Parallel simulation of radiative heat transfer using an unstructured finite-volume method

A spatial domain-based parallel algorithm is developed for simulating radiative heat transfer in a distributed computing environment. The radiative transfer equation is solved using an unstructured finite-volume method that is applicable for any 2D planar, axisymmetric, and 3D problems with structured, unstructured, or hybrid grins. The domain decomposition is carried oat by equally partitioning the spatial domain into many subdomains along the longer geometric dimension. Communication among the subdomains on each processor is performed through a message-passing interface library. In order to examine the parallel performance of the unstructured radiation code, two benchmark problems are investigated for different absorption coefficients, scattering coefficients, and grid sizes in a parallel computer. To help us understand the change of parallel performance, a new parameter the total inner iteration number, is introduced to analyze the results. For all the cases examined, as expected, the parallel performance is seen to degrade rapidly with an increase of the processor number. However, in contrast with other studies, the parallel performance is found to degrade with an increase of absorption coefficient for a temperature-prescribed problem. Also, the global iteration number is found to be not necessarily independent of the grid size.