A cache size based timestep limiter for Implicit Monte Carlo
Alex Long · OSTI OAI (U.S. Department of Energy Office of Scientific and Technical Information) · 2024
Implicit Monte Carlo (IMC) is known to be an expensive component in multiphysics HEDP simulations.Load balance and particle communication issues make IMC difficult to parallelize and on-node performance is memory latency bound.To improve parallel scaling issues, a ghost cell region equal to the timestep size multiplied by the speed of light has been used to eliminate parallel communication during transport.We take a similar approach to on-node performance and explicitly limit the timestep size such that that the memory needed to process the particles sourced in a single cell can always fit in last level cache.We also look at using a cache-sized batch of cells and particles as an event separator in very coarse event based Monte Carlo algorithm to allow arbitrary timestep sizes.