Research On Energy-Efficient Scheduling Techniques For Real-Time Systems

Recently, research on energy consumption problem becomes more and more important in the design of real-time systems. Energy-efficient scheduling techniques can reduce processor energy consumption effectively by combining real-time scheduling theory with dynamic voltage scaling (DVS), and can be applied into the real-time energy-efficient design. The thesis makes a thorough study on energy-efficient scheduling techniques for hard real-time systems.Firstly, this paper respectively discusses the effects of practical DVS processor limitations on the ideally optimal offline energy-efficient scheduling method, such as the effect of transition time, energy trasition overhead and that of the voltage-level and frequency-level discretization. For the multidimensional limitations mentioned above, this paper proposes a new method called selecting idle time, which has turned out better in theory than the one called selecting busy interval, and also describes a better unified technique to optimize offline energy-efficient scheduling.Secondly, this paper deeply analyses an excellent online energy-efficient scheduling algorithm called OLDVS, and shows the causes which result in the algorithm's shortcoming. In order to improve energy savings of this algorithm, this paper also presents various new means and designs. To evaluate the performance of the energy- efficient algorithm, a simulator is implemented. Moreover, this paper also proposes the design structure of simulation experiment on energy-efficient scheduling techniques.Thirdly, according to the shortcoming of OLDVS algorithm such as no adaptive abilities to dynamic changes of task execution, this paper proposes an energy-efficient scheduling algorithm of hard real-time mixed tasks based on Assistant Queue (OLDVS-AQ). This algorithm can use dynamic slack time more effectively to reduce energy consumption largely. The feasibility of the algorithm is proved, and the simulation results show that the proposed algorithm always improves energy saving about 10% on average more than OLDVS algorithm. Meanwhile, the performance gap becomes much larger as the aperiodic factor increases.Lastly, according to the fact that OLDVS algorithm achieves less energy saving when the ratio of the computation requirement of periodic tasks to total computation requirement is higher, this paper proposes an energy-efficient scheduling algorithm for hard real-time mixed tasks based on Simple Feedback (OLDVS-SF). This algorithm can reduce energy consumption more effectively when scheduling periodic tasks and consequently achieve more energy saving. Moreover, the feasibility of new algorithm is also proved, and the simulation results show that the new one always outperforms OLDVS algorithm, and can improve energy saving about 10% at most and about 5% on average when the ratio of the computation requirement of periodic tasks to total computation requirement is higher.