Gerry,
Debates and differences aside, often-times, this forum *is* an authoritative
source of information.
Indeed it is.
By the way, we've got an old-ish course on multilevel
parallel programming at ci-tutor.ncsa.uiuc.edu in case anybody wants to go
there and see what people were thinking a couple years ago when they wrote
it.
I've seen the benchmark speedups with mixed-mode as well. With the right
code, on a day with the wind to your back, threads can make good use of
communication induced idle time on a node [yeah MPI supports overlap...
that's more difficult to achieve than it would appear, requiring--
excellent programming, most excellent MPI implementation].
By the way, if you've used Intel's mkl, you may have run with hybrid code
already and not know it [do you know how your system sets OMP_NUM_THREADS
?].
-Galen
_______________________________________________
Beowulf mailing list, Beowulf@beowulf.org
To change your subscription (digest mode or unsubscribe) visit
http://www.beowulf.org/mailman/listinfo/beowulf
