Open Problems in Computational Steering of Massive Parallel Unstructured Grid Based CFD Simulations

Abstract

Traditionally, analysis of flow fields resulting from computational fluid dynamics (CFD) calculations is a sequential process. The flow area defined by surrounding geometry is tessellated, a mesh is generated and divided into subregions, transferred to a cluster or supercomputer and the result is transferred back. Then, a variety of post-processing tasks should give insights to the physical problem. At that point, parameters chosen wrong can be identified and the simulation has to be done again with tweaked parameters. This is an iterative process that can be time consuming, especially if one iteration lasts more than a few days. In general, aiming at reducing the simulation times by shortening the time used to identify wrong parameters results in high productivity enhancements. 
In this paper, the need for on-line monitoring and computational steering approaches for massive parallel unstructured flow simulators are presented with aircraft design as one of many possible application domains. This involves software integration aspects, data streaming and explorative visualization. Many challenges still have to be solved and this paper summarizes most important ones.