移动网络加密YouTube视频流QoE参数识别方法  

A Method for Identifying the QoE Parameter of Encrypted YouTube Traffic in Mobile Network

作　　者：潘吴斌[1,2] 程光[1,2] 吴桦[1,2] 徐健 PAN Wu-Bin;CHENG Guang;WU Hua;XU Jian(Key Laboratory of Computer Network and Information Integration,Ministry of Education,Nanjing 210096)

机构地区：[1]东南大学计算机科学与工程学院,南京210096 [2]计算机网络和信息集成教育部重点实验室(东南大学),南京210096

出　　处：《计算机学报》2018年第11期2436-2452,共17页Chinese Journal of Computers

基　　金：本课题得到国家“八六三”高技术研究发展计划项目基金(2015AA015603);江苏省未来网络创新研究院未来网络前瞻性研究项目(BY2013095-5-03);江苏省“六大人才高峰”高层次人才项目(2011-DZ024);中央高校基本科研业务费专项资金和江苏省普通高校研究生科研创新计划资助项目(KYLX15_0118)资助.

摘　　要：移动视频业务应用广泛,流量占比高且持续增长.针对有限的移动网络带宽,如何合理地规划网络服务、提供优质的移动视频体验,需要客观的视频体验评估反馈网络服务提供商和视频服务运营商以改善网络利用率及传输方案.当前大多数视频服务质量评估方法都基于DPI(Deep Packet Inspection)方法获取视频播放信息以计算视频QoE(Quality of Experience).然而,为了保护用户隐私和网络安全,越来越多的视频采用HTTPS加密传输,使得传统的DPI方法无法获取码率和清晰度等QoE评估参数.因此,文中提出一种基于视频块统计特征的加密视频QoE参数识别方法(以代表性网络视频YouTube为例).首先,根据SSL/TLS协议握手过程中未加密部分识别HTTPS加密的YouTube流量.然后,根据视频流前若干个包的4种特征识别出HLS、DASH和HPD传输模式,再根据视频块统计特征建立机器学习模式识别视频块的码率和清晰度.实验结果表明该方法传输模式、码率和清晰度识别平均准确率分别达到98%、99%和98%,可以有效用于加密YouTube的QoE评估.YouTube is one of the most popular and volume-dominant video streaming services in today’s Internet.Almost 50%of the YouTube views are from mobile users,and this trend is expectedly increasing in near future.For a limited mobile network bandwidth,how to reasonably plan network services in order to provide high-quality mobile video experience?The answer requires the objective feedback of video experience for network service providers and video service providers to improve network utilization and transmission strategy.Due to the feasibility and cost effectiveness,objective video QoE(Quality of Experience)assessment is commonly used to estimate the user perception on the quality of video streaming services.Active probing can only provide instant samples but cannot accurately represent the actual network condition over the entire period of video streaming service.In contrast,passive measurement can be performed either at the client devices or in network.However,client-side measurements are more intrusive as end users are directly involved,who can provide accurate view from an individual’s perspective on several objective Key Performance Indicators(KPIs).Many studies regard Stallings and initial delays on the video playback as the most relevant KPIs for QoE in HTTP video streaming.In the case of adaptive streaming,quality switches is also regarded as an important factor on QoE.Although it is clear that all these KPI factors indeed have impact on QoE assessment result,there is very few QoE assessment framework proposed to systematically combine multiple KPIs to present their joint effect on QoE.Therefore,cooperated with human factors engineering lab,Huawei maps the subjective feelings to objective KPIs to establish video MOS(vMOS)evaluation framework to synthetically assess video QoE,which is composed of video source quality,initial buffering latency,and stalling ratio.However,the resulting QoE has to be estimated from traffic characteristics typically via DPI(Deep Packet Inspection),which becomes infeasible after

关 键 词：HTTPS视频流量 机器学习 QoE参数识别 体验质量评估 加密YouTube 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]