Deadline scheduling and power management for speed bounded processors

Abstract

In this paper we consider online deadline scheduling on a processor that can manage its energy usage by scaling the speed dynamically or entering a sleep state. A new online scheduling algorithm called SOA is presented. Assuming speed can be scaled arbitrarily high (the infinite speed model), SOA can complete all jobs with reduced energy usage, improving the competitive ratio for energy from 22α-2αα+2α-1+2 (Irani et al. (2007) [17]) to αα+2, where α is the constant involved in the speed-to-power function, commonly believed to be 2 or 3. More importantly, SOA is the first algorithm that works well even if the processor has a fixed maximum speed and the system is overloaded. In this case, SOA is 4-competitive for throughput and (αα+α24α+2)-competitive for energy. Note that the throughput ratio cannot be better than 4 even if energy is not a concern. © 2010 Elsevier B.V. All rights reserved.