| In recent years,as the Internet continues to expand,many services with different requirements have emerged.Different types of Internet applications have put forward new requirements for end-to-end delay.They need to be able to effectively manage large-scale networks,distinguish different types of network applications and provide stable and high-quality services.In view of these challenges of the future network,Some communication-related researchers from China have put forward the concept of network 5.0,adhering to the idea of "network-centric",aiming to provide strong endogenous management capabilities for the network.In addition,the Internet Engineering Task Force proposed the concept of deterministic network,aiming to provide deterministic services with stable end-to-end delay and extremely low packet loss rate for applications in the network,and carried out research on this.How to manage the network in the future large-scale network scenario and plan a stable transmission path for it according to its characteristics has become the focus of this paper.According to the reference model of Network 5.0,an end-to-end deterministic routing algorithm for large-scale networks is designed.The complete algorithm consists of two parts.The first part is to partition the network topology.A topology partition algorithm based on graph attention mechanism is proposed,which extracts the corresponding characteristics of each node and link in the network,divides the complete topology into several sub-domains,and evaluates the partition results using the modular degree.The results show that the algorithm has good partition effect,and its performance is improved compared with other neural network models.The second part is to make end-to-end routing decisions for traffic flows.The layered reinforcement learning model is used for modeling.The complete routing solution includes two stages:path between subgraphs and path within subgraphs.The purpose of the algorithm is to provide end-toend paths that meet its constraints for traffic flows with deterministic QoS requirements.Subsequently,experiments and comparative analysis are carried out based on the simulation environment.The experimental results show that has better performance in terms of deterministic flow satisfaction compared to traditional routing algorithms,and better performance in terms of throughput and resource utilization.This paper implements a network subsystem based on the architecture of Network 5.0.It uses the idea of network virtualization and software to define the network to implement each network element device in the network,and then tests the functions of the subsystem,including the realization of specific message forwarding function,the calculation and configuration of routes,resource allocation and deterministic ability test. |
