Research On Placement Algorithm For Video Processing Service In Edge Computing

With the rise of the Internet of Things,various intelligent devices emerge in endlessly,providing users with various services,in which video services such as intelligent surveillance have developed rapidly,and cameras have flooded people’s lives.However,due to the limitations of camera size and cost,the computing power is not enough to support video processing,therefore,it is necessary to deploy computing power at the edge to support it,then,edge computing is introduced.Edge computing is different from centralized cloud computing,which has a large number of homogeneous resources.There are two main problems in edge computing: first,resources are very limited? second,edge computing servers have great heterogeneity.In addition,in edge computing scenarios,video processing services are divided into multiple microservices and form a microservice chain,rather than a whole.So,how to place video processing services has become a problem to be solved.To address this issue,thesis studies the placement algorithm of video processing services in edge computing scenarios.While in edge computing,the core challenge is the latency issue,so the main purpose of this paper is to reduce service response delay.Thesis mainly studies two application scenarios: one is the static scenario,in which several cameras need to access the edge computing network at the same time,so multiple microservice chains need to be placed jointly? The other scenario is a dynamic scenario.The cameras will dynamically access the network at different times,and may exit at a later time.The contributions of this thesis mainly include two points.First,aiming at the situation that multiple cameras’ microservice chains are placed at the same time in static scenes,thesis proposes a multi-objective static placement algorithm for video processing services in edge computing based on ant colony algorithm,which is called MSP-VPS.This algorithm proposes a method to reduce the average response delay and improve the square sum of remaining resource ratios of servers in the case of heterogeneous resources between edge computing servers,differences in the amount of resources that can be allocated to microservices,and incomplete connections between servers.Besides,thesis uses markov decision process(MDP)to model the dynamic placement process of video processing services in edge computing,and proposes the DP-VPS(Dynamic placement algorithm of video processing service)algorithm based on the idea of deep reinforcement learning(DRL)and the proximal policy optimization(PPO)algorithm.The DP-VPS algorithm optimizes the average response delay of the microservice chain in a future period of time when the cameras is time-varying,so that the edge computing server can provide users with low latency high-quality video processing services with limited resources.Finally,by using randomly generated camera datas to design simulation experiments for the algorithms in thesis,in two different problem scenarios,I compared the performance of proposed algorithms with various algorithms in terms of average service response delay,edge computing server resource utilization and weighted target value.The experimental results show that the algorithms proposed in thesis have better performance in reducing the average response delay of services and increasing the utilization of server resources.