Michael Bader, IPVS, Universität Stuttgart
Memory-Efficient Simulations on Dynamically Adaptive Meshes: From Algorithms to Software
The talk will focus on an inherently memory and cache efficient approach for simulation problems that require dynamically adaptive refinement and coarsening of the underlying discretisation mesh. For element-based discretisation methods in particular (Finite Element or Discontinuous Galerkin methods, e.g.), we will discuss the respective algorithms, their parallelisation, and software concepts to develop such approaches into more flexible simulation environments.
The approach combines structured adaptive grid refinement and element orders defined by the Sierpinski space filling-curve. The locality properties induced by the Sierpinski curve are exploited for memory-efficient implementation following a stack&stream approach, and for efficient parallelisation and load balancing. Current applications are a Finite Volume/discontinuous Galerkin solver for the shallow water equations (in the context of tsunami simulation) and coupled problem as they occur in porous media flow.