| With the continuous development of IoTs,5G and other technologies,a lot of terminals and devices are connected to the network,many of which are very sensitive to time delay.In order to make up for the high network delay caused by long-distance transmission in cloud computing,edge computing is introduced.On this basis,edge operators who do not have basic hardware resources rent edge servers and pay a certain fee to provide edge services for users,so as to effectively deal with massive computing tasks in the network.However,the amount of edge server resources is limited by software and hardware.Decentralized deployment makes the network structure complex,the quantity and types of equipment in the edge network are also complex,and the network state changes rapidly and frequently,so that the complexity of task unloading and resource allocation often leads to high network delay and consumption,which can’t fully meet the requirements of today’s network,Therefore,strategy optimization is needed.Therefore,based on deep reinforcement learning,this paper proposes task unloading and edge computing resource allocation strategy,and designs and implements heterogeneous edge network resource management module.For the task unloading problem in edge computing,this paper mainly considers the selection of computing task unloading direction of equipment.n order to reduce the overall network delay,this paper establishes the edge Internet of things network model,and considers the performance and requirements of various types of equipment in the model to form a more perfect model.Meanwhile,considering the complexity and variability of the scene,this paper designs a knowledge driven model based on the PPO(Proximal Policy Optimization)algorithm,connects the neural network with the PPO algorithm,preprocesses the complex network data,realizes the transformation from data-oriented to knowledge-oriented,and forms a more efficient and stable task unloading strategy.After the computing task is transferred to the edge server by the device,the server needs to allocate appropriate computing resources for the computing task.For the problem of computing resource allocation of edge server in edge computing,in order to reduce the network delay and server consumption as much as possible,this paper establishes the structural model of edge server network,and introduces A3C(Asynchronous Advantage Actor Critical)algorithm for resource allocation.For server failure in the network,this paper adopts the task migration strategy,decomposes the task migration into the process of target selection,migration and resource redistribution,designs the target scoring and selection scheme,and integrates it into the process of resource allocation,so as to form a computing resource allocation mechanism with excellent performance and rapid response to network failure.Based on these two algorithms,this paper analyzes the various needs of users,designs the system architecture,functional sub modules,system process and database,and provides the trust and storage scheme of historical behavior,so as to realize the heterogeneous edge network resource management module based on the lightweight Kubernetes architecture,test the system function,and provide users with the function of simulating or connecting the actual edge network,resource management and benefit allocation,It has certain research and practical significance. |
