Parallel computation on windows 8

kont87 · February 21, 2020, 14:20

Hello,
I googled it but could not find the answer. My problem is, when I use "parallel_computation.py" script, all cores run the whole mesh instead of a partitioned one, and each core performs each iteration (4 cores = 4 identical solution), but if I use mpiexec -n #cores SU2_CFD.exe [config file] instead, It works. I have no problem with that, but for the optimization tutorials, I could not solve the issue and running the optimization cases on a single core is pretty annoying.

I followed the steps -which is very straightforward- for a Win machine, I did not compile the code by using cygwin. I have the latest version of SU2, python version is 3.8.1, numpy's is 1.18.1, scipy's is 1.4.1

I have MS-MPI, version is 10.1.12498.18. SU2_RUN, path for MPI and python are set accordingly.

Any help is appreciated.
Regards.
S.

pcg · February 22, 2020, 13:22

Do you have more than one mpi version in your system?
The optimization script might be calling mpirun instead (not sure what is the default).
If you run mpirun -n # SU2_CFD do you get the same problem?

kont87 · February 22, 2020, 14:58

Quote:

Originally Posted by pcg

Do you have more than one mpi version in your system?
The optimization script might be calling mpirun instead (not sure what is the default).
If you run mpirun -n # SU2_CFD do you get the same problem?

It seems mpirun is not recognized in my system whereas mpiexec is, do you have a suggestion?
Thank you.

pcg · February 26, 2020, 07:18

You can define a custom mpi command via environment variable SU2_MPI_COMMAND, this is used in $SU2_RUN/SU2/run/interface.py, if all else fails you can always hack that script.

kont87 · February 27, 2020, 12:19

Quote:

Originally Posted by pcg

You can define a custom mpi command via environment variable SU2_MPI_COMMAND, this is used in $SU2_RUN/SU2/run/interface.py, if all else fails you can always hack that script.

Problem is solved, thank you for your help. I have now another issue, which is exactly as described in here:
https://github.com/su2code/SU2/issues/559

I am just going over the tutorials, and not changing anything in the config files, yet I can't detect the problem. Do you have any idea?

February 21, 2020, 14:20	Parallel computation on windows 8	#1
kont87 New Member serg Join Date: Dec 2015 Posts: 29 Rep Power: 11	Hello, I googled it but could not find the answer. My problem is, when I use "parallel_computation.py" script, all cores run the whole mesh instead of a partitioned one, and each core performs each iteration (4 cores = 4 identical solution), but if I use mpiexec -n #cores SU2_CFD.exe [config file] instead, It works. I have no problem with that, but for the optimization tutorials, I could not solve the issue and running the optimization cases on a single core is pretty annoying. I followed the steps -which is very straightforward- for a Win machine, I did not compile the code by using cygwin. I have the latest version of SU2, python version is 3.8.1, numpy's is 1.18.1, scipy's is 1.4.1 I have MS-MPI, version is 10.1.12498.18. SU2_RUN, path for MPI and python are set accordingly. Any help is appreciated. Regards. S.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
The problem when i use parallel computation for mesh deforming.	Hiroaki Sumikawa	OpenFOAM Running, Solving & CFD	0	November 20, 2018 03:58
Error running openfoam in parallel	fede32	OpenFOAM Programming & Development	5	October 4, 2018 17:38
Windows 7 parallel calculations	curky	SU2 Installation	23	May 14, 2013 03:17
problem in the CFX12.1 parallel computation	BalanceChen	ANSYS	2	July 7, 2011 11:26
about parallel computation	akang	CFX	5	June 22, 2005 11:51

February 22, 2020, 13:22		#2
pcg Senior Member Pedro Gomes Join Date: Dec 2017 Posts: 466 Rep Power: 14	Do you have more than one mpi version in your system? The optimization script might be calling mpirun instead (not sure what is the default). If you run mpirun -n # SU2_CFD do you get the same problem?

February 26, 2020, 07:18		#4
pcg Senior Member Pedro Gomes Join Date: Dec 2017 Posts: 466 Rep Power: 14	You can define a custom mpi command via environment variable SU2_MPI_COMMAND, this is used in $SU2_RUN/SU2/run/interface.py, if all else fails you can always hack that script.