On Thu, 8 Jul 2004 [EMAIL PROTECTED] wrote:
> > The "HT" isn't a full/complete cpu (dual core cpus are in the process
> > of being developed by Intel (and others))...
> scheduler in the 2.4 series kernel is not HT "aware". Thus the processor afinity
> is incorrectly treated like a traditional second CPU, which HT is not. This
Exactly.
> has the effect of clobbering the CPU cache and net result is EVERYTHING runs
> slower.
Sometimes very much slower - it can be a real system performance
killer. You get the illusion of all 4 cpus running but the system
is slower overall.
> The 2.6 kernel is fully HT enabled and knows exactly how to treat the HT
> enabled CPU's... So, enabling HT should provide some boost again.
It's not _just_ the cpu scheduler (although that is an important part),
but the types of instructions that are being executed. The other
part of the equation (in order to get the maximum benefit of HT) is
to have a HT aware compiler that will schedule instructions so two
cpus do not interfere-with or stall waiting for each other.
Getting 10% or so improvement is nice (I will have to rerun the
tests on a 2.6 based kernel now that I have SuSE 9.1 on a P4/Xeon
system - with a 2.4 kernel the system slowed down quite a bit) but
is a long way from ~2x that one might have been expecting from a
SMP system.
What I am looking forward to is the new dual core cpu technology -
at that time having two cpu chips will yield a real 4-way system.
Wheee!
Cheers,
Steven Schultz
-------------------------------------------------------
This SF.Net email sponsored by Black Hat Briefings & Training.
Attend Black Hat Briefings & Training, Las Vegas July 24-29 -
digital self defense, top technical experts, no vendor pitches,
unmatched networking opportunities. Visit www.blackhat.com
_______________________________________________
Mjpeg-users mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/mjpeg-users
