Algorithms for scheduling periodic jobs to minimize average error

Several preemptive, priority-driven algorithms for scheduling periodic jobs on systems that support imprecise computations are described and evaluated. The algorithms are designed to keep the average error in the results produced over consecutive periods small. The approach taken here is to consider each task as consisting of two parts: a mandatory part that must be completed in order for the task to produce an acceptable result, and an optional part that refines the result produced by the mandatory part to reduce the error in the result. The mandatory parts of all tasks have hard deadlines; the rate-monotone algorithm is used to schedule them to meet all deadlines. The optional parts have soft deadlines; different algorithms are used to schedule the optional parts to minimize the average error. The performance of these algorithms is evaluated in terms of the average error over all jobs as a function of their total utilization factor