OpenMPI in SU2
 

If someone could describe how the Open MPI is used in the SU2, that would be very helpful.

Essentially, what SU2 information is transmitted over the Open MPI?

Right now, the Open MPI is referenced in quite many classes. Our plan is to explore the feasibility of encapsulating all the parallel computing in one class and perhaps using a new protocol to leverage some native low latency cloud platforms.

Basically, we want to make the SU2 core business logic independent of whatever underlying parallel computing platforms.

Thank you for your question.
If you want to change the way that message passing is handled in SU2, unfortunately the best way to ensure that all the necessary information is passed may be to read through the MPI calls and make sure that you are able to replicate that. I'm not sure I understand what you are trying to do, however, since there is already some abstraction of the MPI calls (see Common/include/mpi_structure.h). I believe that this was already the case at the time of your post, however since it has been a couple of months please do take a look at the most recent release on github.

In summary, solution information is transferred when quantities integrated over the whole or part of the domain are needed, and when neighbor quantities need to be updated. As the solution iterates, the value at each node is updated dependent on the values at their neighbors, and so this information needs to be updated at the boundaries between partitions in the domain. Many output quantities rely on integrals over surfaces or the volume.