Research On Node Deployment And Resource Allocation For Mobile Edge Computing

With the rapid development of the mobile communication network,the demand for lower latency and more diverse mobile applications is increasing.Therefore,in order to provide various and higher-quality application services and get rid of the resource limitations of users' equipments,Mobile Edge Computing(MEC)came into being.Mobile edge computing sinks computing power from the cloud to network edge nodes(MEC nodes),and users access these edge nodes to obtain required services.Due to the influx of a large number of smart devices,a large number of MEC nodes are required to provide corresponding services.However,deploying too many MEC nodes will result in waste of cost.With insufficient MEC nodes,user requests need to be remotely scheduled or wait for close but heavy loaded MEC nodes to process,which will greatly increase the request response delay.Therefore,how to place MEC nodes to provide users with high-quality services and control operator costs under dynamically changing network conditions has become an urgent problem to be solved.Based on the above problems,this thesis designs a MEC node placement mechanism based on online adjustment from the perspective of meeting users' needs while minimizing operating costs.Furthermore,considering the diversified needs of users for services and preventing the remote cloud from being overburdened,a MEC service placement mechanism based on federal reinforcement learning was designed.The specific research contents are as follows:(1)MEC node placement mechanism.The mechanism consists of the initial placement of MEC nodes and the online adjustment of MEC nodes.First,considering the user requirements and network characteristics in the early stage of the network,based on the historical average load of the node,the initial placement problem of the node is constructed as a coverage problem.The backtracking algorithm of depth-first search is used to obtain the optimal MEC node initial place strategy.With the continuous change of the network state,when the current node placement strategy can no longer meet the requirements,the node placement needs to be adjusted.This thesis takes user QoE as an important indicator and makes a judgment on the adjustment timing through a fuzzy neural network.After deciding to adjust,make intelligent decisions based on deep reinforcement learning algorithms while considering multiple costs and constraints.The simulation results show that the proposed node placement mechanism can guarantee the quality of service in a dynamic environment while minimizing the operating cost.(2)MEC service placement mechanism.Considering the more diversified service requirements of users and the heavy burden of remote cloud,combined with the research trend of more intelligent communication networks,this thesis designs a MEC service placement mechanism based on federal reinforcement learning.This mechanism considers minimizing latency and cost,as well as privacy and security issues during data sharing.At the same time,in order to solve the problem of limited node resources,collaboration between nodes is considered.The mechanism uses reinforcement learning to treat a single node as an agent to make its own caching and collaborative decisions,and to train a decision model that is more adaptable to the environment by combining federated learning.Considering the problem that the incomplete information environment among distributed nodes may lead to resource competition,a priority based conflict mitigation strategy is designed.The strategy makes decisions on the priority of conflicting nodes and avoids conflicts caused by competing resources simultaneously.Simulation results show that the service placement mechanism proposed in this thesis can save more costs and has faster training speed and environmental adaptability.