From owner-freebsd-hackers  Thu Jun 19 16:07:21 1997
Return-Path: <owner-hackers>
Received: (from root@localhost)
          by hub.freebsd.org (8.8.5/8.8.5) id QAA25501
          for hackers-outgoing; Thu, 19 Jun 1997 16:07:21 -0700 (PDT)
Received: from friley01.res.iastate.edu (friley01.res.iastate.edu [129.186.189.1])
          by hub.freebsd.org (8.8.5/8.8.5) with ESMTP id QAA25496
          for <hackers@FreeBSD.ORG>; Thu, 19 Jun 1997 16:07:17 -0700 (PDT)
Received: from friley01.res.iastate.edu (loopback [127.0.0.1])
	by friley01.res.iastate.edu (8.8.5/8.8.5) with ESMTP id SAA13293;
	Thu, 19 Jun 1997 18:08:44 -0500 (CDT)
Message-Id: <199706192308.SAA13293@friley01.res.iastate.edu>
X-Mailer: exmh version 2.0gamma 1/27/96
To: dg@root.com
cc: sthaug@nethelp.no, hackers@FreeBSD.ORG
Subject: Re: Network concurrency problems!? 
In-reply-to: Your message of Thu, 19 Jun 1997 04:33:46 -0700.
             <199706191133.EAA03461@implode.root.com> 
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Date: Thu, 19 Jun 1997 18:08:44 -0500
From: Chris Csanady <ccsanady@friley01.res.iastate.edu>
Sender: owner-hackers@FreeBSD.ORG
X-Loop: FreeBSD.org
Precedence: bulk


>>I can't remember seeing a direct statement that the Pro 100/B driver is
>>simpler due to the design of the Pro 100/B. Can anybody help me here?
>
>   I might have said that...I don't recall. In any case, it's part of the
>reason; the other part is that I simply was very careful about coding and
>spent a great deal of time optimizing the critical paths. There shouldn't
>be much difference in total throughput, but the Pro/100B under FreeBSD
>consumes much less CPU time compared to the DEC chip cards. My vauge
>recollection is that the interrupt time was roughly half. It's entirely
>possible that Matt has improved the performance of his driver since I
>did the tests, however.

For what its worth, we are seeing virtually no difference in performance
between the dec and intel cards with Matt's latest driver.  However, the
intel card does seem to be a lot more efficent.

BTW, does the intel card use DBDMA?  From the driver, it looks like
something similar.  How about the dec?

Also, this is sortof an aside, but I am writing a new driver, and was
wondering about using mem mapped vs io mapped registers.  How much of
a difference in performance is there?  Or should I just do a lot of outl()'s?
(like in linux.. but its really ugly imho..)

The memory/PCI busses are already a significant bottleneck, so I dont think
I can afford to sacrifice too much..

Chris Csanady

>
>-DG
>
>David Greenman
>Core-team/Principal Architect, The FreeBSD Project