Heuristics in Energy Aware Real-time Scheduling Problems

Energy efficient real-time systems have been a prime concern in the past few years. Techniques at all levels of system design are being developed to reduce energy consumption. At the physical level, new fabrication technologies attempt to minimize overall chipset power. At the system design level, technologies such as Dynamic Voltage and Frequency Scaling (DVFS) and Dynamic Power Management (DPM) allow for changing the processor frequency on-the-fly or go into sleep modes to minimize operational power. At the operating system level, energy-efficient scheduling utilizes DVFS and DPM at the task level to achieve further energy savings. Most energy-efficient scheduling research efforts focused on reducing processor power. Recently, system-wide solutions have been investigated. In this work, we extend on the previous work by adapting two evolutionary algorithms for system-wide energy minimization. We analyse the performance of our algorithms under variable initial conditions. We further show that our meta-heuristics statistically provide energy minimizations that are closer to the optimum 85% of the time compared to about 30% of those achieved by simulated annealing over 500 unique test sets. Our results further demonstrate that in over 95% of the cases, meta-heuristics provide more minimizations than the CS-DVS static method.