软件定义的网络故障诊断方法研究

Modern computer networks, especially data centers, are very prone to faults, due to their large scale and complicated protocol stack. Traditional network management tools are over-simplified, and highly-demanding on operators’ expertise. As a result, it takes a relatively long time to debug the fault, and thereby incurring relatively large economic cost. This project studies how to automatically debug network faults, by leveraging the Software Defined Network (SDN). The goal is to systematically solve the fault detection and localization problem due to switch misconfiguration, malicious modification, software bugs, and hardware failures. The key feature of this project is that it combines passive monitoring and active probing, simultaneously uses online verification and offline analysis, and bases on real-time traffic on the data plane, with an aim to automatically detect, localize, and repair network faults, thereby shortening the debugging period. This project attempts to contribute in three aspects: (1) exploiting the global view of SDN controller to verify the correctness of traffic forwarding paths; (2) employing OpenFlow to obtain traffic statistics and construct forwarding behavior matrix, in order to detect network-wide anomaly traffic, (3) leveraging Binary Decision Diagram to fast test the correctness of switch flow tables, and localize and repair the faulty flow entries. This project is expected to offer a relatively complete system for automatic network debugging, thereby offering a strong support for intelligent network management.

现代计算机网络特别是数据中心网络由于规模庞大、协议复杂，在运行中极易发生故障。由于传统网络管理工具过于简单、高度依赖管理员个人经验，故障诊断周期相对较长，造成的经济损失较大。本项目拟基于软件定义网络（SDN）技术，研究自动化的网络故障诊断方法，系统解决交换机配置错误和篡改、交换机软件漏洞、硬件失效等故障的检测与定位问题。本项目特色在于综合利用被动监测和主动探测方式、并结合在线验证和离线分析手段，面向数据平面实时流量，自动化发现、定位并修复网络故障，缩减故障诊断周期。项目力图从三方面寻求突破：（1）利用控制器全局控制视图检测流量转发路径的正确性，（2）利用OpenFlow获取流量统计特征并建立转发行为矩阵，在全网范围内检测异常流量，（3）基于二元决策图快速探测交换机流表的正确性，并定位和修复故障表项。项目预期可形成一套较为完整的自动化网络故障诊断系统，为智能网络运维提供有力支撑。

现代计算机网络特别是数据中心网络由于规模庞大、协议复杂，在运行中极易发生故障。由于传统网络管理工具过于简单、高度依赖管理员个人经验，故障诊断周期相对较长，造成的经济损失较大。本项目拟基于软件定义网络（SDN）技术，研究自动化的网络故障诊断方法，系统解决交换机配置错误和篡改、交换机软件漏洞、硬件失效等故障的检测与定位问题。本项目综合利用被动监测和主动探测方式、并结合在线验证和离线分析手段，面向数据平面实时流量，自动化发现、定位并修复网络故障，缩减故障诊断周期，取得了三项研究进展：（1）设计了流量转发路径监测方法，以控制平面策略为正确性基准，对数据平面的实时流量进行采样，可以高效验证其转发路径是否与控制平面策略一致。（2）设计了异常流量检测方法，利用转发行为矩阵刻画正确转发行为特征，可以有效检测网络中的异常流量。（3）设计了探测包快速生成和增量更新方法，可以对动态网络进行实时探测。以上成果形成了一套较为完整的自动化网络故障诊断系统，为智能网络运维提供有力支撑。