The flux computations and I/O are parallelized if the flux plane spans multiple processors — the only quantities that are sent to all CPUs, I think, are the total powers, which is only one number per frequency and hence should be a small amount of data.
What may be happening in a transmission calculation, however, is that the entire flux plane ends up on one processor, and hence isn't parallelized. See: http://meep.readthedocs.io/en/latest/FAQ/#should-i-expect-linear-speedup-from-the-parallel-meep <http://meep.readthedocs.io/en/latest/FAQ/#should-i-expect-linear-speedup-from-the-parallel-meep> > > I am trying to calculate transmittance spectrum using parallel Meep > with python interface. > When flux data are processed with 'save_flux' or 'get_fluxes' methods, > all the MPI processes seem to get whole data and use huge memory. > Is this behavior normal? > Please let me know if there is a way to reduce memory usage. > > Best regards, > Munehiro
_______________________________________________ meep-discuss mailing list meep-discuss@ab-initio.mit.edu http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/meep-discuss
