Stalling Assessment for Wireless Online Video Streams via ISP Traffic Monitoring

摘要

Nowadays, ISPs witness an increasing network traffic generated by wireless online video services. Facing the intense mutual competition, it is important for an ISP to improve QoE of wireless video services. Since it is difficult for ISPs to collect user experience information at routers and switches, we propose an approach based on ISP traffic monitoring to access the stalling of a video stream, which is the key factor to video QoE. Concretely, we passively sniff out the following parameters to reconstruct video playing and stalling: playback rate of a slice, start-up delay and playback resuming threshold. Our difference from previous works is that we need not intrude into packets to extract HTTP meta information for these parameters. Moreover, we develop a video stalling assessment system based on Apache Storm, which can process the network traffic data in real time. Experiments on popular online video services (Tencent Video, iQiyi and Youku) in China as well as Youtube in US show not bad absolute difference of stalling times and total stalling duration between our traffic monitoring approach and real measurement.