2485 Parallelism in Large a , Radiative

Parallelism for gray participating media radiation heat transfer may be placed in two primary categories: spatial and angular domain-based parallelism. Angular, e.g., ray based, decomposition has received the greatest attention in the open literature for moderate sized applications where the entire geometry may be placed on processor. Angular based decomposition is limited, however, for large scale applications (0(106) to 0( 108) computational cells) given the memory required to store computational grids of this size on each processor. Therefore, the objective of this work is to examine the application of spatial domain-based parallelism to large scale, three-dimensional, participating-media radiation transport calculations using a massively parallel supercomputer architecture. Both scaled and fixed problem size efficiencies are presented for an application of the Discrete Ordinate method to a three dimensional, non-scattering radiative transport application with nonuniform absorptivity. The data presented shows that the spatial domain-based decomposition paradigm results in some degradation in the parallel efficiency but provides useful speedup for large computational grids.