[ericthefatguy]

Hello everyone,

I was following the "Turbulent ONERA M6" tutorial, and when I ran the python script for parallel computation, I noticed that it was actually running in serial simultaneously on 4 cores.

The SU2 installed is version 3.25 built from source in Ubuntu_14.10 with MPI, TecIO, Metis and CGNS using g++, and the configuration setting was:

CXXFLAGS="-g -O3" CFLAGS="-g -O3" ./configure --enable-tecio --with-metis-cppflags="-D_FILE_OFFSET_BITS=64 -DNDEBUG -DNDEBUG2 -DHAVE_EXECINFO_H -DHAVE_GETLINE" --with-MPI=/usr/bin/mpicxx --with-CGNS-include=/usr/local/include --with-CGNS-lib=/usr/local/lib

Attached is a .txt file written from the terminal to explain the problem I encountered. It contains the output after I typed parallel_computation.py -f inv_NACA0012.cfg -n 4. I set the total iteration number to 5 for illustration purpose.

After a close examination of the txt file, it appears that SU2_PRT was running simultaneously on 4 cores without partitioning the mesh. It just read the configuration and mesh then exited.

I fear that I might set the configuration wrong when I built SU2 from source and the Metis was not installed properly.

Any input will be appreciated!
Eric
log.txt

Updates:

Problem Solved!

Well I just figured it out that the reason behind this issue is that I installed both OpenMPI and MPICH. SU2 was built with MPICH and the python script used mpiexec/mpirun of OpenMPI to run it, hence this problem occurred. Deleting OpenMPI did the trick.

[fpalacios]

Hi,

Thanks for your post, by the way, we are working on a ParMetis implementation... it will be available very soon (hopefully in SciTech 2015, release v3.4)

Thanks for using SU2,

Francisco Palacios
SU2 lead developer