NUMA-AWARE DATA MANAGEMENT FOR NEUTRON CROSS SECTION DATA IN CONTINUOUS ENERGY MONTE CARLO NEUTRON TRANSPORT SIMULATION

The calculation of macroscopic neutron cross-sections is a fundamental part the continuous-energy Monte Carlo (MC) transport algorithm. MC simulations full nuclear reactor cores are computationally expensive, making high-accuracy impractical for most routine analysis tasks because their long time to solution. Thus, preparation simulation algorithms next generation supercomputers extremely important as improvements in computational performance and efficiency will directly translate into achievable accuracy. Due stochastic nature algorithm, cross-section data tables accessed highly randomized manner, resulting frequent cache misses latency-bound memory accesses. Furthermore, contemporary non-uniform access (NUMA) computer architectures, featuring very high latencies less space per core, exacerbate this behaviour. absence topology-aware allocation strategy existing high-performance computing (HPC) programming models major source problems NUMA systems. improve we propose strategies that allow control over location structures within hierarchy. A new management library, known AML, has recently been created facilitate mapping. To evaluate usefulness AML context simulations, have converted two lookup “proxy-applications” (XSBench RSBench) utilize library. In study, use these proxy-applications test several (the nuclide grid, unionized logarithmic hash multipole methods) with number schemes on variety node architectures. We find library speeds up 2x current hardware (e.g., dual-socket Skylake-based system) compared naive allocation. These exciting results also show path forward efficient next-generation exascale supercomputer designs feature even more complex hierarchies.