Multilevel Parallel Strategy and Efficiency Optimization in NECP-X

Long computing time and abundant memory requirement are two most critical issues for the method of characteristics (MOC) applied for the three-dimensional whole-core transport calculation. The massive parallel is an effective method to solve these problems. With the rapid development of supercomputers in China, massive parallel computing is becoming possible. How to develop parallel algorithm becomes a top priority. In this paper, a multilevel parallel strategy of space, angle and characteristic rays was studied in the numerical reactor neutronics code NECP-X. To realize efficient parallel computing, space parallel was developed based on domain decomposition method. To get good load-balance of angle parallel, a greed algorithm for angle domain decomposition was applied. To save memory and increase efficiency in the characteristic ray parallel, a dynamic scheduling feature within shared memory architectures was used and analyzed. The numerical results show that the parallel efficiency of space, angle and characteristic rays in NECP-X is high. NECP-X can take full advantage of parallel resources and achieve large-scale parallelism.