On Mon, 29 Mar 1999, Greg A. Woods wrote:

> ...you *cannot* just mix
> things arbitrarily.  There are issues w.r.t. whether or not you've got
> the devices in the correct order on the bus....

You can't attach two wide devices together with a piece of narrow
cable and expect them to negotiation wide transactions, no. But
they will still work, at least.

> ...whether or not you're
> meeting the minimum cabling and connector standards for the highest
> negotiated speed device on the bus...

Right. You can't use bad cables and expect it to work. This is just
as true if you have only one device on the bus.

> whether or not the upper and lower
> halves of the bus are properly terminated

Which also has nothing to do with how many devices are on the bus.

>, whether or not the attached
> devices properly follow the specifications for option negotiation

In other words, broken devices don't work very well.

> and
> finally whether or not you can properly program the host adapter to
> control the various options on a per-device level.

And broken controllers don't work very well.

Really, Greg, none of this is a big surprise. And all of this
applies even if you have only one device and one controller on the
bus. It's nothing specific to with mixing multiple devices.

> Even once you got
> everything right in theory there are some fairly tricky timing issues,
> not to mention that SCSI-1 and SCSI-2 devices which barely meet their
> own requirements are unlikely to meet the more stringent requirements
> necessary for Ultra and Ultra-2 speeds (even if they're off at the
> "narrow" end of a bus).

But they won't be using the bus during the sync. phase of the Ultra
or Ultra-2 devices. That's the whole point of SCSI backwards
compatability, and why it works.

> As Matthew Jacob's much more detailed answer suggests, it's much safer
> to keep your slow and "fast" devices on separate buses.

It's always safer to keep different devices on separate buses. And
usually faster. But you're highly unlikely to cause any harm by
putting it all on one bus. And it will most likely all work, if
you have decent cables and termination.

> > In fact, LVD is the only
> > one you can't mix without losing capabilities (LVD devices revert
> > to single-ended if there are any non-LVD devices on the bus), and
> > differential is the only one you can't mix at all.
> 
> LVD (or any "differential signaling") is not equivalent to transfer
> speed.  You're mixing your metaphors!  ;-)

Uh, no I'm not. I said that if you put LVD devices on a single-ended
bus, you lose some capabilities. In this case, they revert to
single-ended, which also means they scale down to Ultra-1 transfer
speeds. What's so hard to understand about that?

> > Again, no. Virtually all but very old narrow disk drives are fast
> > SCSI (10 MHz). A lot of CD-ROM drives appear to be slower, though
> > (4 MHz).
> 
> I've got lots of "slow" SCSI drives (and man are they ever SLOW!).
> They're sure a heck of a lot newer than my SMD drives, but yes, they are
> quite "old".

Well, I suppose it depends on how you define old. Lots of people
consider 350 MB 3600 RPM drives made in '94 or '95 to be old, and
these are invariably Fast-SCSI (10 MHz) drives. I rarely see anything
older than that that is actually terribly useful in a machine.

Anyway, the summary is: give it a go, this stuff can work just fine
at home. No, it's not going to be the sort of ultra-reliable thing
you need for production use, but then again, if you're that concerned,
go buy a DEC Storageworks rather than futzing about with some weird
conglomeration of drives you found in a dumpster somewhere.

cjs
-- 
Curt Sampson  <cjs@cynic.net>   604 801 5335   De gustibus, aut bene aut nihil.
The most widely ported operating system in the world: http://www.netbsd.org