AsTree: An Audio Subscription Architecture Enabling Massive-Scale Multi-Party...
summary
AsTree replaces full “subscribe-to-all” audio with a cascading tree topology over SFUs and media-plane audio selection that forwards only a few loudest streams per node, cutting signaling and stream fan‑out. Implementation in Lark and benchmarks show dramatically lower CPU peaks versus full subscription, supports rooms >10,000 participants, and reduces audio and video rebuffering by >30% and >50% respectively; user negative reviews drop by 40% under production load.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:21SFU conferencing architecture
- 2:22Audio subscription scalability
- 5:58S3 core idea
- 6:41S3 high-level design
- 7:24Delegate server selection
- 10:34Master delegate election
- 11:59Audio stream selection
- 13:02Evaluation results
- 14:04Conclusion
related talks
19mScaling for Large Group Calls
25mA Stateless Service for Audio Processing
32mBuilding Real-Time Audio/Video Apps with Flutter and WebRTC
17mLive Video Streaming as "tail -f"
19mAudio Streaming - Using WebRTC for building your own Voice AIs
23mIntroduction to Media Streaming Mesh
23mMoQ: Not Another Tech Demo
30mJitsi Videobridge: the state of the art SFU that powers Jitsi Meet