One-way delay measurement based on flow data: Quantification and compensation of errors by exporter profiling
One-way delay measurement based on flow data: Quantification and compensation of errors by exporter profiling
复制标题
基于流量数据的单向延迟测量：通过出口商分析对误差进行量化和补偿
DOI：
10.1109/icoin.2011.5723108
复制
发表时间：
2011
期刊：
影响因子：
--
通讯作者：
Jochen Kögel
中科院分区：
文献类型：
--
作者：
Jochen Kögel
单向时延（OWD）是网络管理的重要手段。它可以指示路由问题，网络拥塞，并有助于跟踪网络影响的应用程序问题。虽然主动测量提供高精度，但在大型网络的所有路径上主动测量OWD会导致高工作量。因此，被动测量具有吸引力。我们测量OWD的方法是基于流数据，这些数据通常是从支持流捕获的路由器（导出器）导出的，用于流量核算和报告。因此，这种方法不需要额外的网络组件，并且通常几乎没有额外的成本。众所周知，由于记录丢失，流量数据往往是不准确和不完整的。然而，关于流量数据的时间戳准确性的信息很少。因此，我们研究了时间戳误差，并开发了一种方法来量化它们，以提高基于流量捕获的OWD测量的准确性。我们的贡献是三方面的：首先，我们分析了时钟分辨率的影响流记录创建的参考模型的基础上。第二，我们开发出口商分析方法，即，从流数据中提取由导出器引入的错误。第三，我们提出了从全球企业网络收集的数据中获得的结果。我们的结论是，流量数据为基础的OWD计算的精度在很大程度上取决于捕获设备，特别是时间戳分辨率。观察到的标准偏差范围为2.07 ms至34.02 ms。
One-way delay (OWD) is an important mea-surand for network management. It can indicate routing problems, network congestion, and is useful for tracking down application problems to network effects. While active measurements deliver high accuracy, measuring OWD actively on all paths of large networks results in high effort. Thus, passive measurements are attractive. Our approach for measuring OWD is based on flow data, which is often exported from flow capturing enabled routers (exporters) for traffic accounting and reporting. Thus, this approach does not require additional network components and often comes at almost no additional cost. It is well-known that flow data is often inaccurate and incomplete due to record loss. However, only little information is available on timestamp accuracy of flow data. Therefore, we investigated timestamp errors and developed a method for quantifying them in order to improve the accuracy of flow capturing based OWD measurements. Our contribution is threefold: First, we analyze clock resolution effects based on a reference model for flow record creation. Second, we develop methods for exporter profiling, i.e., extracting errors introduced by exporters from flow data. Third, we present results obtained from data collected in a global enterprise network. We conclude that precision of flow data based OWD calculation heavily depends on the capturing devices, especially on timestamp resolution. Standard deviations observed range from 2.07 ms to 34.02 ms.