[NCLUG] July 1, 2003 talk - Linux System Performance

[jbass at dmsd.com]

Actually, due to a counter overflow problem, the sources this morning:

	 Thu Jun 26 06:36 327/12259 "Re: [NCLUG] July 1, 2003 talk - Linux System Performance"

is the most current, but looking at some members numbers, it still looks
like there is a pesky overflow problem. I don't have a fast machine to
test against, so it's been a little troublesome tracking down unexpected
promotion to int's in gcc. Hate to have to resort to reading asm again :(

In the mean time, I might suggest running the test with an arg of between 2-5
on very fast machines to keep the counter size managable. ... "./mbench 2"

I noticed that some people are actually looking at the numbers and scratching
heads .... this is a good thing :)  As we collect data from a dozen or two
different types of machines it will become obvious that IA32 machines are not
created equal, with significantly different performance variations due to
cpu and chipset integration tradeoffs between the cache design, fsb, SMP
cache management, and the basic memory subsystem design + optimizations.

See ya next week, it's starting to look like a fun talk.

Have fun!!
John


Benson Chow <blc at q.dyndns.org> writes:
> It's in John's message/first message of this thread, also I have a copy
> at http://q.dyndns.org/~blc/mbench/mbench.c (too lazy to move it
> around... I even left my binaries in that directory too...)

> On Thu, 26 Jun 2003, Jim Wildman wrote:

> > Date: Thu, 26 Jun 2003 13:55:56 -0400 (EDT)
> > From: Jim Wildman <jim at rossberry.com>
> > Reply-To: nclug at nclug.org
> > To: nclug at nclug.org
> > Subject: Re: [NCLUG] July 1, 2003 talk - Linux System Performance
> >
> > Where do you get mbench?  I'll run it on some server class hardware.
> >
> > ------------------------------------------------------------------------
> > Jim Wildman, CISSP, RHCE                                jim at rossberry.com
> > http://www.rossberry.com