面向互联网的SDN流量多粒度处理机制

Traffic multi-granularity processing mechanism for Internet-oriented SDN

软件定义网络(software defined networking,SDN)作为一种新型的网络架构,将网络的控制平面与数据转发平面分离,实现了可编程化控制,为互联网提供了改善网络全局性能的新思路.虽然SDN具有全局视角优势,但在处理互联网海量数据时也存在性能瓶颈:频繁的层间通信会使控制器计算效率下降,海量的流表项数据使得交换机存储压力过大.为了进一步提升SDN的性能使其适应互联网的海量流量处理,本文提出了面向互联网的SDN流量多粒度处理机制,将SDN架构应用到互联网骨干网的流量处理中,分别从路由和调度两个方面设计并实现了流量多粒度处理机制.仿真结果表明:本文设计的流量多粒度处理机制能减少层间通信次数,减少下发流表项,维持负载均衡,提高了路由选取的正确性和有效性,提升了SDN性能,进而提升了处理互联网海量数据的能力.

As a new network architecture, software defined networking(SDN) separates the control and data forwarding planes and implements programmable control. SDN provides a new way to improve the overall performance of the Internet. Although SDN has global view capacity, it also suffers from performance bottlenecks when handling massive Internet data, i.e., frequent interlayer communication between the control and data forwarding planes of SDN reduces the computational efficiency of the controller, and massive flow table entry data incur very high storage costs on switches. To further improve SDN performance and make it adaptable to the massive traffic processing of the Internet, this paper proposes a traffic multi-granularity processing(MGP) mechanism for Internet-oriented SDN. The proposed MGP applies the SDN architecture to traffic processing of the Internet backbone network and implements a multi-granularity traffic processing mechanism in consideration of routing and scheduling. Simulation results demonstrate that the proposed MGP can reduce the amount of interlayer communication and flow table entries, maintain network load balancing, improve the correctness and effectiveness of route selection, improve SDN performance, and improve the processing capacity of mass data on the Internet.