[WBEL-users] Process CPU Binding

[Pete Biggs]

On Fri, 2005-08-05 at 00:15 -0500, Vincent Raffensberger wrote:
> I don't believe it's possible to bind a process or thread to a
> specific physical processor (at least an experimental kernel patch).
> Even if you could, I'd be skeptical of any performance benefit.

It depends on the application.  With computationally intensive programs
it is perfectly possible to design your algorithm/data so that the bulk
of the computation runs on data held in the processor cache.  If the
process then shifts CPU you have to reform the cache with a resultant
performance hit.  If you have to do this on every context switch, then
it can really hammer performance.

When I ran SGi boxes, it was common practice to lock a process to a CPU.
But part of the benefit of this was that the inter-processor/memory
architecture was slow compared to the internal CPU speed - and there was
sufficiently large caches such that even if another process executed on
the CPU, the bulk of the data was still in he cache.

P.

-- 
Pete Biggs :{)        pete @ physchem.ox.ac.uk   pete.biggs @ chem.ox.ac.uk
01865 275490 (Work)   pete1biggs @ gmail.com     pete @ biggs.org.uk