parallel_computation.py cannot do parallel computation normally, and repeatition occurs

[HANSIeltsKing]

Hi, this is an SU2 user who want to use parallel_computation.py to simulate the QuickStart configuration file.
When I access the QuickStart folder and type “parallel_computation.py -f inv_NACA0012.cfg -n 4”, the parallel computation do not work but everything repeat 4 times. The same solution also appear 4 times and the speed become lower.

|           0|  1.1562e+00|   -1.026860|    0.201330|    0.082117|
|           1|  1.0859e+00|   -1.169122|    0.263458|    0.037718|
|           1|  1.1016e+00|   -1.169122|    0.263458|    0.037718|
|           1|  1.1172e+00|   -1.169122|    0.263458|    0.037718|
|           1|  1.1562e+00|   -1.169122|    0.263458|    0.037718|
|           2|  1.0833e+00|   -1.236203|    0.301817|    0.029207|
|           2|  1.1042e+00|   -1.236203|    0.301817|    0.029207|
|           2|  1.1354e+00|   -1.236203|    0.301817|    0.029207|
|           2|  1.1458e+00|   -1.236203|    0.301817|    0.029207|
|           3|  1.0742e+00|   -1.327201|    0.320840|    0.024965|
|           3|  1.1055e+00|   -1.327201|    0.320840|    0.024965|
|           3|  1.1172e+00|   -1.327201|    0.320840|    0.024965|
|           3|  1.1172e+00|   -1.327201|    0.320840|    0.024965|
|           4|  1.0906e+00|   -1.419001|    0.329132|    0.023806|
|           4|  1.0844e+00|   -1.419001|    0.329132|    0.023806|
|           4|  1.1125e+00|   -1.419001|    0.329132|    0.023806|
|           4|  1.1187e+00|   -1.419001|    0.329132|    0.023806|
|           5|  1.0859e+00|   -1.513907|    0.332101|    0.023194|

However, when I type this "mpirun -np 4 SU2_CFD inv_NACA0012.cfg" the parallel computing can work normally.
I am sure I installed mpich and enabled python wrapper.
How can I solve this problem of those repeatition?

Thank you very much!

[pcarruscag]

The parallel_computation script is useless at the moment.
In theory this script will call mpirun, so unless there is another mpirun on your system, which python finds first somehow... it should work.
Try setting SU2_MPI_COMMAND="full path to the mpirun that works" (get the path with which mpirun)

[HANSIeltsKing]

Thanks for your reply.

After I set SU2_MPI_COMMAND in interface.py by using the path from "which mpirun", the repeatition still occurs. Any code related to python cannot work normally (such as parallel_conputation.py, shape_optimization.py, SU2_CFD.py etc.)

Actually I followed the installation guide step by step, the custom mpi and the python wrapper were also enabled. Could this be any other missing detail in the installation process? The SU2 version I use is 7.3.0.

Thank you very much.

[thw1021]

I recently met the same problem.

Firstly, check your mpicc and mpicxx path. Try which mpicc and which mpicxx. The output should be: /usr/bin/mpicc and /usr/bin/mpicxx.

If you get something like /home/username/miniconda3/envs/su2/bin/mpicc and /home/username/miniconda3/envs/su2/bin/mpicxx, then the problem will occur.

I found the key to solve this problem is to make sure mpicc points to /usr/bin/mpicc and mpicxx points to /usr/bin/mpicxx.

Here is the solution I found:

1. uninstall the previously installed mpi4py: conda uninstall mpi4py
2. reinstall mpi4py by python -m pip install mpi4py --user

The problem should then be solved. Hope this can help.

[thw1021]

Of course, always to make sure that you are using a virtual environment.

[HANSIeltsKing]

Thanks for your reply.

I am not sure how to confirm the path is /usr/bin/mpicc?
because the command of "--enable-mpi --with-cc=/usr/bin/mpicc --with-cxx=/usr/bin/mpicxx“ cannot be used in the meson build.

so which command should I type to ensure the meaning of "--enable-mpi --with-cc=/usr/bin/mpicc --with-cxx=/usr/bin/mpicxx"?
otherwise ninja cannot install due to lacking of <mpi.h>

[pcarruscag]

export CC=..../mpicc
export CXX=..../mpicxx
Before the first call to ./meson.py

[HANSIeltsKing]

but if i do it in the virtual environment, meson cannot build because of the absolute path and relative path mixed.
how can i solve this problem?

[thw1021]

This explanation may be better:

make sure mpicc and mpicxx points to system mpi instead of mpi4py

I installed openmpi by sudo apt-get install libopenmpi-dev openmpi-bin. After that, which mpicc returns /usr/bin/mpicc, and which mpicxx returns /usr/bin/mpicxx.

[HANSIeltsKing]

The current installed version only has SU2_CFD as a parallel format supported by openmpi. But other parts, including SU2_CFD_AD, are still solved in series.

So this is not purely a python problem.

Have you ever tried to successfully parallelize calculations in a solver other than SU2_CFD?

[thw1021]

I did a test following this tutorial.

The Steps:

1. mpirun -n 2 SU2_CFD inv_NACA0012.cfg > su2_cfd.log 2>&1, and here is su2_cfd.log

2. mpirun -n 2 SU2_CFD_AD inv_NACA0012.cfg > su2_cfd_ad.log 2>&1, and here is su2_cfd_ad.log

Things seem fine. Could you have a check, please?

[thw1021]

I made a demo to show all I have done when install SU2. Hope it helps.

Please see https://github.com/thw1021/mySU2Docker

[HANSIeltsKing]

it works for me
thank you @thw1021

[thw1021]

Glad it helps :-)