SDN/NFV架构下服务功能链的在线智能部署机制研究

软件定义网络和网络功能虚拟化是下一代互联网的关键技术。基于SDN/NFV架构，服务功能链要求网络流按序经过一系列特定的网络功能，通过灵活利用基础设施的异质资源为用户提供定制化的网络服务。如何针对服务请求的多样性，网络资源的异质性和时变性等特点，对服务功能链进行在线智能部署是目前亟待解决的问题。本项目拟围绕异质资源动态特征与关系的表示方法、面向服务功能链的在线动态资源分配机制等科学问题展开研究，研究内容包括：①刻画服务功能链的差异化需求与多域异质资源之间的内在关系，提出供需适配的服务功能链分配方法；②研究异质资源抽象和深度特征表示方法，提出基于服务质量保障的服务功能链的在线自适应映射策略；③研究融合生成对抗网络与强化学习的决策优化机制，提出快速智能的服务功能链重配置算法。本项目可以为满足用户多元化服务需求，降低网络运营成本，提升服务部署灵活性提供重要的理论依据和关键的技术支撑。

Software-defined networking (SDN) and network function virtualization (NFV) are key technologies for the next generation Internet. Based on the architecture of SDN/NFV, the service function chain (SFC) requires the network flow to go through a series of network functions in a specific order. It could provide users with customized network services by flexibly using the heterogeneous resources of the infrastructures. How to implement the online intelligent deployment of SFC in response to the diversity of service requests, the heterogeneity and the time-varying characteristics of network resources needs to be solved urgently. This project takes the representation method of the dynamic characteristics and relationships of heterogeneous resources and online dynamic resource allocation mechanism for service function chain as the key scientific issues. The research methodologies of this project are as follows: 1) This project studies the relationship between the differentiated requirement of SFC and the heterogeneous resources of multiple domains, and proposes a demand-supply matched allocation method of SFC; 2) This projects studies heterogeneous resource abstraction and deep feature representation methods, and propose an online adaptive mapping strategy for SFC with the quality-of-service guarantee; 3) This project studies the decision-making optimization mechanism of the integration of generative adversarial network and reinforcement learning, and proposed fast and intelligent SFC reconfiguration method. The research on this project will provide theoretical and technical support for satisfying the diversified service needs of users, reducing the network operating costs, and improving the flexibility of service deployment.