Modern network measurement employs several in the substrate network. These points perform measurement tasks to measure traffic and report real-time to in the control plane. These servers convert events to flow statistics and report them to network management applications, which require both (i.e., collecting events within a limited time deadline) and (i.e., bounding the probability of event loss). However, existing solutions fail to satisfy the two requirements because they ignore the transmission latency and reliability when collecting events from measurement points to monitoring servers. In this paper, we propose , a framework that aims to offer low-latency and reliable event collection. provides a near-optimal heuristic that guides each measurement point to select monitoring servers to receive its events and corresponding network paths for transferring events with timeliness and reliability guarantee. We have implemented on a 64 × 100Gbps Tofino-based switch. We conduct testbed experiments to demonstrate the effectiveness of in practice. The results indicate that offers low-latency and reliable event collection while preserving high application-level accuracy.

中文翻译：

---

【公式省略】：网络测量中低延迟、可靠的事件采集


现代网络测量在底层网络中采用了多种测量方法。这些点执行测量任务来测量流量并向控制平面实时报告。这些服务器将事件转换为流量统计数据并将其报告给网络管理应用程序，这需要（即在有限的期限内收集事件）和（即限制事件丢失的概率）。然而，现有的解决方案无法满足这两个要求，因为它们忽略了从测量点收集事件到监控服务器时的传输延迟和可靠性。在本文中，我们提出了一个旨在提供低延迟且可靠的事件收集的框架。提供近乎最优的启发式方法，指导每个测量点选择接收其事件的监控服务器以及相应的网络路径来传输事件，并保证及时性和可靠性。我们已在基于 Tofino 的 64 × 100Gbps 交换机上实施。我们进行测试台实验以证明其在实践中的有效性。结果表明，它提供了低延迟和可靠的事件收集，同时保持了应用程序级的高准确性。