Exploiting Trace Data for Adaptive Mobile Video Streaming with Performance Guarantees

Current mobile video streaming is primarily delivered over HTTP. In addition to conventional web access logs, many service providers also log the data usage for billing and analytics purposes. This work exploits these vast amount of bandwidth trace data to enable the provisioning of streaming performance and video quality guarantees in adaptive mobile video streaming. We develop a post-streaming analysis framework to extract from past trace data statistical correlations between key system parameters that enables the automatic configuration of future streaming parameters including startup delay and rate adaptation such that a given minimum video quality and a target playback rebuffering probability can be maintained. The proposed framework is evaluated and compared to two existing adaptive streaming algorithms via a trace-driven simulator using extensive throughput trace data captured from a production 3G/HSPA network.