Scaling for Large Group Calls
summary
Explains a server-side scenario forwarding architecture that uses dominant speaker detection plus audio capping to forward only the top X active talkers, reducing server→client audio traffic by about 38% for 20-person calls and 63% for 50-person calls. Describes SFU-side ranking using per-packet audio level headers, noise-floor normalization, and two rankers for stability vs responsiveness. For video it combines simulcast with temporal layers and a new stream subscription manager plus plan-from-BW allocator to avoid oscillation, respect pin/viewport priority, and serve super-low-bandwidth users.
AI-written summary · about
chapters (3)
related talks
17mGroup Call End-to-End Encryption and the Challenges of Encrypting Large Calls
30mJitsi Videobridge: the state of the art SFU that powers Jitsi Meet
15mAsTree: An Audio Subscription Architecture Enabling Massive-Scale Multi-Party...
57mMastering WebRTC for Large Conference Calls on Android
20mBeyond P2P: Video Routing in WebRTC
53mA Million Video Calls a Day, or Call Your Mom
10mRTC @Scale, Future RTC Experiences Session - Live Q&A
41mServers for WebRTC: It is not all Peer to Peer