Sen - R 9718 1997

Computational steering is a technique that combines simulation and visualization. The user is continuously provided with visual feedback about the state of the simulation, and can change parameters on the y. Designers can vary parameters to optimize their product, users can detect errors in the input early, researchers can do qualitative sensitivity analyses easily. The implementation of computational steering is very tedious. It requires knowledge of the simulation, visu-alization, user interfacing, and data communication. In this paper we discuss an environment that enables users to implement and use computational steering eeectively without much support from user interface experts. We show how the environment is applied to various applications. 1. INTRODUCTION Visual supercomputing has become a hot topic. Both simulation and visualization have boosted in recent years. Many methods, techniques and tools are available nowadays for the simulation and visualization of complex phenomena. Current computing hardware enables researchers and engineers to perform complex simulations within seconds. The results are visualized with high end graphics workstations, which generate colorful pictures in real-time. Computational steering is located at the cross-point of these developments. With computational steering the user is continuously provided with visual feedback about the state of his simulation, and can change parameters on the fly. Designers can vary parameters to optimize their product, users can detect errors in the input early, researchers can do qualitative sensitivity analyses easily. Of course, standard methods can also be used: Just perform simulations and display the results afterwards. But we and others [2] believe that computational steering accelerates the process of gaining insight in complex simulations with orders of magnitude, and thereby introduces a new paradigm. If computational steering is such a great idea, why isn't it routinely used yet? Why are new values for parameters still entered via text files in obscure formats or even by changing the code itself? Why are simulations done until 99.99 % accuracy is achieved, only to observe after the results are converted and displayed with a visual-ization package that a crucial parameter has a wrong value? The answer is simply that the implementation of computational steering is hard. It requires knowledge of the simulation, visualization, user interfacing, and data communication. An expert in a field such as CFD, molecular dynamics, or mechanical engineering typically will have to cooperate with experts on user interfacing and visualization – which in many cases will be prohibitively expensive. Even if this is …