dDVS: An efficient dynamic voltage scaling algorithm based on the differential of CPU utilization

Abstract

Traditional dynamic voltage scaling algorithms periodically monitor CPU utilization and adapt its operating frequency to the upcoming performance requirement for CPU power management. Predicting CPU utilization is usually conducted by estimating upcoming performance requirement. In order for dynamic voltage scaling algorithms to be effective, the prediction accuracy of CPU utilization must be high. This paper proposes a power management algorithm that improves accuracy of predicting future CPU utilization using process state information. Experiments show that the proposed algorithm reduces power consumption by 11%-57% without any performance degradation.