ISP Video Stalling Detection Based On User Behavior

ISP(Internet service Provider)provides millions of users with network access and other services.As a result,competition among ISPs is extremely fierce.To improve the competitiveness of ISPs has been a major task.Videos are regarded as major part of network traffic.To provide video users with high user satisfaction network services can greatly enhance user experience,and thus effectively improve competitiveness of ISPs.Video Stalling is viewed as a main measurement of video user satisfaction QoE(Quality of experience),is an important entry point,so stalling detection from the ISP side can provide an effective means for ISPS to improve user satisfactionDue to user privacy however,most video websites use HTTPS as the application layer protocol of video transmission,and ISPs cannot obtain video user's current playback by resolving HTTPS packet,which means that Stalling cannot be detected directly.To solve this problem,this paper proposes a video stalling detection method that can adapt to the video stream transmitted using cryptographic protocol.This method provides ISP with effective means to improve the user experience,and then improve the competitiveness.The method proposed in this paper can effectively detect Video Stalling in various major video sources,including YouTube,Youku,LeTV and Tencent Video.The stalling detection model can be divided into two modules,stalling matching module and stalling decision module.Now that user's behaviors may have impacts on ISP's detection on stalling,which are non-negligible,we must design a stalling detection model as practical as possible for ISP.Therefore,this paper introduces the recognition of user's behaviors to the stalling decision module,which can effectively identify the user behaviors,and thus reduce the miscalculation rate of stalling detection.With limited information due to unresolvable HTTPS protocol,this model can identity user behaviors and carry out stalling detection by modeling the IP source of message data,message length,message type,message basic information.Considering that building an experimental environment is not an easy task,we build and containerize a stalling analysis platform to facilitate cross-platform migration.We conduct experiments in large amounts of random video samples from major mainstream video sources at home and abroad.After comparing the stalling matching rate as well as miscalculation rate of video samples with and without User Behavior Analysis,we conclude that the user Behavior Analysis module can effectively reduce the miscalculation rate without significantly reducing the matching rate.According to summarized results of stalling detection in each video source,the model proposed achieves satisfactory robustness and excellent stalling detection effects on major videos at home and abroad.