基于路径及反馈的软件定义网络策略更新方案*

软件定义网络(Software-Defined Networking,SDN)虽是具有逻辑集中控制特点的网络架构,但其底层的数据平面依旧是分布式的系统,由此产生的策略更新不一致问题将导致数据包的错误处理,进而引起一系列不正确的网络行为。就此问题,本文提出了基于路径及反馈的策略更新方案,通过对旧流行踪的准确定位实现新、旧流的并行传输,并基于BP神经网络建立了相应的网络性能反馈模型,根据网络状态动态调节数据包重放时机,旨在确保一致性的同时,获得更加良好的网络性能。MATLAB训练结果以及基于POX、Mininet的仿真实验表明,反馈模型的建立能够准确刻画网络性能的变化,该方案与同类研究相比具有更优良的网络性能、更新效率、通用性、隔离性以及相当低的规则空间消耗。

Path and Feedback Based Policy Update Scheme for SDN

Although Software-Defined Networking has the network architecture of logically centralized control, its underlying network still has the characteristics of distributed systems, and the resulting problem of policy update inconsistency will lead to an error processing of data packets, thereby causing a series of incorrect network behaviors. At this point, the proposed path and feedback based update scheme accurately tracked the location of old flows to achieve the parallel transmission of old and new flows, and the established network performance feedback model which based on BP neural network can dynamically adjust the packets re-released timing according to the network status. This scheme is aimed to obtain better network performance as well as the consistency guarantee. Based on POX, Mininet and MATLAB, the simulation and training results show that the feedback model can accurately depict changes in network performance, and compared with similar studies, our scheme has better network performance, efficiency, commonality, isolation and considerably lower rule-space consumption.