From owner-freebsd-hackers  Mon Jun 26 02:12:49 1995
Return-Path: hackers-owner
Received: (from majordom@localhost)
          by freefall.cdrom.com (8.6.10/8.6.6) id CAA03477
          for hackers-outgoing; Mon, 26 Jun 1995 02:12:49 -0700
Received: from gndrsh.aac.dev.com (gndrsh.aac.dev.com [198.145.92.241])
          by freefall.cdrom.com (8.6.10/8.6.6) with ESMTP id CAA03471
          for <hackers@freebsd.org>; Mon, 26 Jun 1995 02:12:46 -0700
Received: (from rgrimes@localhost) by gndrsh.aac.dev.com (8.6.11/8.6.9) id CAA03217; Mon, 26 Jun 1995 02:12:57 -0700
From: "Rodney W. Grimes" <rgrimes@gndrsh.aac.dev.com>
Message-Id: <199506260912.CAA03217@gndrsh.aac.dev.com>
Subject: Re: FreeBSD as a router
To: rcarter@geli.com (Russell L. Carter)
Date: Mon, 26 Jun 1995 02:12:57 -0700 (PDT)
Cc: hackers@freebsd.org
In-Reply-To: <199506252111.OAA01741@geli.clusternet> from "Russell L. Carter" at Jun 25, 95 02:11:53 pm
X-Mailer: ELM [version 2.4 PL24]
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Content-Length: 1333      
Sender: hackers-owner@freebsd.org
Precedence: bulk

> 
> |> R. Grimes writes.....
> |> 
> |> >The SMC 9332 EtherPower 10/100 is a bus master device with no memory
> |> >on it at all.  It uses host memory for packet buffers and for all
> |> >practical purposes this can be as much memory as you want to through
> |> >at it!
> |> >
> |> As long as you can get the bus. You don't want this on a 100mbs media
> |> device. The card (or controller) needs some buffering or you'll be in
> |> trouble with a heavy burst.
> |
> |I don't think that a 10MByte a second memory demand is going to have
> |much of a problem at all on a 132MB/sec (theroy) or 100MByte/sec (measured)
> |bus like PCI.
> 
> Well, that 100 MBytes/sec was mem to mem, right?  Not going through the
> bus to a device.  For that the best I've ever heard of with the current 
> generation PCI chipsets is 18 MBytes/sec.  If anybody knows of 
> device<->mem faster than 18 MBytes/sec I would appreciate a pointer to it.

18MB/sec is PCI cycles to I/O addresses, it should (and I need to
go measure this on a video card) be 100MB/sec for bus mastered DMA
access for long bursts.  Or for access memory that resides on the
PCI bus.  (Any one seen a PCI memory card yet??)


-- 
Rod Grimes                                      rgrimes@gndrsh.aac.dev.com
Accurate Automation Company                 Reliable computers for FreeBSD