VPAP: VBR Pattern Aware Playback Buffering for video streaming

Abstract

When media is streamed over networks that only provide best-effort delivery, playback interruptions caused by variations of network throughput can be largely eliminated by using techniques such as client-side playback buffering. A larger buffer generally provides stronger protection against playback interruptions, since it accommodates a wider range of download rate variability. However, larger buffers also incur longer start-up delays, and longer delays in re-buffering, should a playback interruption occur. In this paper, we propose a scheme that enables the client-side player to dynamically calculate the minimum possible playback buffer that it needs given an estimation of network throughput, and may thus allow an earlier start to playback, while avoiding buffer under-runs. We have confirmed that a number of popular, real-world videos—encoded using Variable bitrate (VBR) schemes—contain bitrate patterns that allow them to safely start playback earlier than their average bitrate metadata would suggest. Our scheme augments each video stream so that client-side software can use a model of the network throughput to determine whether it can safely start playback early. We tested the proposed scheme against two popular media players using real video sequences. Among the tested videos, in the best case, our scheme could start playback with less than 1% of the delay incurred by one of the tested players, while still avoiding buffer under-runs.