Related work

increase the utility of high-performance simulations for scientists because it lets them drive the scientific discovery process and interact with their data. They can investigate, calibrate, and correct the execution of their simulations. For example, computational steering lets users interactively maneuver molecule docking in a protein-design simulation,1 interactively examine and control structures’ shape optimization,2 and manage the process of simulated annealing.3 Steering provides two major benefits: it lets users diagnose their simulation models, and it provides insights into the complex runtime interactions of the simulation components. Computational steering is the online management of application execution and application resources to improve performance or explore an application (see the “Computational-steering issues and taxonomy” sidebar).4 Initially, implementations of computational-steering systems focused on functionality with limited effort dedicated to understanding the performance of the steering-system architecture. Furthermore, application users customized many of these designs to specific operating environments and applications; therefore, these approaches were difficult to leverage across applications and target platforms. Nevertheless, interest in techniques for computational steering persists (see the “Related work” sidebar). Recent interest in virtual environments4 has further fueled the need for an efficient, dynamic infrastructure for high-performance computational steering. Three reasons motivate the need for high performance in computational steering. The steering system must