WebRTC scalability: size video infrastructure
Short answer
WebRTC scalability relies on SFU (Selective Forwarding Unit) horizontal clustering, simulcast (multiple qualities per stream), SVC (Scalable Video Coding), and proper TURN sizing. Unlike pure peer-to-peer, SFU scales from dozens to hundreds of participants per session.
Why P2P does not scale
In peer-to-peer, each participant sends N-1 streams and receives N-1 streams. At 20 participants, each client handles 19 streams — unsustainable.
The SFU fixes this: each client sends 1 stream to the server, which selectively redistributes.
SFU scaling strategies
1. Horizontal scaling
- SFU node cluster;
- Load balancer (round-robin, geo-routing);
- Session pinning (one SFU per session).
2. Simulcast
The client encodes 3 qualities (low, medium, high). The SFU sends the appropriate quality per receiver based on bandwidth.
3. SVC
Dependent video layers (VP9 SVC, AV1): SFU forwards required layers without retranscoding.
4. Selective forwarding
The SFU forwards only relevant streams (speaker detection, dynamic layout).
Sizing orders of magnitude
| Session | Recommended SFU | SFU bandwidth |
|---|---|---|
| 1:1 | P2P or light SFU | ~2–6 Mbps |
| 10 participants | 1 SFU node | ~20–60 Mbps |
| 50 participants | 2–4 SFU nodes | ~100–300 Mbps |
TURN relay sessions add separate bandwidth — size independently.
Scalability and latency
Scalability must not degrade latency: horizontal SFU (not MCU) is the standard.
How does Leagora scale WebRTC?
Sovereign WebRTC infrastructure with horizontal SFU, simulcast, monitoring. France hosting or on-premise. Contact.
FAQ
Does scaling require MCU?
No. SFU horizontal scaling is preferred for low latency.
How to size TURN vs SFU?
Measure relay session percentage in pilot. TURN bandwidth can exceed SFU in corporate networks.
Can one SFU handle multiple tenants?
Yes with tenant isolation and per-use configuration.
Key takeaways
- SFU horizontal cluster + simulcast = WebRTC scalability standard.
- Size SFU and TURN separately.