Performance evaluation for neutron transport application using message passing

In this paper, recent advances in parallel software development for solving neutron transport problems are presented. Following neutron paths along the characteristics of the system, the transport equation is solved to obtain the scalar flux per region and energy group. Due to the excessive number of tracks in the demanding context of 3D large-scale calculations, the parallelisation of the solver is considered in order to obtain fast iterative solution. Different load balancing strategies are used for the distribution of the tracks along processors. These strategies are based on the calculation load implied by each track length. The performance of the MPI implementation, using different parallel machines, is analysed for realistic applications and numerical results are shown. A comparative study of different networks existing on the market and the influence of their parameters on total communication time is also presented.