跨域场景下的联邦路由机制设计  

Design of federated routing mechanism in cross-domain scenario

作　　者：丛培壮 张宇超 田野[1] 王文东[1] 李丹[3] CONG Peizhuang;ZHANG Yuchao;TIAN Ye;WANG Wendong;LI Dan(Beijing University of Posts and Telecommunications,Beijing 100876,China;Tsinghua University,Beijing 100084,China)

机构地区：[1]北京邮电大学网络与交换技术国家重点实验室,北京100876 [2]北京邮电大学计算机学院(国家示范性软件学院),北京100876 [3]清华大学,北京100084

出　　处：《电信科学》2020年第10期29-36,共8页Telecommunications Science

基　　金：国家自然科学基金资助项目(No.61802024);中央高校基本科研业务费专项资金资助项目(No.2482020RC36);国家重点研发计划项目(No.2019YFB1802603)。

摘　　要：随着多网融合的发展,独立自治域网络之间的高效互联互通已成为一个关键理论技术问题。由于域内信息的私密性,各自治域之间不会共享相关的数据,传统的域间路由协议无法提供高效的跨域路由机制。为克服现有协议存在的不足,提出了一种基于机器学习的联邦路由机制,该机制通过各自治域的神经网络模型和参数隐式地共享域内信息,不仅可打破现存的数据孤岛问题,还可极大地减少域间信息共享所需要传输的数据量,进而降低全网路由信息收敛时延,基于该联邦路由机制,边界路由器也能够根据全网状态信息制定全局的优化路由转发策略。With the development of multi-network integration,how to ensure efficient interconnections among multiple independent network domains is becoming a key problem.Traditional interdomain routing protocol fails due to the limitation of domain information privacy,where each autonomous domain doesn’t share any specific intra-domain information.A machine learning-based federated routing mechanism was proposed to overcome the existing shortcomings.Each autonomous domain shares intra-domain information implicitly via neural network models and parameters.It not only breaks data islands problems but also greatly reduces the amount of transmitted data shared between domains,then decreases convergence delay of entire network information.Based on the federated routing mechanism,border routers can formulate global optimal routing strategies according to the status of entire network.

关 键 词：跨域网络 联邦路由 网络融合 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]