Managing several instances of code-aster with MPI

paul.oumaziz

Hello,
I am trying to implement a domain decomposition method and I use code-aster as the non-linear solver over the subdomains. Therefore I want to execute several independent instances of code-aster and I manage the communication between the processes with mpi4py.
To do so, I split the global communicator to initiate code-aster with the specific sub-communicators (with CA.init(comm=comm_split)) otherwise the instances of code-aster are getting in each other's way. But even with such sub-communicators I have the impression that the instances are interacting with each other. Maybe the STAT_NON_LINE of each instance uses objects in the bases which belong to others instances ?
I also tried to use a bash script to execute separated command files in separated folders with mpirun but it doesn't change anything.
Maybe I should install a sequential version of code-aster ?

Or do you have a way to execute multiple instances of code-aster with MPI which are completely independent from each other ?

Thank you for your help,
Paul

nicolas_p

Hello Paul,

It should work. I did it for test zzzz178d. A thermo-mechanical coupling with two instances of code_aster where I use the sub-communicator. Each instance has its own life and can exchange data with the global communicator.

If you have a small example which reproduces the problem