From owner-freebsd-hackers  Sat Sep  2 08:59:18 1995
Return-Path: hackers-owner
Received: (from majordom@localhost)
          by freefall.FreeBSD.org (8.6.11/8.6.6) id IAA26574
          for hackers-outgoing; Sat, 2 Sep 1995 08:59:18 -0700
Received: from mail.htp.com (mail.htp.com [199.171.4.2])
          by freefall.FreeBSD.org (8.6.11/8.6.6) with ESMTP id IAA26565
          for <hackers@freebsd.org>; Sat, 2 Sep 1995 08:59:17 -0700
Received: from et.htp.com (et.htp.com [199.171.4.228]) by mail.htp.com (8.6.5/8.6.5) with SMTP id LAA22490; Sat, 2 Sep 1995 11:59:34 -0400
Date: Sat, 2 Sep 1995 11:59:34 -0400
Message-Id: <199509021559.LAA22490@mail.htp.com>
X-Sender: dennis@mail.htp.com
X-Mailer: Windows Eudora Version 2.0.3
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
To: Bruce Evans <bde@zeta.org.au>
From: dennis@et.htp.com (dennis)
Subject: Re: novell ne2000 cards with FreeBSD...opinions?
Cc: hackers@freebsd.org
Sender: hackers-owner@freebsd.org
Precedence: bulk

>>Yeah, I just looked at the driver and saw the programmed IO cruft in there
>>to deal with the novell cards.  Bummer.  Someone mentioned that some of the
>>cards really do have shared memory and can run just like the SMC and 3COM.
>>Do you know anything about this?
>
>Shared memory on my NE2000/WD8013EBT card is 50% faster, but that isn't
>saying much.  The driver spends half its time bcopying to the shared
>memory instead of more that half its time doing explicit i/o instructions.
>The speeds are something like:
>
>	i/o instructions:	2MB/sec
>	bcopy:			3MB/sec
>	IDE disk controllers:	3.3MB/sec	PIO mode 0 (or 1?) spec
>	8MHz ISA bus max:	4MB/sec
>
The bus transfer itself is a pretty small piece of the latency puzzle. But
with a PIO card you have setup operations for every transfer...whereas with
shared memory you can access individual bytes and the buffer without setup
latency.

db