Steering and visualization: Enabling technologies for computational science

Abstract

Computational steering is an investigative paradigm whereby the parameters of a running program can be altered according to what is seen in the currently visualized results of the simulation. For certain problems, interactive computation brings specific benefits: parameter sweeps can be completed more efficiently by quickly identifying combinations of input values that yield nonsensical results; ‘what-if’ studies may elucidate a number of related hypotheses without computing all of them from scratch; even simply tracking the development of a computation on-line may allow its early termination (with consequent resource savings) if it turns out its set-up was flawed for some reason. Having first come to prominence in the 1990s, the take-up of steering has accelerated in recent years: high performance computer (HPC) facilites have become more widely available, usage of the Grid is increasing, and visualization is experiencing take-up beyond the walls of the pioneering graphics laboratories that first nurtured it. All these factors combine to expand the range and difficulty of the scientific problems that can be tackled by steering. However, this benefit comes at the cost of increasing hardware and software complexity which itself may defeat further progress. This paper discusses these issues and, in the light of some users’ recent experiences, charts the challenges that still face us.