It is well known - small simulations are better with single core than multicore. In order to get a reasonable speed-up on a multicore a simulation should exceed e.g. 100 Mb (in occupied memory). Just use a higher resolution and see how the speedup will change. If increase of resolution leads to better speedup then there is nothing wrong with your setup. Besides, MPICH is known to have poor performance with multicore systems, use OpenMPI instead, it is much better.
2010/5/14, Jonathan Hodges <js...@columbia.edu>: > **Dear Meep Users, > > I have a quick question about using meep-mpi on a multicore processor, > specifically the new CORE i5 from Intel. MPICH2 recognizes the availability > of the multicore processor and initializing the meep-mpi command using (from > the apt-get binaries): > > mpirun -np 4 /usr/bin/meep-mpi some_structure.mpi > > successfully starts meep, recognizes the 4 processors and is off > calculating. > > The problem is that the performance is worse with 4 cores than it is with > one core. For example, running the holey-wvg-cavity.ctl example takes 27s > and 76s with 1 and 4 cores, respectively. > > Am I doing something wrong in configuring mpirun that causes the performance > to suffer? Am I missing some critical idea about MPI/MPICH2 on multicores > that can explain the performance degradation? > > Thank You, > Jonathan Hodges > Columbia University > _______________________________________________ meep-discuss mailing list meep-discuss@ab-initio.mit.edu http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/meep-discuss
