For all that's worth: we run a program doing lots of symbolic computations
and allocating lots of small memory blocks (< a few k) on a variety of
machines. The P4 performs roughly proportionally to its frequency compared
to the P3; the Athlon is slightly faster than the P3.

We ran the program under Linux + OProfile.

Roughly 12% of the time is spent within the GC, representing roughly 30%
of the cache faults. Surprisingly, it seems that accesses to unboxed
floats in some comparison procedures take up 10% of the time and 4% of the
page faults.

"Profile - don't speculate.". You should probably get OProfile or any
other kind of profiler that enables you to count events such as cache
faults.

David Monniaux            http://www.di.ens.fr/~monniaux
Laboratoire d'informatique de l'École Normale Supérieure,
Paris, France



-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners