Parallel Unstructured Grid Generation for Complex Real-World Aerodynamic Simulations

PAGE The focus of this thesis is the design and im plem entation of a framework for Parallel U nstructured Grid G eneration based on the Advancing Front (AF) technique. Parallel U nstructured Grid G eneration is a challenging problem for two reasons. F irst, Domain Decom position (DD) is a necessary pre-processing step by which the com putational do­ m ain is decomposed in several smaller sub-dom ains th a t can be meshed in parallel. The DD problem is difficult because the size and shape of the separators m ust be com patible w ith the local spacing and shape required by the application. Second, Load Balancing (LB) is the assignm ent of the sub-dom ains to a set of processors such th a t the load is evenly d istributed. The LB problem is difficult because accurately estim ating the load of each sub-dom ain is impossible due to the heuristic nature of the AF m ethod. The solution to both problems is crucial for Parallel unstructu red grid generation which is an enabling technology for large-scale and high perform ance sim ulations. Three DD approaches for parallel unstructured grid generation by the AF technique are presented and evaluated in the context of aerodynam ic applications. F irst, a Discrete Domain Decom position approach is presented by which the domain, defined by a coarse mesh, is decomposed in several sub-dom ains using robust graph partition ing algorithm s. This is a well known approach and is used as the base case. The second approach utilizes the oct-tree as an auxiliary d a ta struc tu re to partition an ”em pty” , i.e., unmeshed, dom ain by Recursive Coordinate Bisection (RCB). Lastly, an approach which utilizes the A dvancing-Partition (AP) strategy th a t is native to the underlying grid generator is described. In contrast to earlier dom ain decomposition approaches, the AdvancingP artition strategy benefits from inherent properties of the advancing front algorithm and is particularly suited for parallel unstructured grid generation targeting complex, real-world aerodynam ic configurations.