Department of Computer Science
Complex Flow Simulations using Heterogeneous Supercomputers
We will report on our experiences implementing a massively parallel software framework centered around the Lattice Boltzmann Method (LBM). The talk will have two parts the first one presenting our software framework WaLBerla. The software includes parallel algorithms for treating free surfaces with the goal of simulating fully resolved bubbly flows and foams. Additionally, we will report on a parallel fluid-structure-interaction technique with many moving rigid objects. This work is directed towards the modeling of particulate flows that we can represent using fully resolved geometric models of each individual particle embedded in the flow. In the second part we will present our multi-GPU implementation of the LBM and our particulate flow method.
We will provide a scalability analysis using a performance model to show that our implementation scales well on computers having several hundred GPUs, such as the Tsubame 2 in Tokyo having 1408 compute nodes each equipped with 3 NVIDIA Tesla 2050 GPUs. The talk will end with some remarks on the benefits and challenges of GPU-computing.