SDN中基于机器学习的网络流量分类方法研究

NETWORK TRAFFIC CLASSIFICATION IN SDN BASED ON MACHINE LEARNING

由于缺乏网络流量类别信息,目前软件定义网络SDN控制层难以有针对性地对在线视频流量和下载流量进行速率管控。当带宽有限时这会严重影响用户同时观看在线视频和进行下载时的体验。针对此问题,提出一种在SDN中基于机器学习的在线视频流量和下载流量分类方案。该方案选取新的、可以有效区分在线视频流量和下载流量的特征集合。通过测试对比多种机器学习模型的分类效果,在SDN中设计实现了基于随机森林(RandomForest)模型的实时流量分类应用,为在SDN中实现细粒度的网络流量管控、优化QoS等功能奠定了基础。测试结果表明,该方案对SDN中在线视频流量和下载流量的实时分类效果较理想,实时分类平均准确率较高。

Due to the lack of information of network traffic types,it is difficult for the SDN control layer to implement the rate control over online video traffic and download traffic.When the bandwidth is limited,it seriously affects the user experience while watching online video and downloading.Aiming at this problem,we proposed a traffic classification scheme of online video and download based on machine learning in SDN.This scheme selected a new feature set that could effectively distinguish online video traffic and download traffic.Through testing and comparing the classification effects of various machine learning models,we designed and implemented a real-time network traffic classification application based on random forest model in SDN.It laid the foundation for fine-grained network traffic control and optimization of QoS in SDN.The test results show that the scheme has a good real-time classification effect on online video traffic and download traffic in SDN.And the average accuracy of real-time classification is higher.