Subject: Re: RAIDFrame problem
To: Matt Dainty <matt@bodgit-n-scarper.com>
From: Greg A. Woods <woods@weird.com>
List: port-alpha
Date: 07/05/2003 20:02:09
[ On Sunday, July 6, 2003 at 00:01:16 (+0100), Matt Dainty wrote: ]
> Subject: Re: RAIDFrame problem
>
> I understand the disks won't be using LVD as they're on a UW bus, but
> I've never read anything that says I can't do that, without accepting
> that the disks will slow down to the speed of the adapter.

It's not just the speed which changes -- it's the termination
requirements and signalling which change as well.

> I also read
> somewhere that due to the spec of LVD, for whatever reasons, the
> termination must/should be done on the cable, hence none of the LVD
> disks I own have TERM jumpers on them anywhere.

I think that's just a general update of the overall specifications and
not something specific to LVD.  It was silly to put termination
electronics on every target device (though it must have been cheap
enough for the manufacturers to do or they wouldn't have done it :-).

> In all of this, I've been using a properly terminated cable, just
> initially, this was a 1.5m long variant, which I thought would be okay
> as before LVD, the maxmimum bus length is 2m? So I thought I'd still be
> inside the constraints of UW.

It should be OK, but there are some other critical factors that are
important to SE Ultra (SCSI-3 aka Fast20) buses, including the required
separation between connectors (30cm for SCSI-2 and newer buses).

Maximum SE (SCSI-3) total bus length is 1.5m (and with 30cm spacing
between stubs that limits you to just five devices, including the host
adapter!).

Maximum LVD bus length is a full 12 meters, provided that there are
_only_ LVD devices on the bus.  (and HVD max len is 25m :-)

See:

	http://www.scsifaq.org/

and especially:

	http://scsifaq.org:9080/scsi_faq/scsi_quick_start.html

See also:

	http://www.ba-stuttgart.de/~schulte/htme/ebuss12.htm
	http://serv.dr.qc.ca/scsi.htm

> Your comments regarding signal noise, etc. prompted me to try and use a
> shorter cable, but it still retains a terminator on the end, just the
> cable length is reduced by about half.

The question remains:  what kind of terminator are you using?

I would expect all LVD internal cables to have SE compatible termination
as well (also called "multimode LVD/SE terminators"), but I wouldn't
want to bet on it.

If you can find a normal non-terminated internal 68-pin cable with at
least 30cm between the connectors you use, and a female 68-pin "Active"
or "FPT" (aka "negation") terminator, then you might try them in
preference to the LVD cable.

The LVD cable and connectors should be fine, electrically, but if you
can't confirm that its terminator is a full multimode LVD/SE terminator
(i.e. will work properly for SE/UW) then you want to avoid using it on
the 2940UW.

> My card has both an external and internal wide channel, and an internal
> narrow channel. I can use any two of the three, so I have just those two
> disks on the internal wide channel, and a TLZ09 DAT on the narrow
> channel, properly jumpered for termination.

Pull all the narrow devices off the bus -- they may cause problems.

SE/UW buses are not really happy to have multiple different speed
devices connected (at least not if there's traffic for both interleaved
on the bus) and they are definitely not happy if you have both narrow
and wide devices on them.

Separating target devices with different attachment speeds and different
widths onto separate buses should always be one of the first changes to
try when diagnosing SCSI bus problems.

I've seen lots of problems with Adaptec cards when the card is the end
of the wide bus and the narrow bus extends beyond the card (e.g. on the
50-pin connector).  I have similar problems with an Asus (Symbios Logic
53c875) card -- I can't get anywhere near the full SE/UW bus length when
my tape library is also connected.  I believe the issue is with the card
not doing the high-bit termination well enough on on wide bus (though
I've got to try a shorter jumper cable on the disk shelf I'm testing
before I'll know for certain about the 53c875 card).

> I can't access the cards BIOS from the Alpha, but I know from having it
> in a PC previously, it was set to 'auto'.

That should work fine then as long as you remove all the narrow devices,
but it may cause problems if you do not pull the narrow devices.

-- 
								Greg A. Woods

+1 416 218-0098;            <g.a.woods@ieee.org>;           <woods@robohack.ca>
Planix, Inc. <woods@planix.com>; VE3TCP; Secrets of the Weird <woods@weird.com>