Acceleration of Solving Non-Equilibrium Ionization via Tracer Particles and MapReduce on Eulerian Mesh

摘要

Non-equilibrium ionization (NEI) is an important phenomenon related to many astrophysical processes, but the traditional method, which tightly couples the NEI solver with Eulerian mesh infrastructure, introduced high overhead on computing, memory and communication. In order to overcome the shortcomings of the pure Eulerian scheme, a new approach employing tracer particles and MapReduce model to solve the NEI problem was proposed. We introduce (1) a particle-dumping scheme for tackling the problem of large amounts of small particle snapshots continuously generated at each evolution step, (2) a parallel method based on the MapReduce model to solve the NEI equations along the particle trajectories. Both post-processing and non-intrusive in-situ schemes are supported in the paper's approach. The approach was prototyped and tested based on the FLASH multiphysics simulation framework, and it is easily adapted to other simulations modeling reactive flow on Eulerian mesh. Evaluations on up to 192 cores show that our approach can improve the end-to-end performance of a real world simulation by 3-fold above.