Parallel unstructured grid generation for computational fluid dynamics simulations of three-dimensional, complex aerodynamic configurations is a challenging problem for two reasons: (1) domain decomposition, which decomposes the computational domain into several smaller subdomains that are meshed in parallel and (2) load balancing, in which the subdomains are assigned to a set of processors so that the load is evenly distributed across them.
At the same time, the desire to perform higher fidelity and more accurate simulations along with the widespread availability of parallel machine architectures and the ability of flow solvers to take advantage of parallel architectures creates the requirement for large-scale, high resolution grids. Generating such grids sequentially is difficult due to both CPU and memory limitations. Thus, parallel unstructured grid generation is an enabling technology for large-scale, high performance simulations and is the primary objective of this work.
Three methods for parallel unstructured grid generation are presented and evaluated in the context of aerodynamic simulations. First, a discrete domain decomposition approach is presented by which the domain, defined by a coarse mesh, is decomposed into several subdomains using robust graph partitioning algorithms. Next, an approach is described by which an oct-tree serves as an auxiliary data structure to guide the partitioning. Finally, an approach is described that, in contrast to earlier domain decomposition approaches, benefits from inherent properties of an advancing front algorithm and is particularly suited for parallel unstructured grid generation targeting complex, real-world aerodynamic configurations.