DDA: Cross-Session Throughput Prediction with Applications to Video Bitrate Selection

Abstract

User experience of video streaming could be greatly improved by selecting a high-yet-sustainable initial video bitrate, and it is therefore critical to accurately predict throughput before a video session starts. Inspired by previous studies that show similarity among throughput of similar sessions (e.g., those sharing same bottleneck link), we argue for a cross-session prediction approach, where throughput measured on other sessions is used to predict the throughput of a new session. In this paper, we study the challenges of cross-session throughput prediction, develop an accurate throughput predictor called DDA, and evaluate the performance of the predictor with real-world datasets. We show that DDA can predict throughput more accurately than simple predictors and conventional machine learning algorithms; e.g., DDA's 80%ile prediction error of DDA is > 50% lower than other algorithms. We also show that this improved accuracy enables video players to select a higher sustainable initial bitrate; e.g., compared to initial bitrate without prediction, DDA leads to 4x higher average bitrate.