Dashlet: Taming Swipe Uncertainty for Robust Short Video Streaming
summary
Implements out-of-order chunk pre-buffering that schedules specific chunks of upcoming short videos based on a simple non-ML swipe model derived from two in-the-wild swipe datasets, then replays real TikTok traces under controlled networks. Compared to the reverse-engineered TikTok client it reports 28–101% QoE improvement and about a 30% reduction in bytes downloaded but never watched, with performance approaching an oracle that knows future swipes.
AI-written summary · about
chapters (10)
- 0:00Start
- 2:14Traditional vs Short video streaming
- 5:55Understand users' swipe pattern
- 6:35Swipe distribution across dataset is stable
- 7:23Buffer order in traditional video streaming
- 8:02Dashlet: play start time
- 8:50Dashlet: Algorithm
- 9:23Implementation
- 10:29Key results
- 13:58Conclusion
related talks
16mImproving Video Delivery and Streaming Performance with...
14mUsing Performance Signals to Improve Video Recommendations
17mPredict, Prune, Play: Efficient Video Playback Optimization Under Device...
16mSwift: Adaptive Video Streaming with Layered Neural Codecs
50mDemuxed 2023 Lightning Talks
49mQwik: Streaming JavaScript framework
45mJavaScript Streaming - A Qwik glimpse into the future
26mMagic Mirror: Understanding the Diversified Network Conditions of Twitch Community