3 Second End to End Latency At Scale
summary
Claim: 3-second camera-to-screen latency at scale. Implementation uses a custom waves over WebSockets stream protocol, an MSE-based player with a quarter-second MSE buffer delivering two frames at a time, H.264/AAC single-bitrate, and an Origin–Mid–Edge Azure deployment. Measurement injects latency beacons to compare timestamps. Lab and preview customers see 1.5–2 s observed end-to-end; ABR and iOS MSE absence remain unsolved, and next steps include SRT for ingest and experimenting with HTTP chunked transfer techniques.
AI-written summary · about
chapters (9)
- 0:00Start
- 4:58Considerations: Chunk Based?
- 6:12Considerations: WebRTC? WebSockets?
- 8:28Prototype
- 10:54Delivery Network
- 13:16Player
- 14:50Limitations
- 17:18Monitoring and Measuring
- 18:54What's next?
related talks
21mThe challenges of deploying Apple's Low Latency HLS In Real Life
19mLow Latency Live from a Different Vantage Point
18mVideo @Scale 2022 LIVE Q&A #1
10mRTC @Scale, Future RTC Experiences Session - Live Q&A
04mSub-frame Latency
17mLive Video Streaming as "tail -f"
19mAdventures in WebRTC-based Content Streaming
16mCDN Challenges of HTTP-based Low Latency Live Streaming Delivery