Parallelization of hierarchical radiosity algorithms on distributed memory computers

Abstract

Computing distribution of light in a given environment is an important problem in computer-aided photo-realistic image generation. Radiosity method has been proposed to address this problem which requires an enormous amount of calculation and memory. Hierarchical radiosity method is a recent cipproach that reduces these computational requirements by careful error analysis. It has its idea from the solution methods of N-body problems. Although hierarchical approach has greatly reduced the amount of calculations, satisfactory results still cannot be obtained in terms of processing time. Exploiting parallelism is a practical way to I'educe the computation time further. In this thesis, we have designed and implemented a parallel hierarchical radiosity algorithm for distributed memory computers. Due to its highly irregular computational structure, hierarchical radiosity algorithms do not yield easily to parallelization on distributed memory machines. Dynamically changing computational patterns of the algorithm cause severe load imbcilances. Therefore, we have developed a dynamic load balancing technique for the parallel hierarchical radiosity calculation.