Re: Benchmarks... (fwd)

Adam 'WeirdArms' Wiggins (awiggins@cse.unsw.EDU.AU)
Tue, 4 Nov 1997 23:36:58 +1100 (EST)

Sorry to spam but this is for those interested in the micro kernel
discusion.
cheers Adam

---------- Forwarded message ----------
Date: Tue, 4 Nov 1997 22:56:10 +1100 (EST)
From: Michael Robert Doyle <michaeld@cse.unsw.EDU.AU>
To: compsoc.unix@cse.unsw.EDU.AU
Subject: Re: Benchmarks... (fwd)

A few weeks (days?) we were talking about the diff between the kernals....

here's some more

Mike

---------- Forwarded message ----------
Date: Tue, 4 Nov 1997 22:48:18 +1100
From: Fabio Riccardi <Fabio.Riccardi@inria.fr>
To: Multiple recipients of list <linux-pmac@samba.anu.edu.au>
Subject: Re: Benchmarks...

Hi,

for those who really want to know what goes wrong in mklinux, there is
a very recent paper discussing the subject:

The Performance of µ-Kernel-Based Systems.
Hermann H&aumlrtig, Michael Hohmuth (Dresden University of
Technology), Jochen Liedtke (IBM, Yorktown), Sebastian Sch&oumlnberg,
and Jean Wolter (Dresden University of Technology).

available at http://os.inf.tu-dresden.de/pubs/sosp97/

The paper compares the linux implementation on top of a modern
microkernel (L4) with respect to that of OFS's mach. The authors find
a typical slowdown of about 50% or more for the mach kernel, which is
much more than what originally found on older (late 80's) tests with
mach. The L4 kernel instead suffers a really minimal performance
degradation (few percent), largely balanced by the benefits of a
microkernel architecture. One interesting thing is that the mklinux
port has been largely optimized, while the L4 port was done in a
hurry, just to try things out... amazing...

It is a very nice paper, very readable, lots of thinking matter for
those who really whish to design modern operating systems...

Fabio

-- 
Fabio Riccardi - INRIA, Project SOR  |
email : Fabio.Riccardi@inria.fr      | * time is on our side! *
http://www-sor.inria.fr/~riccardi/   |