Research On Intent-driven Cross-layer Coordination Restructuring Of IP+Optical Network

With the official commercialization of 5th generation mobile communication technology(5G),beyond 5g(B5G)and 6th generation mobile networks(6G)have also attracted extensive research in academia and industry.New applications and business requirements in 6G networks(such as ultra-high definition video streaming,virtual and augmented reality,telemedicine,etc.)have come,it will bring more dynamic and peak to average traffic to the network.The traditional static IP topology based on peak traffic design,even if using load balancing technology,faces the problems of low resource utilization efficiency and high network cost.At the same time,due to the hierarchical management and independent operation and maintenance of traditional IP+optical network,the flexibility of the network is insufficient.If the current IP+optical network architecture is used to support emerging services,it will not only be difficult to ensure the service quality of the network,but also lead to problems such as low resource utilization and high network capital expenditure,which is a huge problem and challenge for the communication network.Therefore,the network needs real-time and dynamic adjustment to meet the high dynamic traffic demand.However,the joint routing and resource allocation between IP network and optical network involves twotier parameters.The whole process is too complex to be applied and deployed in the actual large-scale network scenario.In addition,IP networks usually regard optical networks as static networks.The above limitations make IP network unable to make full use of the flexibility of optical network to adjust in real time to respond to dynamic business requirements.Therefore,how to effectively cooperate with IP+optical network to schedule network resources across layers,improve resource utilization efficiency in the network,reduce operation and maintenance costs,ensure network service quality and save costs as much as possible are the key difficulties faced by operators.Therefore,in view of the above problems,this paper introduces the concept of "intention driven" IP+optical network collaboration,carries out active and reactive topology reconstruction of the network under normal and fault conditions to improve the resource utilization and save costs of the network,and forms a complete closed loop of cross layer collaborative reconstruction,self-optimization and self-recovery of IP+optical network based on intention driven.Specifically,the main research contents of this paper are as follows:1.This paper proposes a topology recommendation mechanism based on intention driven cross layer cooperation in IP+optical networks.Aiming at the problems of low network resource utilization and high network cost caused by the mismatch between static network resource allocation and dynamic service requirements in the existing network,the network topology with the best performance(including link capacity)is recommended to the IP layer through the optical layer,so that the IP layer network can be reconstructed in real time according to the dynamic service requirements.This mechanism recommends the optimal network topology upward through the optical layer,and shows the carrying capacity of the optical layer to the IP layer.Therefore,this mechanism makes full use of the flexibility of the optical network and makes up for the deficiency of the network solid-state rigidity caused by the fact that the IP network can not fully know the use of optical network resources when the underlying physical optical network carries the IP network of multiple tenants.Therefore,the IP network topology can be reconstructed in real time according to the dynamic business requirements,which greatly improves the resource utilization of the network.In addition,in order to verify the feasibility of this mechanism,this paper designs an experimental platform for cross layer collaborative reconfiguration of IP+optical network for the research and verification of subsequent algorithms.2.This paper presents a proactive network topology restructuring(PNTR)algorithm based on deep neural networks.Based on the above cross layer collaborative topology recommendation mechanism of IP+optical network,the IP network topology with the best performance(including link capacity)is calculated through the proactive network topology reconstruction recommendation algorithm,so as to dynamically reconstruct the IP network to provide customized services for the upper intention and realize the self-optimization of the network.The algorithm takes the lowest cost of the network as the optimization goal,calculates the optimal network topology suitable for the current business intention in real time according to the business requirements of the network,so that the network can ensure the user’s quality of service with low cost and efficient utilization of network resources,and the actual effect of the algorithm is tested by simulation program.The simulation results show that the algorithm greatly improves the utilization of network resources,and has better real-time and generalization ability than heuristic algorithm.3.In this paper,a reactive network topology reconstruction(ATLRNTR)algorithm based on adaptive transfer learning is proposed.When the core IP node fails and there are no idle dormant devices in the network for service migration,the traditional rerouting based method has poor recovery effect due to its long recovery time and dynamic service distribution.At the same time,the pre-trained algorithm model and the set network strategy will no longer be suitable for the current network failure,and the retraining of the model needs massive data sets and certain computing resources.Based on this,the reactive network topology reconstruction algorithm based on adaptive migration learning is proposed to reconstruct the network topology in real time to recover the network fault,realize the self-recovery of the network and improve the reliability of the network.The simulation results show that the algorithm has obvious advantages in the required samples and iteration times compared with the retraining of the model when it reaches the benchmark solution.