Research On Parallel Task Scheduling Algorithm In Edge Computing

With the development of the Internet and big data technology,edge computing has constantly integrated into people's lives.In some edge computing scenarios,mobile users can offload tasks which are produced by their own devices to nearby edge clouds for low-latency,high-efficiency services.How to efficiently and reliably schedule tasks offloaded to the edge cloud is an important issue in edge computing.Mobile devices in the daily live has been more and more intelligent,and the types of services that can be provided are also increasing.However,due to computing resources and battery capacity limitations of these devices,they could not satisfy the demand which include the quality of service and response time of some computation-intensive services.One way to solve these problems is offloading these computationally intensive tasks to remote cloud computing centers which has more computing power and more energy than user's intelligent device.However,these cloud computing centers have a long distance from the user's mobile devices.Offloading user's tasks to remote cloud computing centers will generate a large communication delay,and some delay-sensitive services cannot get prompt response.Edge computing technology has satisfied the requirements of computationally intensive and delay sensitive applications for computational resources and response time.But the computing resources and battery capacity of edge servers are limited relative to the remote cloud,task processing efficiency and total resource among edge servers are often not the same exactly,resource competition always exists in the tasks.Therefore,ho w to design a scheduling algorithm for competitive tasks on edge servers is a significant scientific problem in edge computing.A good scheduling algorithm should make edge servers could handle numerous user's tasks with limited computing resources and battery capacity,and ensure user's service experience.Int order to solve the task scheduling problem in the edge computing,We proposed two different methods.The main work is as follows:(1)We invoked the P-BPM task schedule mode to solve the schedule problem in edge cloud,proposed a Parallel-batch multi-objective job scheduling algorithm based on the ant colony algorithm,called the P-MACO algorithm.We proposed a method to save running overhead for the phenomenon that the difference in processing efficiency due to different performance parameters between edge servers and processing jobs timely with considering their deadlines.(2)We designed an online parallel scheduling algorithm with reinforcement learning theory that can be applied in the edge computing environment and suitable for random task generation,called Online Parallel-batch Q-Learning Algorithm(OnPQ-Learning algorithm).We simulated the multi-edge servers task scheduling problem to a Markov decision problem,and find the appropriate scheduling strategy with Q-learning method.In order to ensure the service response timely and save the resources of the edge server,we defined the optimized objectives as minimize the number of fail tasks which overstep their response time deadline,energy required of the edge server and the cost and the average task waiting time,so that the edge server could provide users with a large number of high-quality services with their limited resources.In addition,we made some insights about calculate the edge server's time and energy in our algorithm.(3)We designed the simulation experiments to compare our algorithms with some existing scheduling algorithms to evaluate the performance based on Google cluster dataset.We compared our algorithms with MACO algorithm,FCFS algorithm and OnDisc algorithm in different application scenarios.The results of experiments proved that our algorithm outperformed all of the compared algorithms significantly in terms of saving resources of edge servers and guaranting processing tasks timely.