Parallelization methods for efficient simulation of high dimensional population balance models of granulation

In order to solve high resolution PBMs to simulate real systems, with high accuracy and speed, a comprehensive and robust parallelization framework is needed. In this work, parallelization using just Message Passing Interface (MPI) and a more advanced method using a hybrid MPI + OpenMP (Open Multi-Processing) technique, have been applied to simulate high resolution PBMs on the computing clusters, SOEHPC and Stampede. We study the speed up and the scale up of these parallelization techniques for different system sizes and different computer architectures to come up with one of the fastest ways to solve a PBM to date. Parallel PBMs ran approximately 50-60 times faster, when using 128 cores, than the serial PBMs ran. In this work it is found that hybrid MPI + OMP methods which account for socket affinities led to the fastest PBM compute times and about 80% less memory than a purely MPI approach.