Computational steering on distributed systems: Indoor comfort simulations as a case study of interactive CFD on supercomputers

摘要

This paper presents the current state of a computational steering system for interactive computational fluid dynamics (CFD) simulations, allowing engineers to simulate interactively indoor climate and to evaluate human comfort. The tools presented support cooperative planning and design by providing means for interactively adding, removing and modifying geometry and boundary conditions online during a CFD simulation. To ensure interactivity and short-latency updates even for high-resolution runs the parallel Lattice-Boltzmann-based simulation kernel is optimized for high-performance computing systems. Emphasis is placed on the computational steering architecture, connecting a supercomputer with one or more visualization workstations. In particular, we show how a high-performance communication between simulation and visualization or steering front-end can be achieved together with preserving a flexible mechanism of on-the-fly attachment of multiple cooperation clients.