*BSD News Article 73851


Return to BSD News archive

Path: euryale.cc.adfa.oz.au!newshost.anu.edu.au!harbinger.cc.monash.edu.au!munnari.OZ.AU!news.mel.connect.com.au!news.mira.net.au!vic.news.telstra.net!act.news.telstra.net!imci3!newsfeed.internetmci.com!info.ucla.edu!nnrp.info.ucla.edu!galaxy.ucr.edu!cs.ucr.edu!grif
From: grif@cs.ucr.edu (Michael Griffith)
Newsgroups: comp.os.linux.networking,comp.unix.bsd.netbsd.misc,comp.unix.bsd.freebsd.misc
Subject: Re: TCP latency
Date: 16 Jul 1996 03:15:28 GMT
Organization: UC Riverside, Dept. of Computer Science
Lines: 40
Message-ID: <4sf1gg$ebs@galaxy.ucr.edu>
References: <4paedl$4bm@engnews2.Eng.Sun.COM> <31E3D9E2.41C67EA6@dyson.iquest.net> <4s5bl2$qpg@linux.cs.Helsinki.FI> <31E664EB.167EB0E7@inuxs.att.com>
NNTP-Posting-Host: cs.ucr.edu
Xref: euryale.cc.adfa.oz.au comp.os.linux.networking:45361 comp.unix.bsd.netbsd.misc:4068 comp.unix.bsd.freebsd.misc:23704

In article <31E664EB.167EB0E7@inuxs.att.com>,
John S. Dyson <dyson@inuxs.att.com> wrote:
>Geesh, do you understand that your example tests only three connections
>to the same machine?  You are not showing scalability at all.  (Mostly
>you are showing that you arent' busting the cache.)  The scalability
>issues on the old Linux context switch didn't come into effect until
>about 20processes did it?  Herein, you are showing that the localhost
>code under very little if NO load runs the same speed (at least to me.)
>But you STILL are not addressing the issue of scalability (especially
>to/from multiple TCP/IP addresses.)

Scalability is a desirable goal, but as operating system designers, we
need to keep in mind that one should focus on making the general case
fastest.  The general case for many metrics happens to be the unloaded
case.  It is also the easiest to measure and analyze.

For example, the run queue has way less than 20 processes in the
common case.

If we want to make the general case fast, we need to benchmark that
general case.  I think that lmbench does a pretty good job at doing
that.

If someone is interested is measuring less common, but interesting
situations (like a highly loaded server), then new benchmarks will
need to be written.  Once the benchmark numbers have been released,
the OS designers will again hit the drawing board trying to be the
fastest.  Until then, predicting who will be fastest is pointless.  So
is suggesting than interesting, but highly rare, environments are the
only ones someone should care about.

To my way of thinking, the competition between free operating systems
can be quite useful.  After all, none of the commercial UN*X vendors are
much fun to challenge anymore.  I'm just sad to see the competition take
such an ugly turn.  I'm sure that none of you would say in person what
you have written in Usenet.
-- 
Michael A. Griffith (grif@cs.ucr.edu) | http://www.cs.ucr.edu/~grif/
Department of Computer Science        | "My freedom of speech implies
University of California, Riverside   | your freedom to be offended."