Real-time Assessment Of Stalling For Online Video Based On Transport Layer

With the rapid development of mobile Internet,especially the popularization of 4G network,more and more users watch the network video.Facing a large number of Internet video users,ISP as Internet service providers,to provide Internet access services for users.If user suffer network delay,jitter,packet loss and other issues of network,video likely appears stalling phenomenon which reduces the quality of experience.Thus,ISP needs to take action to improve the quality of experience before many videos occur stalling.Facing severe market competition,it is not only inconvenient extract information from terminal but also expensive to purchase the relevant information from media service providers.However,ISP as an Internet access service provider,it can get video traffic data between server and client.How to improve the user experience through the network traffic has been gradually concerned.In this paper,our main work is reconstructed the process of video playback and detect number and duration of video stalling by using video traffic data,which is mainly composed of HTTP and TCP packets.The HTTP packet header contains the basic information of the video,and the TCP packet header contains the encoded video data.However,based on the protection of user privacy,ISP cannot intrude the HTTP packet to extract basic information of video,so we mainly reconstruct the process of video playback and detect the number and duration of video stalling by using TCP packet information.Since the video data of the TCP headers are encoded video clips,we estimate the basic information of the video before reconstructing the process of video playback.We propose methods to estimate video definition,video initial and video stalling and video resuming and reconstruct the process of video playback based on queue theory.Finally,we detect number and total duration of video stalling.In fact,the network traffic consists of text,pictures and video,etc.,which are complexity and great.ISP needs to filter non-video data and detect the video stalling,so we deploy detection system based on Apache Storm which is real-time deployment system.Experimental video detected are from Tencent video,IQiyi,Youku in China and You Tube in US.Finally,we evaluate the performance of our system with absolute error,relative error and least square fitting.The results show that our method has a better detection effect.