Although software-defined networking (SDN) provides a flexible way to provision and control networks, it also makes network debugging and troubleshooting more complex. In SDN, the network is fully managed by software programs that increase flexibility and sophistication but are prone to bugs. Pinpointing those bugs is challenging because they can occur at multiple locations, such as the forwarding plane, the controller OS, and the network services running on top of the controller. Compared to functional bugs, performance bugs are particularly irritating due to their non-failure semantics. They can lead to performance loss (reduced throughput, increased latency, and wasted resources) while maintaining the network connectivity. The literature reports several tools and techniques to diagnose SDN bugs, but, unfortunately, they are mostly ineffective against performance bugs. In this paper, we propose a novel monitoring and diagnostic framework capable of diagnosing performance bugs in the SDN data plane. The proposed tool works within Open vSwitch (OVS), a popular software switch, albeit it can easily be adapted to any OpenFlow switch. Tracing techniques are used to collect low-level performance data from monitored switches. Our tool derives adapted performance metrics from kernel and userspace traces, and then displays them in time-synchronized graphical views. These views provide valuable insights into OVS operation. They also enable practitioners to discover performance-related issues and analyze their root causes. A few use cases are presented to demonstrate the efficiency of our tool in optimizing OVS performance and diagnosing its performance bugs.

中文翻译：

---

使用内核和用户空间跟踪对启用 OpenFlow 的交换机进行诊断和故障排除

尽管软件定义网络 (SDN) 提供了一种灵活的方式来配置和控制网络，但它也使网络调试和故障排除更加复杂。在 SDN 中，网络完全由软件程序管理，这些软件程序增加了灵活性和复杂性，但容易出现错误。查明这些错误具有挑战性，因为它们可能发生在多个位置，例如转发平面、控制器操作系统和运行在控制器之上的网络服务。与功能性错误相比，性能错误因其非故障语义而特别令人恼火。它们可能导致性能损失（吞吐量降低、延迟增加和资源浪费），同时保持网络连接。文献报道了几种诊断 SDN 错误的工具和技术，但不幸的是，它们大多对性能错误无效。在本文中，我们提出了一种新颖的监控和诊断框架，能够诊断 SDN 数据平面中的性能错误。建议的工具在 Open vSwitch (OVS) 中工作，这是一种流行的软件交换机，尽管它可以轻松适应任何 OpenFlow 交换机。跟踪技术用于从受监控的交换机收集低级性能数据。我们的工具从内核和用户空间跟踪中得出经过调整的性能指标，然后在时间同步的图形视图中显示它们。这些视图提供了对 OVS 操作的宝贵见解。它们还使从业者能够发现与性能相关的问题并分析其根本原因。提供了一些用例来证明我们的工具在优化 OVS 性能和诊断其性能错误方面的效率。