Performance evaluation of on-line scheduling algorithms for imprecise computation

This paper presents an algorithm for scheduling imprecise tasks to minimize maximum error and analyzes its performance through intensive simulation. The imprecise computation is used to manage transient overload of computation. Each task for imprecise systems consists of a mandatory part and an optional part that can be skipped when systems are overloaded. The imprecise computation trades accuracy for meeting the deadline of tasks by skipping their optional parts. To increase accuracy for imprecise computation, we present a simple on-line scheduling algorithm that minimizes the maximum error. This algorithm executes mandatory parts of tasks first, then the optional parts in order to minimize maximum error. The proposed algorithm also increases schedulability by executing mandatory parts first. The proposed algorithm is simple and does not require any additional data structures such as the reservation list for maintaining mandatory parts.