Subject: Re: NetBSD/i386 and single board computers
To: Greg A. Woods <port-i386@NetBSD.ORG,>
From: Matthias Drochner <>
List: port-i386
Date: 12/06/1997 17:35:43
Excerpts from netbsd: 5-Dec-97 Re: NetBSD/i386 and single .. Greg A.
Woods@most.weird (2614)

> Good to hear about that Teknor board -- that's one of the ones I've been
> looking at.  Have you worked with any of their CompactPCI boards yet?

No, my systems (one 3U, one 6U) are both by "or"
(german company).

The PICMG Teknor board works well in general.
One weak point: It has the Intel 82557 as ethernet
chip. One doesn't get documentation for it, and
diskless boot didn't work for me. (The autonegotiation
was "too smart".)
I'd fear that the same chip (or perhaps the 82558
which is not better wrt documentation) is used on the
CompactPCI boards as well.
Their www-accessible documentation doesn't
tell much...

The Pro-Log CompactPCI systems looked well
for me. (I would have bought a 3U system from
them if I had found a german distributor.)
I've got a couple of their PICMG boards, sold
by DEC (yes, I have too much money...), and they
are more robust than all other boards I've seen.

In any case, after my bad experiences I'd make
sure that the external PCI bus is decoupled by a bridge.

> Note that many of those "industrial" systems are now looking like
> embedded flat-panel systems which can be programmed just like PCs, so
> having a home-PC style bios on them isn't too surprising....

I was more referring to "slot-CPUs". In my application
(particle physics experiments) they are only accessible
via 30m of cable at runtime.
Because the most common maintanance operation for PCs
consists in power-cycling we are making the power
supplies controllable via a fieldbus (Profibus).

> which is why I suspect that many PC
> motherboards only have two or three PCI slots -- they just can't design
> in enough quality to make any more slots work reliably.

Well, then it is close to miracle how a CompactPCI
can work at all. (I know, they have simulated it. But
I'm curious what happens if I put such a system
in my experiment's frontend area which suffers
from strong magnetic fields, radio frequency senders
and fast signal slopes.)

> I think in the end the VME320 stuff will be much more reliable and give
> similar performance, but for now it's very proprietary.

I'm not yet convinced that the new VME standards
(64X and 320) will be used very widespreat.
There are probably compatibility problems, and the
2 additional connector rows are not a great idea imho
(they should have taken a completely new connector).
And there are too many decition makers who follow
everybody who says "industry standard" or even
"Windows NT compatible".

> The industrial Dec Alpha's are just too expensive, and most everything
> else would require more work to get netbsd running on than I can afford
> to spend right now.

I'll try to get an alpha PICMG board next year. Without
DEC UNIX, the list price is $3995 for 366MHz.
(Don't know yet how to get SRM firmware.)
The hardware is close to existing systems and well
documented, so a NetBSD port should be straightforward.
(It is not even unreasonable: If I want more CPU power
in a frontend system the choices are limited. Pentium Pro
is "dead end", and useless without a wellperforming chipset.
Pentium 2 is no option for mechanical reasons - the CPU
will be in the way.)

best regards