*BSD News Article 87852


Return to BSD News archive

Path: euryale.cc.adfa.oz.au!newshost.carno.net.au!harbinger.cc.monash.edu.au!munnari.OZ.AU!news.ecn.uoknor.edu!feed1.news.erols.com!tezcat!cam-news-hub1.bbnplanet.com!news.bbnplanet.com!su-news-hub1.bbnplanet.com!arclight.uoregon.edu!super.zippo.com!zdc!szdc!szdc-e!news
From: "John S. Dyson" <dyson@freebsd.org>
Newsgroups: comp.os.linux.misc,comp.os.linux.networking,comp.os.linux.setup,comp.unix.bsd.misc,comp.os.ms-windows.nt.advocacy,comp.os.os2.advocacy
Subject: Re: Linux vs whatever
Date: Thu, 30 Jan 1997 11:10:46 -0500
Organization: John S. Dyson's home machine
Lines: 49
Message-ID: <32F0C806.167EB0E7@freebsd.org>
References: <32DFFEAB.7704@usa.net> <5cm5h3$bfg@lynx.dac.neu.edu> <5cmv5q$k6d@cynic.portal.ca> <5co7vd$lvt@lynx.dac.neu.edu> <5coib1$jvv@cynic.portal.ca> <5cqe32$d22@halon.vggas.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
X-Mailer: Mozilla 3.01 (X11; I; FreeBSD 3.0-CURRENT i386)
Xref: euryale.cc.adfa.oz.au comp.os.linux.misc:155207 comp.os.linux.networking:66704 comp.os.linux.setup:94661 comp.unix.bsd.misc:2143 comp.os.ms-windows.nt.advocacy:51412 comp.os.os2.advocacy:264482

James Youngman wrote:
> 
> In article <5coib1$jvv@cynic.portal.ca>, cjs@cynic.portal.ca says...
> 
> >Two years ago it did not. I've not done any testing recently or
> >seen any tests that I would consider more than tentative, but from
> >everything I'm still hearing, Linux networking still does not
> >perform as well as BSD networking. However, this is not an issue
> >that I really care to debate; so if you wish to assert that Linux
> >networking is just as good without some solid testing to support
> >you, I'm not going to refute that.
> 
> Let's not get carried away here.   Even BSD networking is not the best.  A few
> years ago VJ did a reimplementation of the BSD stack that entirely did away
> with mbufs and (IIRC) nearly doubled its speed.   Unfortunately this
> implementation isn't widely available (at least, I've never seen it).
> 
You have brought up an interesting point, and I would like to
expand on it.  I think that we are many times too quick to run
a simple benchmark and infer too much from it.  It is amazing
to me that someone will run a tcp(latency|transfer rate) benchmark
and conclude that OS1 networking is 10% faster than OS2 networking.

Frankly, a 10% difference running an LL benchmark is most of the
time overshadowed by other performance aspects of the OS.  Bragging
rights should be reserved for measurable true (real application)
performance differences :-).  Speed isn't the only
measure of quality though, and the end users who stress the systems
should carefully qualify their OS (esp if free.)  (Take a look
at the async vs. sync metadata -- it is a design decision, perf
vs. reliability.)

So what if Linux networking is 10% faster/slower (even 20%) in
benchmarks?  What tradeoffs and design decisions are accompanied
with the speed difference?

It used to be that Linux networking was ALOT slower and much less
standards compliant than *BSD's.  It IS better than it used to be,
and in some ways might be faster/slower than BSD, but is not twice
as fast, and is probably not twice as slow.  Considering that it
is a reimplementation (learning from BSDs mistakes), and probably
performs approximately (+-10%-20%ish) the same in many areas
(speed-wise), indicates that it might be harder to create a fully
functional networking stack that is TWICE as fast as *BSDs than
first glance implies.

John
dyson@freebsd.org