Subject: Re: Dual SCSI, single chain?
To: Curt Sampson <cjs@cynic.net>
From: Matthew Jacob <mjacob@feral.com>
List: port-alpha
Date: 03/28/1999 16:46:54
You can't mix LVD chains and SE or Differential chains.

LVD, Single Ended, and Differential are all voltage and nominal current 
levels. You may not mix these. The interesting property of LVD capable
devices is some of them (not all- disks usually, but not all HBAs) can
sense the voltage levels involved and know whether they're connected to a
LVD, SE or Differential bus. That doesn't mean LVD busses can be mixed and
matched- it just means that some LVD devices can attach to multiple
busses.

Data transfer speeds such as FAST-10, FAST-20 (Ultra), FAST-40 (Ultra2) 
(and soon to be FAST-80 (Ultra-3), in theory, may in fact be intermixed as
the speeds only apply during DATA IN or DATA OUT phases and are negotiated
per-target. This is further complicated by the fact that Ultra2 requires
and LVD bus to work at all. In practice, however, it is foolish to mix
Ultra/Ultra2 devices with a bus that contains older (e.g., < FAST-10) 
devices due to the extreme likelihood of crappy signal reflection caused
by designs that were not aware of these kinds of issues.

This is also complicated by mixing wide and narrow bus widths and some of
the signal termination issues this raises.

In practice, I lay out SCSI busses as follows:

	Differential Only

	LVD/Ultra2 Only

	Ultra - <= 4 disks only. No mixing of internal or external cables.
	No mixing of Wide/Narrow.

	All others. fine.

As usual, your mileage may vary. This is not an entirely exact science. I
have configurations with mixed and matched cabling in FAST-10 mode and
with wide and narrow segments that *only* works when I leave the narrow
stub end of *that configuration only* unterminated. But I also don't make
this a production environment.

(this is Ross' cue to trot out his *much* superior h/w knowledge to
correct my mail....)


-matt

On Sun, 28 Mar 1999, Curt Sampson wrote:

> On Sun, 28 Mar 1999, Greg A. Woods wrote:
> 
> > Also, unless you're using full differential, or LVD, there are many
> > restrictions on the mixing of "slow", "fast", and "ultra-fast" devices
> > on the same bus...
> 
> No, this is certainly not the case. You can mix any combination of
> stuff you want (excepting differential)--narrow, wide, slow, fast,
> ultra, etc., so long as you don't mind the bus contention (i.e.,
> a 4 MHz CD-ROM drive transferring 1 MB/sec is going to use a quarter
> of the bus bandwidth regardless of the width and speed of the
> controller and other devices on the bus). 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.
> 
> Given that, yes, you probably do want to segregate your drives on
> to different buses to avoid performance limitations, if you've got
> low-performance stuff on the bus that you'll actually be using. (I
> keep an old 2X CD-ROM on the same bus as my pair of fast new
> Barracudas, but I almost never use it, so it doesn't bother me.)
> 
> > ...and given that most "narrow" devices are always either
> > "slow" or at best "fast"...
> 
> 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).
> 
> 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
>