Hybrid parallel strategies for the neutron transport code PANDAS-MOC

Abstract

PANDAS-MOC (Purdue Advanced Neutronics Design and Analysis System with Methods of Characteristics) is being developed to find high fidelity 3D solutions for steady state and transient neutron transport analysis. However, solving such transport problems in a large reactor core could be extremely computationally intensive and memory demanding. Because parallel computing is capable of improving computing efficiency and decreasing memory requirements, three parallel models of PANDAS-MOC are designed using the distributed memory and share memory architectures in this article: a pure message passing interface (MPI) parallel model (PMPI), a segment OpenMP threading hybrid model (SGP), and a whole-code OpenMP threading hybrid model (WCP). Their parallel performances are examined by the C5G7 3D core. For the measured speedup, PMPI model > WCP model > SGP model. The memory consumed by the WCP model is about 60% of that consumed by the PMPI model. This study also demonstrated that the performance of WCP parallelism is limited by the hybrid reduction in the CMFD calculation and omp atomic clause in the MOC sweep. Once they are optimized, the WCP model can outperform the PMPI model.