high performance computing on graphics processing units: hgpu.org

Despite the experience of several decades the numerical simulation of computational fluid dynamics is still an enormously challenging and active research field. Most simulation tasks of scientific and industrial relevance require the modeling of multiple physical effects, complex numerical algorithms, and have to be executed on supercomputers due to their high computational demands. Facing these complexities, the reimplementation of the entire functionality for each simulation task, forced by inflexible, non-maintainable, and non-extendable implementations is not feasible and bound to fail. The requirements to solve the involved research objectives can only be met in an interdisciplinary effort and by a clean and structured software development process leading to usable, maintainable, and efficient software designs on all levels of the resulting software framework. The major scientific contribution of this thesis is the thorough design and implementation of the software framework WaLBerla that is suitable for the simulation of multi-physics simulation tasks centered around the lattice Boltzmann method. The design goal of WaLBerla is to be usable, maintainable, and extendable as well as to enable efficient and scalable implementations on massively parallel supercomputers. In addition, a performance analysis of lattice Boltzmann simulations has been conducted on heterogeneous supercomputers using a MPI, a hybrid, and a heterogeneous parallelization approach and over 1000 GPUs. With the help of a performance model for the communication overhead the parallel performance has been accurately estimated, understood, and optimized. It is shown that WaLBerla introduces no significant performance overhead and that efficient hardware-aware implementations are possible in WaLBerla. Furthermore, the applicability and flexibility of WaLBerla is demonstrated in simulations of particle flows and nano fluids on CPU-GPU clusters. By the successful application of WaLBerla in various simulation tasks, and the analysis of the performance and the software quality with help of quality criteria, it is shown that the design goals of WaLBerla have been fulfilled.