Research On Task Energy-Efficient Scheduling For Multicore Processors

With the advancement of semiconductor technology,the scale of real-time embedded systems has grown tremendously.The traditional method of improving processor performance by increasing the processor frequency is no longer suitable for modern processors.Researchers in industry and academia have shifted their research focus on multi-core processor architecture,especially heterogeneous multi-core processors with better performance and greater applicability.However,most real-time embedded systems are battery-powered or energy-limited devices.Therefore,the research of task energy-efficiency scheduling algorithm based on heterogeneous multicore processor platform has important research significance.This thesis studies the energy efficiency optimization of task scheduling on heterogeneous multi-core processors,and proposes a task energy-efficiency scheduling algorithm base on reinforcement learning methods.In addition,this thesis designed a hybrid DVFS(Dynamic Voltage-Frequency Scaling)technology,combined it with task allocation algorithms,and applied to periodic task scheduling on heterogeneous multicore processor platforms.The main contents of this thesis are as follows:1.Propose a task energy-efficiency scheduling algorithm based on deep reinforcement learning,Utilize the Markov property in the task scheduling process,use deep reinforcement learning methods with experience replay to interact with the system environment,complete task scheduling decisions.It can be seen from the simulation experiment results that compared with the traditional heuristic algorithm,this algorithm has the highest scheduling reliability.When scheduling task sets with medium and low load,the energy consumption of this algorithm is 10%?51% lower than traditional heuristic algorithm.In extreme cases,the scheduling success rate of this algorithm is more20% higher than traditional heuristic algorithm.2.Improve on the basis of task energy-efficiency scheduling algorithm based on deep reinforcement learning,and use the double deep Q learning network as the training model to avoid overestimation in the scheduling decision process and enhance the stability of the algorithm.A hybrid DVFS scheduling algorithm composed of three DVFS technologies is designed.The appropriate DVFS strategy is selected according to the state of the processor core,and the processor core voltage and clock frequency are flexibly configured.Simulation experiments show that compared with traditional heuristic algorithms,energy consumption is reduced by 25% to 50%.Compared with the algorithm proposed in Chapter 3,when the processor core parameters fluctuate,the energy efficiency advantage of the algorithm is greater,so the algorithm is more adaptable and more stable in the changing task scheduling environment.