SC²S Colloquium - November 20, 2013

From Sccswiki
Jump to navigation Jump to search
Date: November 20, 2013
Room: 02.07.023
Time: 3 pm, s.t.

Kilian Röhner: Effiziente Algorithmen zur Lösung der Black-Scholes Gleichung auf dimensionsadaptiven dünnen Gittern

With the Lattice Boltzmann Method, fluid and areodynamical flow problems can be simulated without solving the Navies-Stokes equation. Flow problems with practical application are dominated by turbulent structures. To resolve these turbulent scales in space and time with the Lattice Boltzmann Method (LBM) on equidistant grids, tremendous computational resources are required. Thus, grid refinement techniques were developed for the LBM which are based on different coupling formulations. The accuracy is determined by the resolution of the underlying grid and increases with decreasing the cell width. Grid refinement techniques raise the possibility of performing high resolution simulations and decreasing the computational effort at the same time.

In this talk, several approaches to refine the grid will be discussed. Some of them were implemented into case examples written in the Open-Source Lattice-Boltzmann Framework OpenLB. The adaption of the grid geometry as well as the interaction between different refinement levels will be explained in detail. To validate the properties of the refinement strategies, generic benchmark test cases will be presented and discussed.

Arash Bakhtiari: MPI Parallelization of GPU-based Lattice Boltzmann Simulations

In this thesis, an MPI parallelized LBM Code for a Multi-GPU platform has been designed and implemented. The primary goal of the thesis is research on efficient and scalable LBM code, which exploits advanced features of the modern GPUs, to adopt optimization techniques like overlapping of work and communication in heterogeneous CPU-GPU clusters. Therefore, three techniques of SBK-SCQ, MBK-SCQ and MBK-MCQ have been designed and implemented. In addition, based on the profiling results, three optimization techniques for efficient boundary values memory access pattern on the GPU memory are developed.

In weak scaling experiments on the MAC GPU cluster with 8 GPUs, the SBK-SCQ has achieved 97% efficiency by 4 GPU as the baseline. Finally, the software is extended to the Smagorinsky subgrid-scale turbulence model.