Partition deactivation with load balancing for parallel flow simulations

An algorithm is presented to save CPU time by dynamically deactivating partitions of a decomposed computational mesh during parallel flow simulations. The procedure targets classes of problems where the propagation behavior, inherent in the equations solved, can be exploited, such as detonation or scalar transport. Combined with dynamic load balancing based on real-time CPU-measurement, available, e.g., by coding the solver using the Charm++ runtime system, yields considerable savings in solution times for both shared-, and distributed-memory calculations. The complete source code is available at https://xyst.cc . • An algorithm is presented to save CPU time during parallel flow simulations by deactivating mesh partitions. • Combined with CPU-based load balancing the procedure is demonstrated to yield considerable savings in time-to-solution. • The complete source code is available at https://xyst.cc .