Design and Analysis of Processor Scheduling Policies for Real-Time Systems

In this paper we study the problem of scheduling jobs with real-time constraints in single and multiprocessor computer systems. We show the optimality of the minimum laxity (ML) scheduling and earliest deadline (ED) scheduling policies on multiprocessors under general workload assumptions for systems in which jobs need not be served once they miss their deadlines. We also describe a discrete-time model of these policies operating on a single processor when deadlines of all jobs are uniformly bounded. The ML and ED policies described incur an overhead that has computational complexity of O(m) or O (log (m) (depending on the implementation) where m is the queue length. Hence we propose and study several efficient policies (with O (1) overhead) that provide most of the performance of the ML and ED policies. Last, we consider the problem of scheduling jobs with real-time constraints that provide increased reward as a function of execution time. We propose several greedy policies that attempt to equalize the amount of service that all jobs attain and show, through simulation and analysis, that they attain performances close to an unachievable optimistic bound.