PARALLELIZATION OF THE DISCRETE ORDINATES METHOD

Two different parallelization strategies of the discrete ordinates met hod (DOM) are described and implemented in a distributed-memory comput er. In the angular decomposition (ADP), each processor performs the ca lculations for the whole domain but deals only with a few directions, while in the spatial domain decomposition (DDP) each processor treats all the directions but only for a subdomain. In the ADP the number of iterations is independent of the number of processors, contrary to the DDP, and higher efficiencies are obtained. The influence of the order of quadrature, grid size, and absorption coefficient of the medium is also investigated.