FSSteering: A Distributed Framework for Computational Steering in a Script-based CFD Simulation Environment

Abstract

In order to get insight into interesting flow phenomena, the traditional work-flow of computational fluid dynamics (CFD) consists of setting up and computing the flow field followed by a consecutive post-processing analysis. Only after this analysis one can identify parameters that may have been set wrongly in a configuration stage. Once these parameters are corrected, another time-consuming loop has to be started. To identify inadequate parameter settings already during the simulation run, online monitoring concepts were introduced. Combined with computational steering methods, parameter values can additionally be adjusted which eventually reduces the number of required iterations to yield satisfactory results. At the German Aerospace Center, a comprehensive framework called FlowSimulator has been developed to offer a generic Python-based interface for the management of CFD simulations. It can easily be enhanced by add-ons. One of these extensions is FSSteering which is described in this paper in more detail. As a computational steering environment, FSSteering provides functionalities essential for interactive visualization and explorative analysis. Besides existing computational steering environments and frameworks, a user-centred and domain-specific view is proposed. Existing functionality can be reused without rewriting simulation code to enable for effective steering in CFD. To be more efficient, components of the architecture are distributed across different resources. Whereas the CFD simulation typically runs on a parallel supercomputer, the visualization is carried out on a high-performance virtual reality system which allows interactive data exploration. The post-processing in between can be performed on the supercomputer or on a separate parallelization cluster. But it is also possible to switch between different existing post-processing toolkits. This is just possible because of the very flexible configuration management of the distributed steering framework. We will demonstrate the steering capabilities and the system flexibility by two current research examples. An outlook for future steps concludes this paper.