Virtual Radiometers for Parallel Architectures

Background: Parallelization of traditional ray tracing algorithms for radiation applications in participating media suffers from the large amount of inter-processor communication required to hand off rays between processors. As a result, strong scaling of these algorithms falls short of the required scaling to utilize modern super computers. Method of approach: We have developed a parallel ray tracing scheme and demonstrated its ability to scale when coupled with a virtual radiometer model. This model incorporates a reverse monte-carlo ray tracing algorithm that avoids the messagepassing interface of parallel architectures, allowing for strong scaling to high numbers of processors. Results: Verification testing of the parallel model demonstrated the expected order of convergence. When implemented in parallel, the model attained strong scaling up to 1728 processors. Conclusions: By virtue of the reverse monte-carlo scheme, the solutions to the rays are mutually exclusive, allowing for massive parallelism. This model attains scaling that is an order of magnitude greater than previously demonstrated in the literature, making it possible to run radiative ray tracing calculations on massively parallel architectures.