From owner-freebsd-hackers  Mon Dec  2 18:10:15 1996
Return-Path: owner-hackers
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.5/8.7.3) id SAA29638
          for hackers-outgoing; Mon, 2 Dec 1996 18:10:15 -0800 (PST)
Received: from dyson.iquest.net ([198.70.144.127])
          by freefall.freebsd.org (8.7.5/8.7.3) with ESMTP id SAA29633;
          Mon, 2 Dec 1996 18:10:10 -0800 (PST)
Received: (from root@localhost) by dyson.iquest.net (8.8.2/8.6.9) id VAA01252; Mon, 2 Dec 1996 21:09:46 -0500 (EST)
From: "John S. Dyson" <toor@dyson.iquest.net>
Message-Id: <199612030209.VAA01252@dyson.iquest.net>
Subject: Re: TCP/IP bandwidth bragging
To: davem@jenolan.rutgers.edu (David S. Miller)
Date: Mon, 2 Dec 1996 21:09:42 -0500 (EST)
Cc: dyson@freebsd.org, dennis@etinc.com, kpneal@pobox.com, hackers@freebsd.org
In-Reply-To: <199612030133.UAA18131@jenolan.caipgeneral> from "David S. Miller" at Dec 2, 96 08:33:44 pm
X-Mailer: ELM [version 2.4 PL24 ME8]
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Sender: owner-hackers@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk

> 
> You can say whatever you want.  And whats more, I am told often by
> disgruntled Solaris performance engineers that lmbench is "bush
> league", that is perfectly fine with me.  My response is, if it is so
> bush leage, why is it so difficult for these systems to get better
> numbers than Linux?  Stay down.
>
lmbench measures a specific data point.  "Bush league" is not descriptive
and just as silly as incompletely specified performance numbers.  I don't
agree with namecalling as a valid criticism of a benchmark.  Note that
my criticism is not condemning.

> 
> And watch out, I have gigabit ethernet and FDDI coming very soon as
> well.  SGI cannot even touch my bandwidth and latencies over 100baseT.
> 
I really don't care how fast SGI, Linux, etc are.  FreeBSD generally maxes
out hardware also.  I do care about integrity.  I really have few complaints
about your bragging -- I do reserve the right (and anyone intellectually
honest would agree) to ask for the benchmark and interpret what it really
measures.  The lmbench TCP/UDP benchmarks are pretty much single connection
(or perhaps two) single threaded benchmarks...  So what?

Lmbench is NOT bush league, the results just need to be interpreted.

BTW, FreeBSD (x86) bw_pipe measures approx 200-230MBytes per second on my
machine, beat that!!!  In fact bw_mem_rd measures 600+ MBytes/sec on my
machine -- beat that!!!  I know what the benchmark measures, and it really
doesn't mean that much (IMO), but my statement is meant to illustrate
the fact that the benchmark results need to be interpreted.  I know what
bw_tcp/lat_tcp/lat_udp, etc... measure also.  I also claim that they do
not mean enough by themselves to judge the suitability of an OS (or a
(TCP|UDP)/IP) stack to a task.

John