Subject: Re: scheduler woes on MPACPI kernel
To: Thor Lancelot Simon <firstname.lastname@example.org>
From: James Chacon <jmc@NetBSD.org>
Date: 01/18/2005 23:43:04
On Wed, Jan 19, 2005 at 12:17:39AM -0500, Thor Lancelot Simon wrote:
> On Tue, Jan 18, 2005 at 11:14:40PM +0100, Johnny Billquist wrote:
> Which leaves you with no win from hyperthreading -- but all the "lose"
> that you get from the multiprocessor locking, TLB contention, and so
> forth of a real multiprocesor system.
> It is not surprising that it is slower.
Perhaps on NetBSD right now it's slower. I ran extensive tests on a 2 real
CPU w. HT linux box I use to cross build on. Turns out there that HT gives
you about a 10% boost. I could get a build of x86 w. X11 to build in about 1h
even w/o HT and 55m with HT with -j0 -> -j8 on both runs in an attempt to find
the sweet spot. Granted...I beleive they have HT aware scheduling to try
and keep processes on the same physical cpu's so you don't lose as much then.
Now, a single cpu box w. HT I can easily beleive will lose.