On Thu, 2005-02-03 at 11:22 +0000, Vincent Bray wrote:
> Please forgive me if I'm being dumb here... In the case of
> 'hyperthreading' CPUs (the Xeon), is there an advantage to locking a
> process to one of these 'CPUs' (ie half of the hardware chip).
Yes. In earlier 2.6 (or 2.5.x... it was a while ago) linux kernels (and
before), the OS would sometimes move processes between the "CPUs" in a
very silly way (esp in SMP+SMT).
In later 2.6's (or 2.5's... i forget and can't find the changelog entry)
the scheduler is aware of HT CPUs and that they really are just one
physical CPU and assigns them one (not two) runqueues. This stops some
needless jumping around and along with some other things implemented
later on, the linux kernel is quite aware (and well behaved) with SMT
CPUs (and in a SMP+SMT environment).
for a bit more info:
http://kerneltrap.org/node/391 (Ingo's patch for SMT)
http://lwn.net/Articles/80911/ (Scheduler Domains)
So... you may not need to tie processes down on SMP+SMT (or just SMT)...
linux may be smart enough to do it for you. A bit of experimenting
couldn't hurt though.
--
Stewart Smith, Software Engineer
MySQL AB, www.mysql.com
Office: +61 4 3 8844 332
Jumpstart your cluster:
http://www.mysql.com/consulting/packaged/cluster.html
Attachment: [application/pgp-signature] This is a digitally signed message part signature.asc
Attachment: [application/pgp-signature] This is a digitally signed message part signature.asc
