软件定义网络已成为未来网络发展的趋势。数据中心网络是典型的高速、大规模网络，这使得软件定义数据中心网络流量测量面临高负载问题。针对该问题，本课题研究面向软件定义数据中心网络的流量测量机制，内容包括：1）利用基于卷积神经网络的网络流量空间特征提取和迁移学习，来研究基于终端主机的网络大流检测机制，目标是有效降低大流检测的负载；2）基于贝叶斯网络的网络流量时间特征提取和基于复杂网络的自适应流抽样机制的研究，其目的是降低抽样频率、提高抽样效率；3）针对全网络范围的网络测量，通过QoS指标统一化表征和面向QoS的网络流量测量部署优化机制的研究，目标是在以保证QoS为基础优化网络流量测量的负载。项目研究成果对于缓解网络规模急剧增长带来的网络流量测量高负载问题具有重要意义。

The software defined network has been the trend of our networks in the future. As a typical high-speed and large-scale network, the software defined data center network has a high overhead for network traffic measurement. Motivated by that, this project studies software defined data center network-oriented network traffic measurement, which includes: 1) by using the convolutional neural network to extract the spatial feature of network traffic and transfer learning, this project proposes an end host-based strategy for elephant flow detection, aiming at decreasing the overhead of elephant flow detection; 2) the extraction of temporal feature of network traffic based on Bayesian network and the self-adaptive flow sampling strategy based on complex network are studied in order to decrease the frequency of flow sampling and improve the efficiency of flow sampling; 3) considering the direct measurement of the whole network, the modeling of quality of service (QoS) and the strategy of QoS-oriented optimization of network traffic monitor are studied, with the target of improving the overhead of direct measurement and guaranteeing QoS. This project has important theoretical significance and application value to alleviate the problem of high measurement overhead as the scale of network increases sharply.