Research On Privacy-Enhanced Routing And Spectrum Allocation Mechanism In Multi-Domain Elastic Optical Networks

The development of Internet technology and the emergence of emerging industries have put forward higher requirements for optical network capacity.In order to improve network capacity,elastic optical network technology with flexible spectrum allocation capability emerges as the times require.At the same time,in order to facilitate network management and control,low-redundancy and lightweight multi-domain elastic optical networks have become an inevitable trend in network development.Due to the huge scale of multi-domain elastic optical networks,machine learning technology needs to be introduced to handle massive connections and provide adaptive services.However,the huge network scale and the high requirement of intra-domain information privacy make it difficult to collect network data,which makes it difficult to train machine learning models based on data centralization.The traditional method adopts a method of abstracting the intra-domain topology first and then centrally processing it to generate cross-domain service provisioning strategies.However,this approach requires a trade-off between the accuracy of the generation strategy and the privacy of intra-domain information.In addition,this method not only cannot dynamically adjust the strategy according to the state of the network,but also leads to the transmission of a large amount of service-irrelevant data in the network,causing network burden.In view of the above problems faced by the current cross-domain service provisioning of multi-domain elastic optical networks,this thesis focuses on the privacy-enhanced routing and spectrum allocation in multi-domain elastic optical networks.The main research contents and results are as follows:(1)This thesis proposes a privacy-enhanced cross-domain service provisioning framework for multi-domain elastic optical networks.In view of the huge scale and the high requirement of information privacy in the domain,which makes the adaptive service provisioning based on machine learning difficult,this thesis proposes a privacy-enhanced cross-domain service framework.The framework adopts a distributed execution and centralized strategy generation method.It does not need to centralize the intra-domain information.It only needs to exchange model parameter information between the domain manager and the federation coordinator.Cross-domain service strategies can be generated.In theory,this framework not only ensures the privacy of intra-domain information,but also reduces the transmission of service-irrelevant information in the network,and can also perform adaptive services provisioning according to the network state.Furthermore,this thesis designs a privacy-enhanced cross-domain service platform for multi-domain elastic optical networks to verify the feasibility of the proposed framework.(2)This thesis proposes a dynamic single-service routing and spectrum allocation mechanism in a multi-domain elastic optical network.Aiming at the difficulty of routing and spectrum allocation in a multi-domain elastic optical network under the condition of ensuring the privacy of intra-domain information,this thesis proposes a routing and spectrum allocation algorithm based on multi-agent federated reinforcement learning.At the same time,this thesis introduce a multi-agent bidirectional activation and multi-auxiliary graph parallel computing mechanisms to accelerate the algorithm.The proposed algorithm only needs to exchange Q values between the federation coordinator and the domain manager to generate routing and spectrum allocation strategies.The platform and simulation results show that:under the condition of ensuring the privacy of intradomain information,the proposed algorithm can adaptively adjust the strategy according to the network state;the algorithm has good performance in running time,decision success rate and decision accuracy.(3)This thesis proposes a concurrent multi-service routing and spectrum allocation mechanism in multi-domain elastic optical network.Aiming at the problem of decision conflict between concurrent service routing and spectrum allocation under the condition of ensuring intra-domain information privacy,this thesis proposes a routing and spectrum allocation algorithm based on FQMIX.The algorithm does not need to pre-order concurrent services.It can realize collaborative decision-making of multiple services through centralized learning and distributed execution.The algorithm adopts the Mixing network to effectively evaluate the advantages and disadvantages of concurrent service joint decision-making strategies.The platform and simulation results show that the proposed FQMIX algorithm performs well in routing decision success rate and conflict rate under the condition of ensuring the privacy of intra-domain information.