Design And Implementation Of Container-Based Edge Intelligence Platform

With the continuous development of IoT and AI technologies,the deployment of AI applications on end devices can provide great potential for industry application and convenience to daily life as well.However,cloud computing suffers from the inability to fully utilize infrastructure resources on edge nodes.Great computing demand also brings about huge burden on cloud resources,leading to large service response time,risks to user privacy and security.Edge computing solves the problems of cloud computing resource pressure and long service response time by deploying computing resources on network edge,while the complexity of AI poses a great challenge for the integration of AI applications and edge computing.Therefore,it is important to provide an efficient edge intelligence platform for deploying AI applications in edge computing environments.To address the above problem,this thesis aims at designing and implementing a containerized edge intelligence platform for data-intensive application.Selected cloud-native Kubernetes cluster framework as the infrastructure,considering containerization technology as the application carrier,microservice architecture as the service paradigm,and face recognition as a typical data-intensive AI application,the main research work and innovation of the thesis include the following aspects:(1)Based on the investigation of existing edge computing platforms and containerization technologies,a framework is proposed called Container Based Edge Intelligence Platform(CB-EIP)by adopting Kubernetes as container orchestration platform.The communication problem at the edge of the cloud is solved based on the cloud-edge communication module and edge-device management module.The edge network’s autonomous management of edge nodes is achieved through the edge controller module in case of network disconnection.Monitoring of the CB-EIP platform is implemented based on the microservice orchestration module and microservice monitoring module.The traffic scheduling function based on user priority is implemented based on the microservice traffic scheduling module.(2)In order to realize the microservice deployment of face recognition application on CB-EIP,face recognition is splitted into five microservice components,including front-end display,face detection,face correction,face characterization and database comparison.By monitoring and analyzing infrastructure resources and microservice components,the feasibility of CB-EIP supporting microservice based face recognition application is verified,and the effectiveness is presented in improving usage of computing resource.(3)A microservice-oriented communication delay optimization scheduling algorithm(CDOS)is proposed for microservice scheduling in edge scenarios,and the effectiveness of the algorithm in optimizing communication delay is verified.