Research On Parallel Algorithms For Deterministic Particle Transport Problems

In the conditions of high temperature and high pressure, the unknown particle fluxes of particle transport equations are defined in energy, time, velocity phase-space and high dimensions geometry space. We need to deal with many problems such as multi-media, multi-group, complex initial and boundary conditions, which contribute to solving the equations. The huge amount of calculation is one of the biggest problems that can't get around. So it is necessary to design scalable parallel algorithms on solving those equations.It's a hard problem to design and implement a parallel Sn algorithm based on geometry domain decomposition in unstructured grids. The communicating delay and sorting algorithm limit the performance of current algorithms, which can hardly get large scalable parallel computing.We present an improved algorithm which is based on [1] about parallel pipelined Sn sweeping algorithm. The improved algorithm can lead to less dependence of original algorithm on network latency. Therefore, the parallel performance of the new algorithm can increase 15% than the old one.Ordering algorithm plays an important role in the parallel algorithm that greatly influences the parallel performance. Based on DFDS priority strategy, we design the PDFDS algorithm for parallel ordering. The performance of PDFDS algorithm can increase 13% than the traditional parallel ordering algorithm on solving large scale problems.At last, using the improved parallel Sn sweeping algorithm and PDFDS algorithm, we solved a two dimension particle transport equations on a parallel computer. Performance results show that the algorithms have well scalability. Parallel efficiency relative to 64 is 96% on 1024 processors. In best cases, the parallel execution time decrease 24% than the similar codes.