Hi Aubrey,

Li, Aubrey wrote:
> If a thread has been waiting longer than the *cpu_rechoose* value(3 ticks),
> the potentially benefiting from a warm hardware cache is diminished, so
> the dispatcher will choose cpu again.
>
> Back to the original issue, do we need a per-thread basis cmt policy?
>   
I guess that's what we're discussing. :) One reason I could see wanting 
to implement the policy on a per CPU basis, is if there was something 
about the CPU (or it's power domain) that made it so you would want to 
do something different from the other CPUs in the system. But I wasn't 
trying to suggest that's what I wanted to shoot for...what I meant to 
say when I said "implemented on a per CPU basis", was that we would 
implement the system wide policy change by changing the CPU's load 
balancing lineages (which exist on a per CPU basis), but across all the 
CPUs. (which is what I describe in bug 4202).

I could see applications for a per thread policy, but what I'm not sure 
of is if that's something we want to bite off for this initial 
integration, or something we want to do when that mechanism is needed by 
something that sets policy on a per thread basis (which we currently 
don't have).

-Eric

Reply via email to