Shedding light on the structure of internet video quality problems in the wild

摘要: The key role that video quality plays in impacting user engagement, and consequently providers' revenues, has motivated recent efforts improving the of Internet video. This includes work on adaptive bitrate selection, multi-CDN optimization, global control plane architectures. Before we embark deploying these designs, need to first understand nature problems see if this complexity is necessary, simpler approaches can yield comparable benefits. To end, paper a attempt shed some light structure problems. Using measurements from 300 million sessions over two-week period, identify recurrent using hierarchical clustering approach space client/session attributes (e.g., CDN, AS, connectivity). Our findings are that: (1) small number (2%) critical clusters account for 83% join failure (44--84% other metrics); (2) many problem events (50%) persist at least 2 hours; (3) majority 60% failures, 30--60% metrics) related content provider, or client ISP issues. Building insights, evaluate potential improvement by focusing addressing find fixing just 1% reduce problematic 55% failures (15%--40% metrics).