-end guys are all going with
FiberChannel disks....

> ObRescue: For all its faults, rescuers of old hardware
> owe a great debt to SCSI: it means we can still buy disks
> that will (in general) still work with our old systems
> (since disks and tape drives are the least reliable parts
> due to their moving parts, this is a Good Thing).

This advantage will not last for long, though there will be a vast array
of standard SCSI devices available in surplus for a long time and indeed
they'll often have many years of life in them even then.

-- 
							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>