PH>> Might this be the reason for the slowdown (two  threads/apps access
PH>> the same memory areas) ? 
DR> But, just as a practical matter, if multiple threads 
DR> accessing the same memory causes that bad
DR> a performance degredation, then what good is SMP?

There are also all those kernel spinlock semaphores. Lots of
kernel stuff only runs on one of the multiple processors. The
story is, not much time is spent in kernel code, so dropping
to one processor is OK. That's the party line. It's like "32-bit"
OS/2 and "32-bit" Windows (can you say "thunk"?). It's one
thing to have Gene Amdahl laughing quietly in the distance and
quite another to invite him into the party, let the prettiest girl
sit on his lap, and make drinks for him.

it's easy to see that affordable SMP will suffer the h/w stalls
you suspect, and that SMP that does not will be even more
expensive. still, i'd give h/w the benefit of a doubt until you
have shown solid improvement on one SMP box vs another. the
s/w may also need to evolve.

___
 X KWQ/2 1.2e X Intel: Accruacy Inside

--- Maximus/2 2.02