Video streaming on mobile devices is prone to a multi-tude of faults and although well established video Qual-ity of Experience (QoE) metrics such as stall frequencyare a good indicator of the problems perceived by theuser, they do not provide any insights about the natureof the problem nor where it has occurred. Quantifyingthe correlation between the aforementioned faults andthe users' experience is a challenging task due the largenumber of variables and the numerous points-of-failure.To address this problem, we developed a frameworkfor diagnosing the root cause of mobile video QoE is-sues with the aid of machine learning. Our solutioncan take advantage of information collected at multiplevantage points between the video server and the mobiledevice to pinpoint the source of the problem. More-over, our design works for different video types (e.g.,bitrate, duration, ..) and contexts (e.g., wireless tech-nology, encryption, ..) After training the system witha series of simulated faults in the lab, we analyzed theperformance of each vantage point separately and whencombined, in controlled and real world deployments. Inboth cases we find that the involved entities can inde-pendently detect QoE issues and that only a few vantagepoints are required to identify a problem's location andnature

Peer Reviewed