| As a complement to cloud computing,edge computing,for some scenarios(e.g.,Telematics),can effectively share the data processing pressure of cloud data centers and provide low-latency computing services to end users.Containers,as a flexible and efficient way to deploy applications,already have widespread use in the cloud computing space.In recent years,there has been some research and practice in industry and academia on how to apply container technology to edge computing.One of the representatives is Huawei’s open source KubeEdge,which extends Kubernetes containerized application orchestration capabilities to the edge,making Kubernetes better suited for the deployment and management of containerized applications in edge computing scenarios.However,the container orchestration system based on Kubernetes and KubeEdge still has some issues in container scheduling,container startup and container communication that need to be improved.In this thesis,the above points will be researched,analyzed,and optimized,and the main work is as follows:(1)To address the problem that the default scoring algorithm in edge-side container scheduling is simple and does not consider storage and network factors,this thesis designs an algorithm that dynamically adjusts the weight of scores based on the variance of historical data,and dynamically connects to the original system in the form of a plug-in and can manually control the factors involved in scoring.(2)To address the problem of slow container start-up speed at the edge,this thesis designs a container runtime agent module based on image data caching,which enables the container engine to efficiently use the existing container image data of neighboring nodes when starting containers and improve the container start-up speed.(3)To address the problem that container applications located in different network segments in edge cloud collaboration cannot communicate with each other at the application layer,this thesis designs a container communication scheme across network segments based on the existing network communication interface of Kubernetes and implements the corresponding network communication module by coding.(4)The three modules above are integrated into a complete prototype system,providing a visual management interface for each module.In summary,this thesis provides solutions to the problems in container scheduling,container startup and container communication when Kubernetes and KubeEdge are applied to edge computing scenarios and implements the corresponding modules in the form of plug-ins to the original system through coding.The above work has improved the accuracy of the scheduler’s scoring of nodes,reduced the latency of container starts,enabled containers to communicate across network segments,and the visual interface provided by the prototype system simplifies the management of the modules. |
