openEMS.de

Hello, I have already recompiled the code OPENEMS but you must first have the MPI libraries pre-installed: (see topic : https://openems.de/forum/viewtopic.php?f=3&t=335) The benefit of MPI is real on a server made up of NUMA nodes: on my Opetron server with 8 NUMA nodes I had divided by 6 the comput...

I identified 2 potential problems: 1- The versions of "CalcNF2FF.m" are not the same for version 0.33 (MPI version) and version 0.35: for version 0.33 there was an option to add in the arguments (... "MPI", true). Is this option still useful for version 0.35 (--with-MPI version). See respective file...

Hello Thorsten, During previous discussions (quite old), you had managed to create an MPI version of OpenEMS (version 0.33) with the parallelization of the calculation of the radiation patterns (see link here: https://openems.de/forum/viewtopic.php?f=3&t=335&start=30#p2002 . Recently, I tried to rec...

Hello, By examining the "AddCircWaveguidePort.m" Matlab function, it seems that only the TE11 mode can be used: the expressions of "funct_Er, func_Ea, funct_Hr, func_Ha" do not conform to those mentioned in table 3.5 of D. Pozar's "Microwave Engineering" book when "n" or "m>1"? TE_TM_modes.JPG Could...

Hi,

You can use Aegmesher : https://bitbucket.org/uoyaeg/aegmesher/ ... Jet/Jet.md and the function "meshCreateLines" to generate meshes from a STL object.
Regards.
Pascal

Hello,

Maybe this mesher could help you :
AEG Mesher: An Open Source Structured Mesh Generator for FDTD Simulations.

https://bitbucket.org/uoyaeg/aegmesher/overview

Hi, I finally solved the problem of affinities using another pthreads wrapper both for the operator creation and the engine. I abandoned the utility "likwid-pin" and I replaced it by http://www.poempelfox.de/workstuff/pthread-overload.c . The startup script is attached and I can also modify the memo...

Hi, I tried to test MPICH but I still have some problems with likwid-pin. The reason is that for each MPI process, the "n" threads initially created were destroyed after a few seconds and then recreated, is that true ? Likwid-pin correctly distributes the threads in the cores at the beginning but wh...

Hi, 1) Concerning the calculation of the radiation pattern : Yes, it seems much better with the latest version of CalcNF2FF.m, at least at first glance... ! 2) Concerning CPU load (at 100%) of the main threads : No, the problem is the same with or without nf2ff dumps.... Do you think it could be lin...

I forgot to mention that the best score on my 48 cores (quadri-socket Opteron 6176 with 12 cores/socket) is obtained using 8 MPI processes with 6 threads per process:

Final score: 8 * 75 mcells / second equivalent to 600 Mcells / s

For comparison, with multithreading alone : 350 Mcells/s max