Research On Key Technologies Of Optical Network Resource Scheduling Towards Computing And Networking Integration

With the rapid development of 5G/6G technology and emerging new applications,differentiated and high-quality requirements are prominent,driving communication networks to provide both information transmission and information processing.Computing and networking integration,which breaks the isolation of computing power and network and greatly improves network capacity and service quality,is becoming an important direction of future networks.As the underlying network,the optical network urgently needs to expand the transmission capacity and enhance the interaction with computing power to cope with the demands of largecapacity carrying and ubiquitous computing power interconnection.In this context,studying the joint resource scheduling of large-capacity optical networks and computing power is imperative.The joint resource scheduling of large-capacity optical networks and computing power faces challenges from "complex coupling of physical layer effects" and "large scale resources coordination":(1)The nonuniform wide-spectrum physical layer effect affects the situation of the lightpath,and makes efficient lightpath configuration difficult;(2)Widespectrum interchannel stimulated Raman scattering(ISRS)causes conflicts among multiple lightpaths in a single service,and makes efficient coordination among multiple lightpaths difficult;(3)The combination of computing power and optical network optimization is a high-dimensional decision problem,and makes efficient mappings of computing and network requirements difficult;(4)The service number largely increased under the flexible trend of the optical network which results in multiplication in resource dimension,and makes parallel management of services difficult.In response to the above challenges,this paper takes "intelligent and efficient computing and optical network resources scheduling" as the goal and carries out research on single lightpath configuration,multiple lightpath coordination,single computing-network integration service scheduling,and multiple computing-network integration services decoupling.The main research contents and contributions are summarized as follows:(1)This thesis proposes an explainable artificial intelligence-based adaptive lightpath margin configuration method in multi-band optical networks.To tackle the complex impact of non-uniform physical layer effects on the lightpath situation in multi-band optical networks,an artificial intelligence(AI)-based lightpath margin estimation model is established.It fits the coupling effect of multiple factors on the generalized signal-to-noise ratio(GSNR),including lightpath information,network status,and traffic mode;Analyzing the AI model through explainable technology bridges the contradiction between the black-box characteristics of AI and the high-reliability requirements of the optical network.Compared with the traditional margin configurations,the proposed method ensures the lightpath transmission quality while avoiding overestimating degradation;the explainable analysis of the AI model assists network trustworthy management and reduces the complexity of AI model training.(2)This thesis proposes a virtual network embedding(VNE)method over multi-band optical networks based on cross-matching and hypergraph theory.Innovation works are conducted from node mapping and link mapping to tackle the coupling between complex physical layer effects and high-dimensional resources.A cross-matching method is proposed for node mapping,which changes independent node ranking to evaluate the mapping tendency.It improves the adaptability of node mapping;A hypergraph is introduced to represent conflicts among multiple lightpaths caused by accumulated degradation of GSNR under the ISRS effect.The hypergraph maximum weight independent set algorithm is proposed to solve the parallel scheduling of lightpaths.Compared with the benchmarks,the proposed method reduces the cost of link mapping by 20.8%and reduces the blocking rate of services by 34.3%.(3)This thesis proposes an edge-cloud collaborative multi-layer VNE method based on approximately lossless model compression.A model compression method is proposed to tackle the challenge of the high complexity of joint optimizing computing and optical network resources.Reserve resources are pre-calculated and link mapping costs are estimated to be reflected in the model as parameters rather than optimization variables,which greatly reduces the decision variables and constraints.Through penalty function conversion and vectorization,solving the compressed model is losslessly transformed into finding the Hopfield neural network(HNN)stable point.Compared with the optimal solution,the proposed method increases the speed of strategy generation by more than 3 times with approximately lossless performance.(4)This thesis proposes an efficient sub-problem decoupling method for large-scale computing-network integration services scheduling under the flexible optical network.To solve the parallel scheduling of multiple services,the sub-problems are decomposed from the service level and network level:A service association graph is constructed,and a service subset partitioning method based on community detection is proposed to ensure high service similarity in each subset;According to the similar characteristics of services in each subset,the resources that are not inclined to be selected are pruned in advance to reduce the underlying network size in the sub-problems.Simulation results show that the subset partitioning method is more efficient than random partitioning.Compared with the optimal solution,the method greatly reduces the solution complexity with a tiny loss in performance.In general,this paper systematically studies the key technologies of optical network resource scheduling in computing and networking integration.The research is conducted to solve single lightpath configuration under multiple factors coupling,multiple lightpath coordination under physical layer accumulated degradation,approximately lossless compression of computing and network optimization models,and multi-objective multiple services decoupling.Simulation results have verified the advantages of the research results in improving the performance of resource scheduling strategies,reducing the complexity of resource scheduling,and ensuring the service quality of computingnetwork integration services.