Abstract:
We present the results of a study of the parallel algorithms based on MPI and OpenMP for vector splitting schemes in heat transfer problems. The following parallel implementations are compared: MPI, “simple” MPI/OpenMP (#pragma omp directives applied to MPI-based code), and MPI/OpenMP with “postman” threads. The main idea of the last algorithm is to split off one thread within each computational node to perform data transfer. This approach allows one to implement overlapping of useful computations and data transfer. The presented results show that introducing postman threads can significantly improve performance of MPI/OpenMP implementation; nevertheless, for the considered class of numerical algorithms, it is more reasonable to use MPI implementation.