Energy-Aware Scheduling for Tasks with Mixed Energy Requirements

Two widespread techniques, Dynamic Voltage/Frequency Scaling (DVFS) and Dynamic Power Management (DPM), rely on speed reduction and low-power state exploitation, respectively, to reduce energy consumption in modern computing platforms. Traditionally, dynamic power was the main source of processor power consumption, therefore DVFS was more effective than DMP to reduce energy consumption in processing elements, whereas DPM was mainly used to shut down system components not currently in use. As the technology continues to miniaturize processors, leakage power has become dominant with respect to dynamic power consumption, hence today most people believe that DPM is the most effective technique also in processing units. This paper shows that, considering a more realistic task model derived from the analysis of actual executed code, tasks may exhibit different energy requirements, so requiring an integrated DVFS-DPM approach. I. I NTRODUCTION