*BSD News Article 92617


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!super.zippo.com!zdc!szdc!news
From: "John S. Dyson" <dyson@freebsd.org>
Newsgroups: comp.unix.bsd.freebsd.misc,comp.unix.bsd.bsdi.misc,comp.sys.sgi.misc
Subject: Re: no such thing as a "general user community"
Date: Wed, 02 Apr 1997 08:09:44 -0500
Organization: John S. Dyson's home machine
Lines: 40
Message-ID: <33425A98.167EB0E7@freebsd.org>
References: <331BB7DD.28EC@net5.net> <5hnam9$393@hoopoe.psc.edu> <5hp7p3$1qb@fido.asd.sgi.com> <33402E4E.6937@indy.celebration.net> <5hso11$4ji@fido.asd.sgi.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.unix.bsd.freebsd.misc:38307 comp.unix.bsd.bsdi.misc:6552 comp.sys.sgi.misc:29619

Steve Alexander wrote:
> 
> In article <33402E4E.6937@indy.celebration.net>,
> John S. Dyson <dyson@indy.celebration.net> wrote:
> >A read/write system call is not null unless it is special cased.
> 
> I agree that this test is not well named, but I have to side with Larry in that
> it is a much more representative measure than say, getpid().
>
My argument isn't that the measurement isn't relevent, it is that
the LMBENCH suite doesn't always measure what it implies.  Additionally,
it requires understanding what it is really doing, and it's limitations
in order to interpret it's results.

For me, it is a very valuable tool, but it is difficult to draw
conclusions about real world behavior from it, without understanding
what it is.  I think that the community is being done a disservice
when even the numbers are posted, without a disclaimer that the
LL benches are limited in scope.  BOGOMIPS is also a LL benchmark,
and of course LMBENCH is much more comprehensive and complete (and
frankly accurate), but one needs to recognize that the benchmarks
are of the same kind.  (I know, I know, BOGOMIPS is practically
a CPU only benchmark, but the question that I am posing is "how
does it relate to application performance?")

So, when you say "more representative measure", I am going to ask,
"measure of what?"  Don't get me wrong, as a developemnt tool and
as a measure of the efficiency of non-scaled kernel performance
or light load kernel performance, it is a valuable tool.  Note
that it is difficult from the benchmark results themselves to
determine if the kernel performance will have any impact on
application performance.  If OS(1) is two times faster at a
given operation than OS(2) on a single application, non
scaled operation, and both of the LL performance figures are
in the very low percents of overhead, then what value are the
results for those who are running applications, when the higher
load results are vastly different?

John