Abstract:  The characteristics of current data center are that the network scale becomes huge, and the flow behavior becomes complex, studying large-scale data center network measurement method and positioning the current network performance bottleneck become increasingly important. This paper analyzes the current status of network measurement research, and presents an automatic measurement-distributed automatic measurement and performance bottleneck analysis method (AM-DMPA). The method can automatically generate the measurement task set, issue measurement tasks, and according to the measurement results, it can convergence the size of the network to be measured and can fast locate the network performance bottleneck link. Based on the typical large-scale data center of Tianhe 2, six different sizes of network environment are selected to compare AM-DMPA method with traditional methods of network measurement, the AM-DMPA method can quickly find and locate the bottleneck link in the network.

Key words: data center, distributed automatic measurement, performance bottleneck link

摘要： 针对当前数据中心网络规模巨大，流量行为复杂等特点，研究大规模数据中心网络测量方法，定位当前网络性能瓶颈变得日益重要。分析了当前网络测量研究现状，提出了一种基于分布式自动化测量的性能瓶颈分析方法（automatic measurement-distributed automatic measurement and performance bottleneck analysis method，AM-DMPA）。该方法能自动生成测量任务集合，下发测量任务，并能根据测量结果不断收敛待测网络规模，快速定位网络中的性能瓶颈链路。在典型的大规模数据中心天河2中，选取6种不同规模的网络环境，使用AM-DMPA方法和传统的网络测量方法进行对比，AM-DMPA方法能够更快速地发现和定位网络中性能瓶颈链路。

关键词: 数据中心, 分布式自动化测量, 性能瓶颈链路