On Thu, 23 Nov 2006, Pieter G wrote:
I was actually thinking more along the lines of "integrating" SPICE type
models into the simulation. Maybe measuring currents at certain points and
feeding that into SPICE simulation and then taking output from SPICE and use
it as a source somewhere else inside the FDTD simulation space.
Is this a good idea with the way MEEP works? It seems to be the case
especially if I use the C++ interface to MEEP.
I don't see any obstacles to that from the Meep side, although I'm far
from an expert on circuit simulations.
And another question. If MEEP divides the simulation into MPI jobs, how much
information flows between the jobs in relation to the total amount of
information contained in sampling space?
The parallelization is very simple: it just divides space (i.e. the
overall grid) into roughly equal sized chunks (rectangular blocks), one
per MPI process. The commmunication is basically just the adjacent
surfaces of the chunks.
The reason why I am asking is to get a feel of how long a job will take
approx if I use commercial beige boxen with say 2GB RAM + gigabit ethernet
connecting them.
It really depends on the size of the problem; the bigger the problem, the
better it parallelizes on a fixed number of CPUs.
We regularly run large jobs on supercomputers with 64+ processors, and are
usually not too far off from linear speedup. YMMV.
Steven
_______________________________________________
meep-discuss mailing list
[email protected]
http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/meep-discuss
