Problems with using MPI

Praveen_krish · December 22, 2022, 13:34

Hey everyone, I am right now working on a linux machine with 104 cores and 128GB RAM. I installed MPICH and SU2 binary that supports mpi.
When I run the SU2 software with "mpiexec -n 10 SU2_CFD <configfile.cfg>" or "mpirun -n 10 SU2_CFD <configfile.cfg>" commands , instead of parallel computation over 10 cores the process is getting duplicated 10 times. I see the same iterations running on all cores.
This problem doesn't happen on my windows system. Where I installed Microsoft MPI and it works as expected. The computation occurs parallelly and hence it reaches convergence much faster.

Does anyone know what's the problem? If yes, what can be the solution?

PS: The MPICH is perfectly installed on the device (ran all the tests mentioned in the installation guide).

wkernkamp · December 22, 2022, 16:56

It must be that you compiled SU2 without the parallel options and defaults were not activated because the mpi libraries were not found.

naffrancois · December 23, 2022, 04:03

Hello,

Additionnaly to a problem in SU2 setup, this problem typically happens when you have two concurrent mpi installed in your system, e.g. openmpi and mpich. Then it might happen that you launch mpirun from mpich using libs from openmpi or vice versa.

You should check which executable and libs are used by default and change environment variables accordingly (PATH and LD_LIBRARY_PATH) in your bash file.

wkernkamp · December 23, 2022, 17:57

To compile SU2 parallel on linux, I used openmpi 3.1.6 which I compiled from source after removing the standard openmpi (it was 4.???).

December 22, 2022, 13:34	Problems with using MPI	#1
Praveen_krish New Member Praveen Join Date: May 2022 Posts: 8 Rep Power: 4	Hey everyone, I am right now working on a linux machine with 104 cores and 128GB RAM. I installed MPICH and SU2 binary that supports mpi. When I run the SU2 software with "mpiexec -n 10 SU2_CFD <configfile.cfg>" or "mpirun -n 10 SU2_CFD <configfile.cfg>" commands , instead of parallel computation over 10 cores the process is getting duplicated 10 times. I see the same iterations running on all cores. This problem doesn't happen on my windows system. Where I installed Microsoft MPI and it works as expected. The computation occurs parallelly and hence it reaches convergence much faster. Does anyone know what's the problem? If yes, what can be the solution? PS: The MPICH is perfectly installed on the device (ran all the tests mentioned in the installation guide).

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
How to allow MPI code run in serial mode with a single CPU core?	aerosayan	Main CFD Forum	15	May 30, 2022 14:05
Problems when compile SU2 v4.X with openmpi	BopngMa	SU2	1	September 6, 2015 07:30
MPI InterFoam	metro	OpenFOAM Running, Solving & CFD	7	March 27, 2013 04:24
Error using LaunderGibsonRSTM on SGI ALTIX 4700	jaswi	OpenFOAM	2	April 29, 2008 11:54
MPI PROBLEMS	gtg627e	OpenFOAM Running, Solving & CFD	20	October 5, 2007 05:02

December 22, 2022, 16:56		#2
wkernkamp Senior Member Will Kernkamp Join Date: Jun 2014 Posts: 371 Rep Power: 14	It must be that you compiled SU2 without the parallel options and defaults were not activated because the mpi libraries were not found.

December 23, 2022, 04:03		#3
naffrancois Senior Member Join Date: Oct 2011 Posts: 242 Rep Power: 17	Hello, Additionnaly to a problem in SU2 setup, this problem typically happens when you have two concurrent mpi installed in your system, e.g. openmpi and mpich. Then it might happen that you launch mpirun from mpich using libs from openmpi or vice versa. You should check which executable and libs are used by default and change environment variables accordingly (PATH and LD_LIBRARY_PATH) in your bash file.

December 23, 2022, 17:57		#4
wkernkamp Senior Member Will Kernkamp Join Date: Jun 2014 Posts: 371 Rep Power: 14	To compile SU2 parallel on linux, I used openmpi 3.1.6 which I compiled from source after removing the standard openmpi (it was 4.???).