Intelligent Collaborative Management And Orchestration Of Cloud-Network-Edge-End Based On Microservices

Cloud Computing(CC)has been widely used with abundant computing and storage resources.However,due to the centralized architecture of CC,it cannot provide low-latency services and increase the network load.Edge Computing(EC)provides low-latency services and reduces network load pressure,but it is limited by its resources and cannot fully meet increasingly stringent requirements.In the era of Internet of Things(Io T),huge number of devices have many resources that are not utilized.Therefore,it is necessary to study cloudnetwork-edge-end multi-domain collaboration technology.In addition,microservice has been widely used recently because of agility and scalability.However,while microservice bring advantages,it also complicates the application architecture,which makes service management and orchestration in cloud-network-edge-end systems more difficult.Thence,the cloud-network-edge-end intelligent collaborative management and orchestration technology based on microservices is studied in this paper: firstly,a distributed collaborative architecture of “four-layers,four-domains and three-plane” is proposed to realize the management and orchestration of network services rapidly and efficiently by designing the virtual network function of the distributed control layer;On this basis,machine learning technology is used to solve the problem of microservices deployment to further reduce service response time.The research content of this paper is as follows:(1)A microservice-based cloud-network-edge-end distributed collaborative management and orchestration architecture is proposed in this paper.Firstly,a distributed architecture which can achieved the efficient allocation of network resources,on-demand service customization,and flexible and reliable management is proposed by incorporating the cloudnetwork-edge-end into the “four-layer,four-domain and three-plane” architecture.“Fourlayer” include infrastructure layer,virtualization layer,function layer and application layer.“Four-domain” include end domain,edge domain,cloud domain and network domain.The cloud domain,edge domain and end domain provide distributed computing and storage resources for the architecture,and the network domain provides communication capabilities for cloud-edge-end information interaction and collaboration.Each domain includes “threeplane”,control plane,management orchestration planes and intelligent plane.Secondly,in order to realize distributed collaborative management and orchestration,virtual network functions combined with microservices,such as end device management function,edge service deployment function,remote service deployment function and edge service registration function are designed.All network functions communicate with each other through predefined interfaces to completes the provision and acquisition of services.Finally,a cloud-network-edge-end distributed collaborative management and orchestration platform based on microservices is built,and compared with existing technologies such as traditional cloud computing and centralized cloud-edge collaborative computing architecture,it is verified that the proposed architecture can effectively reduce the service instantiation time,service response time and end-to-end transmission delay,etc.(2)A microservice deployment algorithm based on Graph Neural Network(GNN)and Deep Reinforcement Learning(DRL)is studied in this paper,which can choose the deployment location for microservices.Firstly,the microservices in the cloud-network-edge-end system are constructed as a Directed Acyclic Graph(DAG),in which the vertices of the graph represent microservices,and the edges of the graph represent the dependencies between microservices.Secondly,by minimizing the service response time and satisfying the constraints that each microservice can only be deployed on one physical node,load balancing and available resource thresholds,the optimal microservice deployment scheme is obtained.Then,since the traditional Deep Learning(DL)cannot process graph data,this paper adopts the GAT-DDPG algorithm based on Graph Attention Networks(GAT)and Deep Deterministic Policy Gradient(DDPG)to solve this problem,where GAT is used to extract the features of complex DAG and the features is a part of the state set of the DDPG algorithm.The simulation results show that the microservice deployment algorithm proposed in this thesis can effectively reduce the service response time compared with random deployment algorithm and nearby deployment algorithm.