SPAWN: An Iterative, Potentials-Based, Dynamic Scheduling and Partitioning Tool

Many applications of physics modeling use regular meshes on which computations of highly variable cost can occur. Distributing the underlying cells over manycore architec-tures is a critical load balancing step that should increase the period until another step is required. Graph partitioning tools are known to be very effective for such problems, but they exhibit scalability problems as the number of cores and the number of cells increases. We introduce a dynamic task scheduling approach inspired by physical particles interactions. Our method allows cores to virtually move over a 2D/3D mesh of tasks and uses a Voronoi domain decomposition to balance workload among cores. Displacements of cores are the result of force computations using a carefully chosen pair potential. We evaluate our method against graph partitioning tools and existing task schedulers with a representative physical application, and demonstrate the relevance of our approach.