Research On Energy-efficient Scheduling For Parallel Applications On Heterogeneous Computing System

In the wave of the rapid development of computer information technology and integrated circuit technology,more and more large connected Internet of Things,high-performance processors whose functions tend to be specialized and subdivided,and large computing platforms with more and more complex computing requirements all promote the trend of isomerization of computing systems.In the current information society in the era of big data,access to information and computing resources at any time and anywhere has become a common basic life demand of the public.It can be predicted that in today's computing system,the amount of computing tasks required by the system is showing explosive growth.Corresponding to this will be a huge demand for energy,and more and more data centers will consider energy as the first factor in location.Therefore,the study of high energy efficiency scheduling algorithm under heterogeneous system is not only in line with the current actual economic demand,but also in line with the direction of technological development.In recent years,the topic of efficient computing task scheduling strategy has been widely studied.Among them,Dynamic Voltage and Frequency Regulation(DVFS)has become an important energy management technology by reducing the power supply voltage and frequency of the processor simultaneously to regulate energy consumption.This paper focuses on the application scenarios of heterogeneous distributed computing systems,scheduling the static,non-independent parallel applications with priority constraint relationship between tasks,and solving the problem of optimizing the completion time of parallel applications under a basic energy consumption constraint.In this paper,we first discuss the traditional scheduling algorithm based on priority queue and introduce the energy preallocation mechanism to deal with the problem of energy consumption constraint.By analyzing the limitations of the existing pre-allocation strategy,a heuristic scheduling algorithm ESSPP based on proportional pre-allocation of energy relaxation space is proposed to further enhance the rationality of the pre-allocation scheme.Finally,the idea of genetic evolution random search is further introduced in the completed work,and the proposed efficient scheduling algorithm is extended.The adaptive problems of the following genetic algorithms in the task scheduling scenario are discussed in detail:how to encode chromosomes?How to realize the three genetic operations of natural selection,chromosome crossing and chromosome variation?On the basis of ESSPP algorithm,a high efficiency scheduling algorithm ESBGA based on genetic algorithm is proposed.Finally,the effectiveness of the two proposed algorithms is verified by several simulation experiments.