Subject: Re: 2.0_RC1/i386: SMP not working?
To: Geert Hendrickx <email@example.com>
From: Manuel Bouyer <firstname.lastname@example.org>
Date: 10/03/2004 20:00:44
On Sun, Oct 03, 2004 at 12:54:46PM +0200, Geert Hendrickx wrote:
> On Thu, Sep 30, 2004 at 05:57:30PM +0000, Christos Zoulas wrote:
> > Is it a true dual P4 or a hyperthreaded one? If you have two physical
> > processors, then this is a bug. If you have two logical ones, then you
> > need to use an ACPI kernel inorder for them to be found.
> > christos
> Apparantly, it's a hyperthreaded one. I'm not sure whether I fully
> understand the concept of logical CPU's. With an ACPI-disabled kernel,
> only one CPU appears. Does the one physical CPU run at full potential
> then, or not?
There is a little theorical gain with hyperthreading. From tests I ran,
running a NetBSD SMP kernel on a P4-HT gives a few percents gain on a kernel
make -j4 compile over a UP kernel. With linux 2.6, SMP gives a few percents
loss (I guess, because of finer-grained locking in the kernel).
So there may be a little benefit in using a MP kernel with a HT CPU,
but you want to benchmark your application(s) to know if it's worth it or
Manuel Bouyer <email@example.com>
NetBSD: 26 ans d'experience feront toujours la difference