Dynamic voltage scaling for real-time scheduling of multimedia tasks

Abstract

Real-time multimedia applications are usually played back many times. For those applications, the distribution of actual execution time is no longer unknown from the second playback. In this paper, we propose a novel dynamic voltage scaling (DVS) algorithm, called CLDVS, for scheduling real-time multimedia applications. In order to minimize energy consumption, CLDVS determines the processor's operating frequency and supply voltage using the distribution of actual execution time during a time interval at the beginning of the interval. For that, all active tasks in the time interval are identified and incrementally placed on the time vs. scaling factor space in order to reduce variations of the scaling factor for minimum energy consumption. Simulation experiments show CLDVS achieves enormous energy savings and outperforms the existing DVS algorithms with different dynamic workload characteristics. © Springer-Verlag Berlin Heidelberg 2005.