Subject: Re: 1.6.1 on dec5000/200 questions
To: Aaron J. Grier <agrier@poofygoof.com>
From: Maciej W. Rozycki <macro@ds2.pg.gda.pl>
List: port-pmax
Date: 05/27/2003 14:59:42
On Mon, 26 May 2003, Aaron J. Grier wrote:

> > > fza (DEC FDDIcontroller 700 (DEFZA; fiber optic)) at tc0 slot 5 offset
> > > 0x0 not configured
> > > even if it's PIO, it's still got to be better than 10mbit ethernet.
> > Well, give me docs and time and I may come up with a driver. (If I don't
> > get distracted to much by my PPC toys...) 
> 
> Matt Thomas stated he had the docs a few years back and I think Maciej
> Rozycki and Jason Thorpe have mentioned something about it too.

 I think you've got it backwards about me -- I'm looking for docs and not
having ones. :-/  OTOH, I have some information that let me made an
initial attempt for a Linux driver (hack?) and I can help making one for
NetBSD, too. 

> it may not be able to go the full 100mbit, but it's still got to be
> faster than 10mbit ethernet.

 The difference in throughput is noticeable, especially when being able to
use the full MTU (e.g. transmitting directly between FDDI nodes).  The PIO
transmits are hogging the system, indeed, though it may probably be a bit
improved with write-back caches like these of R4k DECstations and Alphas
-- cache invalidation is faster than using uncached stores, especially
given cache lines are typicaly wider then 32 bits, e.g. the /260 has it
configured to 128 bits by the firmware and can optionally be set up to 256
bits.  I've been able to get up to 40Mbps real throughput without trying
to optimize the code, e.g. trying to play tricks with caches.

 The FZA is nice for FDDI diagnostics -- it lets you handle all the SMT
traffic it receives and transmits. 

-- 
+  Maciej W. Rozycki, Technical University of Gdansk, Poland   +
+--------------------------------------------------------------+
+        e-mail: macro@ds2.pg.gda.pl, PGP key available        +