From owner-freebsd-hackers  Fri May 16 13:14:35 1997
Return-Path: <owner-hackers>
Received: (from root@localhost)
          by hub.freebsd.org (8.8.5/8.8.5) id NAA17041
          for hackers-outgoing; Fri, 16 May 1997 13:14:35 -0700 (PDT)
Received: from haldjas.folklore.ee (Haldjas.folklore.ee [193.40.6.121])
          by hub.freebsd.org (8.8.5/8.8.5) with ESMTP id NAA17030
          for <hackers@FreeBSD.ORG>; Fri, 16 May 1997 13:14:14 -0700 (PDT)
Received: from localhost (narvi@localhost)
          by haldjas.folklore.ee (8.8.4/8.8.4) with SMTP
	  id XAA14446; Fri, 16 May 1997 23:07:23 +0300 (EEST)
Date: Fri, 16 May 1997 23:07:22 +0300 (EEST)
From: Narvi <narvi@haldjas.folklore.ee>
To: Christoph Kukulies <kuku@gilberto.physik.rwth-aachen.de>
cc: "David S. Miller" <davem@jenolan.rutgers.edu>, hutton@isi.edu,
        hackers@FreeBSD.ORG
Subject: Re: throughtput measurements for fast ethernet
In-Reply-To: <199705161601.SAA01821@gil.physik.rwth-aachen.de>
Message-ID: <Pine.BSF.3.96.970516230149.14418A-100000@haldjas.folklore.ee>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Sender: owner-hackers@FreeBSD.ORG
X-Loop: FreeBSD.org
Precedence: bulk



On Fri, 16 May 1997, Christoph Kukulies wrote:

> >    From: Christoph Kukulies <kuku@gilberto.physik.RWTH-Aachen.DE>
> >    Date: Fri, 16 May 1997 08:48:13 +0200 (MEST)
> > 
> >    Someone told me some time ago when I was seeking for similar
> >    figures (Garret ?) that FreeBSD can saturate 10/100 Mbit with
> >    appropriate CPU power. The only interesting question would be CPU
> >    utilization during transfer compared to other L-word OSs.
> > 
> > I'd be more interested in seeing FreeBSD get low latencies, but as
> > long as you guys are bzero()'ing a structure on the stack of
> > tcp_input() for every packet that arrives just for T/TCP's sake, it
> > isn't going to happen.
> 
> Interesting.
> 
> a) I don't know how efficient the bzero() is (inline? #idef KERNEL?)
>    over a statementwise zeroing of a 20 byte structure and why this.
> 

I tried it (actrually more). 

After replacing the bzeroing of the tcp options and all the
bzero(&taop,...) in tcp_input.c, tcp_output.c and tcp_usrreq.c with
macros doing by-component zeroing I see 4-5% of improvement in both TCP
latency and throughput on my P75. But it is a kind of hack.

I can post the diffs if anyone is interested.

	Sander

> 
> --
> Chris Christoph P. U. Kukulies kuku@gil.physik.rwth-aachen.de
>