python-meep team mailing list archive

[Martin Fiers <mfiers@xxxxxxxxx>]

Dear Jeff,

Great that you got it working!

The speedup of only 30% for 4 cores is probably due to the fact that the
example is too small for 4 cores. Once the simulation starts, there's no
Python overhead. For higher resolutions you will see more effect of using
multiple cores.

Kind regards,
Martin Fiers

2012/11/11 Jeff Chiles <pseudotexan@xxxxxxxxx>

> I finally got it working.  The problem seemed to be a few things.  But the
> tweaks below (some of which you recommended) seem to work fine:
>
> -changed the "include" path in setup-mpi.py to link to a folder with the
> files from meep-openmpi: meep.hpp and /meep/mympi.hpp and /meep/vec.hpp.
>
> -in the "meep-swig" part of setup-mpi.py, I changed the libraries
> extension to ['meep_openmpi']
>
> -I ran the script with "mpirun -n 4 python scriptname.py" and it works.
>
> I did however notice that the speedup seems to be very small for the
> time-stepping portion, about 30% increase for 4 vs. 1 core.
>
> Thanks for your help with troubleshooting this though.
>
> Best regards,
> Jeff
>
>
> On Wed, Nov 7, 2012 at 3:26 AM, Nizamov Shawkat <nizamov.shawkat@xxxxxxxxx
> > wrote:
>
>> Hello,
>>
>> 2012/11/6 Jeff Chiles <pseudotexan@xxxxxxxxx>:
>> >
>> > Now it has another problem... when I execute a script with mpirun and
>> the
>> > "-np 2" argument, it seems that the program just executes two identical
>> > copies of itself in parallel rather than splitting up any of the work:
>> >
>> > The command I use is
>> >
>> > mpirun -np 2 python python_meep_bent_wg_polygons.py
>> >
>>
>> Please try:
>>
>> python python_meep_bent_wg_polygons.py -np 2
>>
>>
>> With best regards.
>> Shawkat Nizamov
>>
>
>