Research On Task Scheduling And Service Function Migration Of Edge Computing Server

Mobile Edge Computing(MEC)brings powerful computing power near the devices of the Internet of things to ensure latency constraints,making it one of the key technologies to support intelligent applications of the Internet of things in the future.In the edge system with multiple subnets,the number of devices and types of applications of the Internet of things in different subnets are different,which leads to the uneven distribution of computing load.In heavily loaded subnets,because the resources of edge servers are limited,it may be difficult to meet some computing tasks,while edge servers deployed in light subnets may have rich idle resources.If the edge server with free resources can release some overloaded tasks from other subnetworks,it will meet more applications in the edge network.Therefore,collaboration between edge computing servers with limited resources is an effective solution to improve the performance of edge networks.However,there are still many idle resources in the edge network,resulting in a waste of resources.Therefore,aiming at the cooperation scenario between edge servers,this thesis considers making use of the free resources of private edge servers in the edge system to bring private edge servers into the goal of user task unloading by means of paid lease.The main research contents and innovations of this thesis are as follows:This thesis studies the task scheduling and resource allocation strategy under the cooperative computing mode of joint private edge servers.Firstly,the task offload process is modeled,and the delay and resource constraints caused by task offload are considered in order to maximize the overall benefit of the public edge server.Considering the privacy characteristics of the private edge server,in order to make better use of the free resources in the private edge server,the historical load data of the private edge server is used to predict the load through the ARMA model.The process of task unloading and resource allocation is described as a Markov decision process,and the corresponding system reward is obtained according to the actions of different allocation strategies as system actions.Finally,based on Markov decision process,a task scheduling and resource allocation algorithm based on deep deterministic policy gradient reinforcement learning(DDPG)is designed.Numerical results show that the algorithm can improve the success rate of user task execution and the overall benefit of public edge servers while ensuring delay constraints.It can make effective use of the computing resources of private edge servers in a resource-tight environment.The edge server needs to deploy the special code and the service function of the database to respond to the class service request,the user’s service request is time-varying,and the fixed service function deployment strategy will reduce the system performance.Therefore,this thesis studies the service function migration and task allocation(SFMTA)strategy under the collaborative computing mode of federated private edge servers.Firstly,a dual time scale framework is designed to separate the SFMTA decision from the time scale.Then based on this framework,a SFMTA algorithm based on Double-Layer DDPG is designed,which jointly optimizes the SFMTA decision under the storage and computing resource constraints of the edge server to maximize the operator’s revenue.The upper DDPG makes the decision of service function migration on the time frame,taking into account the delay cost and expense in the process of service function migration.The lower-level DDPG makes task allocation decisions on the time slot scale.Finally,the simulation results show that,compared with the traditional method,the Double-Layer DDPG algorithm proposed in this thesis can maximize the average time benefit of the operator,while ensuring the average response time of the system,it can make the uninstall task of almost all users in the system execute successfully.It makes full use of the resources of the private edge server in the edge system,and realizes the reasonable allocation of users’ unloading tasks and the overall optimal allocation of resources.There are totally 29 figures,2 tables,and 85 references in this dissertation.