Providing better video experiences for the next billion users.
summary
A project uses an internal root cause tool to trace stalls from client to CDN and defines a bad session rate per user metric. Encoding changes add 100 and 200 kbps lanes, relax unique resolution rules, and move GOPs from two to five seconds for roughly 6–8% compression efficiency. ABR adds a minimum quality bar and outlier-resistant bandwidth estimation, playback adds phased prefetch and cache limits, and network switches to QUIC, reducing stalls and improving time to first byte and watch time.
AI-written summary · about
chapters (10)
- 0:00Introduction to project
- 1:38Analyzing playback issues
- 3:29Impact of COVID-19
- 5:11Defining success metrics
- 7:18Encoding improvements
- 11:28Adaptive streaming logic
- 14:25Bandwidth estimation
- 15:18Playback side optimizations
- 17:38Network protocol upgrades
- 20:40Results and conclusion
related talks
03mImproving ABR at Netflix
19mUsing application informed pacing to be a friendly internet neighbor
11mFree ABR Renditions for User Generated Content Platforms
10mGoing Beyond Bitrate: Adaptation Based on Video Context
13mSENSEI: Aligning Video Streaming Quality with Dynamic User Sensitivity
26mBandwidth Ballet: Optimizing Video Performance for Web Maestros
41m8K Codec Low-Latency Livestreaming, and AI Streaming
17mMetrics Driven Obsession with Viewer Experience