网络功能虚拟化（Network Function Virtualization，NFV）是支撑下一代运营商网络和数据中心网络的关键技术。为满足未来网络对高性能和高可用性的需求，实现低延时高容错的NFV系统已成为亟待解决的关键性挑战问题。本项目采用创新性方法解决现有NFV系统中存在的报文处理延时高和容错方案开销大的问题，为其在未来网络中的部署和应用提供重要的理论基础和方法支持。主要研究内容包括：（1）提出结合可编程数据平面的、面向服务质量需求的高性能NFV服务链机制，并提出面向NFV软/硬件服务链的串并行编排机制，以降低NFV系统中报文处理的高延时；（2）提出结合可编程数据平面的高效NFV状态管理机制，以降低现有NFV系统容错方案造成的网络流量管理、状态迁移和报文处理的高开销。本项目将完成低延时高容错NFV系统的关键机制研究和原型实现，对推动未来高性能、高可用网络体系结构的发展具有重大意义。

Network Function Virtualization (NFV) has become a key technique of the next-generation ISP networks and data center networks. To meet the demands of high performance and high availability in future networks, a key challenge is to provide a low-latency, low-overhead fault-tolerant NFV system. In this project, we take innovative approaches to solve the issues of high latency of packet processing, and the large overheads of providing fault tolerance in existing NFV systems, both setting the theoretical foundations and providing practical mechanisms for the deployment and application of NFV in the high-performance, highly available future networks. In particular, this research includes the following approaches: (1) To reduce the latency in NFV systems, we propose a QoS-oriented, high-performance NFV service chaining mechanism combining programmable data plane, and also propose the mechanisms of serial/parallel orchestrations of both software and hardware service chains; (2) To reduce the overheads of network traffic management, state migration and packet processing in fault-tolerant NFV systems, we take advantage of the programmable data plane and in-network storage, and propose a high-performance state management service. This project will deliver both key mechanisms and prototype implementations of a low-latency, low-overhead fault-tolerant NFV system, which is of great importance in pushing forward the developement of future network architectures.