[mpich-discuss] Performance of openMP vs MP

[bob ilgner]

What is quicker for generalised looping structures on shared memory
platforms, openMP or MPI.

The few tests I have run on a shared memory machine shows that my
Mpich2 implementation runs quicker than the equivalent openMP for most
situations(see attachment). I am in a discussion at the moment with
someone telling me that the openMP should be quicker thnan the MPI for
a smaller number of cores.

Is there an experience base or rule that asserts that openMP
parallelisation of loops should be quicker than MPI ? Why should
openMP be quicker than MPI on shared memory platforms ?
-------------- next part --------------
A non-text attachment was scrubbed...
Name: SMP_compare_platforms.png
Type: image/png
Size: 56965 bytes
Desc: not available
URL: <http://lists.mcs.anl.gov/pipermail/mpich-discuss/attachments/20110522/d7ee1eab/attachment-0001.png>