JCOGIN. A parallel programming infrastructure for Monte Carlo particle transport

The advantages of the Monte Carlo method for reactor analysis are well known, but the full-core reactor analysis challenges the computational time and computer memory. Meanwhile, the exponential growth of computer power in the last 10 years is now creating a great opportunity for large scale parallel computing on the Monte Carlo full-core reactor analysis. In this paper, a parallel programming infrastructure is introduced for Monte Carlo particle transport, named JCOGIN, which aims at accelerating the development of Monte Carlo codes for the large scale parallelism simulations of the full-core reactor. Now, JCOGIN implements the hybrid parallelism of the spatial decomposition and the traditional particle parallelism on MPI and OpenMP. Finally, JMCT code is developed on JCOGIN, which reaches the parallel efficiency of 70% on 20480 cores for fixed source problem. By the hybrid parallelism, the full-core pin-by-pin simulation of the Dayawan reactor was implemented, with the number of the cells up to 10 million and the tallies of the fluxes utilizing over 40GB of memory. (author)