User-controlled optimization of task scheduling for imprecise computer systems

Abstract The performance of imprecise computer systems with optimized task scheduling is studied. An imprecise computer system trades off computational accuracy for faster response time of tasks (when necessary) and is therefore potentially useful in real-time applications. In the authors' system, tasks arrive randomly during runtime. Each task has two levels of computation time requirements: the full-level computation requirement and the reduced-level computation requirement. The reduced-level computation of a task takes less time to accomplish than the full-level one and produces results that are acceptable as, though less precise than, those of the full-level computation. The tasks are scheduled in the following way: if the total number of tasks in the system is no more than M — a system parameter — the tasks are executed at the full level; otherwise, the tasks are executed at the reduced level. An optimization procedure that is driven by user requirements is studied. It is proposed to select the system parameter M such that an objective function is optimized and a set of user-specified requirements is met. The optimization procedure aims to give the user the control in determining the parameter M according to performance requirements. The performance of the system under the control of the optimization procedure is examined.