Parallel Generation of Hierarchical Cartesian Meshes for CFD on HPC Systems

Speaker: Konstantin Fröhlich (JSC / RWTH Aachen University)
Date: Thursday, 3 December 2015, 15:30-17:30
Session: Numerics I
Talk type: Short talk (15 min)

Abstract: Large-scale simulations of flow problems in intricate geometries require high spatial grid resolutions, i.e., complex flow phenomena such as turbulent boundary layers and free shear layers necessitate fine meshes to obtain sufficient accuracy. Grid generation faces restrictions of memory availability and computational power under the precondition of being stable for manifold geometries. That is, the meshing for such simulations make the use of serial methods unfeasible. The objective of this talk is to present a robust algorithm for massively parallel grid generation that efficiently employs multicore machines with distributed memory. The algorithm automatically generates hierarchical Cartesian meshes with multiple levels of refinement and is capable of reducing load-imbalance by workload redistribution on the process level. The applicability of the mesh generator is demonstrated for biofluid mechanical and technical flows, i.e., for the simulation of the flow in the human nasal cavity and around a generic Apollo-type reentry capsule.