Monte Carlo method is an important method of probability theory in particle transport simulation. It does well in arbitrary geometric description and can obtain more accurate results. While it takes a lot of time to simulate particles repeatedly. Parallel computer technology is one of the important ways to solve this problem. Nowadays, most Monte Carlo codes own the function of parallel computing. But due to their deficient in data communication and parallel algorithm, they are not suited to large-scale parallel computing. How to improve parallel performance and scalability becomes a new challenge in this field.After full study and analysis of the characteristics of Monte Carlo method and parallel computing theory, based on SuperMC program and the message passing interface, this paper designed and implemented the parallel computing function of fixed-source and critical problem. In order to ensure the consistency of serial and parallel results, the original serial random number generator was parallelized. Then, this paper introduced the optimization of data communication and parallel algorithm to improve the parallel performance and robustness.To verify the correctness and parallel performance of the code, Monte Carlo particle transport code MCNP was used as reference, the ITER-Benchmark and BN600-Benchmark were chosen as the representative in a series of contrast tests. The results show the parallel performance of the algorithm based on this paper is significantly higher than MCNP, at the same time, has better scalability. |