Research On Architecture And Resource Allocation Algorithm For 5G End-to-end Network Slicing

Network Slicing(NS)emerged to meet the challenge of the interconnection of all things.Its basic idea is to divide a physical network into several logical networks that meet different requirements,which are isolated from each other.However,the network management objects increase with the diversified development of the vertical industry applications,which causes the change operations to become more frequent,and there are more operational breakpoints between planning,construction,maintenance,optimization and operation links.Therefore,it is urgent to explore the Management and Orchestration(MANO)technology for the 5G end-to-end(E2E)network slicing.In addition,the deployment of the E2E network slices involves resource allocation across domains,and Artificial Intelligence(AI)can be introduced to improve the performance of E2E network slicing with limited system resources.To solve the above problems,the 5G E2E network slicing architecture and resource allocation algorithm are studied in this work.First,a full-service E2E network slicing architecture is proposed through four-tier and three-sided abstraction,which is customized by decoupling and reconstructing the network functions,as well as templates and instantiations of E2E network slicing.Based on this,the joint optimization of E2E network slicing is explored to enhance the communication quality of E2E network slicing.The main contributions of this work are as follows:(1)A full-service E2E network slicing architecture is designed.First,the architecture includes four layers,i.e.,the application layer,the function layer,the virtualization layer and the infrastructure layer,and three planes,i.e.,the control plane,the AI plane and the MANO plane.Then,the AI plane and MANO plane of the 5G core network is designed as microservices:the integrated network elements are decoupled into multiple microservice AI functions and MANO functions,which are interconnected through a unified service interface.Secondly,the idea of template and instantiation is introduced to support customized E2E network slicing:1)A dedicated network slicing template is designed for typical vertical industries;2)The network slice template is activated on demand,that is,instantiation.Finally,the experimental results show that:1)The 5G E2E network slicing designed in our work can meet the personalized requirements of the typical vertical industries and provides customized services for users;2)The isolation between slices can be effectively guaranteed;3)The service response time can be effectively reduced;4)The reliability of the system can be improved.(2)Deep Deterministic Policy Gradient(DDPG)algorithm is used to solve the joint optimization of the E2E network slicing First,an E2E network slicing system model is built and two typical types of network slices are defined:high-throughput and delay-sensitive slices.Then,the joint optimization of the E2E network slicing utility is modeled.To better describe the performance metrics of the E2E network slice heterogeneity,the utility functions of two slices in terms of the throughput and latency gains minus the corresponding costs are defined,respectively.To jointly optimize the allocation of communication and computing resources of the two types of E2E network slices,the long-term total utility of the two types of slices is maximized while satisfying the service requirements of all users.Finally,the feasibility of the proposed algorithm is verified by experiments.With limited system resources,the algorithm can increase the rate of high-throughput slices and reduce the latency of delay-sensitive slices effectively.