[ale] Linux performance vs IRIX performance (fwd)

[Vernard Martin]

I saw this on the linux-kernel mailing list and thought that it would be of
general interest to the list since the recent announcement of SGI port of linux.

"Larry McVoy writes:"
Message-Id: <199807171840.LAA05852 at bitmover.com>
To: linux-kernel at vger.rutgers.edu
Subject: Linux performance vs IRIX performance
 From: lm at bitmover.com (Larry McVoy)
Date: 	Fri, 17 Jul 1998 11:40:23 -0600
X-Orcpt: rfc822;linux-kernel at vger.rutgers.edu
Sender: owner-linux-kernel at vger.rutgers.edu
Precedence: bulk
X-Loop: majordomo at vger.rutgers.edu

Hi folks,
	I thought you'd like this.  I have a regression test (1700 line
shell script) for a source code control system I'm building.  When I run
that test on an AMDK6 at 300 + 1MB L2$ + 100Mhz bus + SDRAM, it takes

	2.82user 3.40system 0:06.37elapsed

When I run the same thing on a SGI MP system, with 195Mhz R10K's + 4MB of
L2$, I get 

	1.357user 10.379sys 13.284elapsed

The interesting thing is that the user code ran almost twice as fast on
the SGI processor (not bad for a processer at 2/3 the Mhz).  That's the
good part.  The bad part is that SGI OS spent 3 times as long doing
exactly the same work.  The work is creating and deleting a bunch of
files, with some small amount of processing on the files.

So what does this mean?  What it means is that Linus' focus on latency,
while it seems pedantic at times to argue about one more cache miss or
one more function call in the code path, actually pays off in a big way.
If that SGI box were running Linux and we could assume the same ratios,
then the numbers should be about

	1.3user 1.6sys 3.0elapsed

A factor of 4.4 difference.  It's not fair to actually expect that good of
a result - the SGI OS is an SMP OS that scales up to approximately 128 
processors, has all sorts of useful and not so useful features that Linux
doesn't have, etc, etc.  None the less, it is likely that Linux on the same
hardware would be about 3 times faster than IRIX.  

I think that is pretty cool and that maybe you wanted to know that the
philosophy actually works and works well.  Kudos to Linus & team for
all the great work.

--lm

P.S.  The Linux was 2.0.34 - there are some changes in 2.1 that should make
this even better...

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo at vger.rutgers.edu
Please read the FAQ at http://www.altern.org/andrebalsa/doc/lkml-faq.html

--------------------------------------------------------------------------------
Vernard Martin (vernard at cc.gatech.edu)  | 2nd Atlanta Linux Showcase!
http://www.cc.gatech.edu/~vernard/      | October 23 - 24, 1998
Georgia Tech, College of Computing      | http://www.ale/org/showcase