SFU WebRTC: Selective Forwarding for enterprise video
Short answer
An SFU (Selective Forwarding Unit) is a WebRTC server that receives each participant’s stream and forwards it selectively to others without transcoding. It keeps low latency, limits client CPU load, and scales multi-party meetings—the standard model for professional video infrastructure.
How does an SFU session work?
- Signaling (HTTPS/WSS): room creation, SDP exchange;
- ICE / STUN / TURN: network path to the SFU;
- Publish: each participant sends audio/video/screen to the SFU;
- Selective forward: SFU routes relevant streams per receiver;
- Adaptation: simulcast/SVC adjusts quality to bandwidth.
SFU vs MCU vs P2P
| P2P | SFU | MCU | |
|---|---|---|---|
| Model | Direct client links | Selective relay | Mixed composite stream |
| Transcoding | No | No | Yes |
| Latency | Lowest (1:1) | Low | Often higher |
| Scale | ~2–3 participants | Dozens to hundreds+ | Variable, costly |
| Modern pro usage | Short 1:1 | Standard | Legacy / hardware rooms |
Why SFU is central for scalability
- Horizontal scaling: SFU node cluster behind a load balancer, with session pinning (one SFU per session);
- Simulcast: client encodes multiple qualities; SFU sends the one suited to each receiver;
- SVC (Scalable Video Coding): dependent video layers (VP9, AV1) for fine adaptation without retranscoding;
- Selective forwarding: speaker detection, dynamic layout, reduction of unnecessary streams;
- Low latency: no intermediate transcoding pipeline.
Sizing strategies, associated TURN capacity and cluster approaches are detailed on scalable WebRTC infrastructure.
Where to host the SFU
The SFU processes real-time media streams: a critical component for sovereignty and compliance.
| Model | Use case | Watch point |
|---|---|---|
| France / EU cloud | Standard deployment, EU data residency | Verify actual SFU node location, not just the web portal |
| On-premise | Regulated sectors, strict IT policies | Plan hardware sizing and redundancy |
| Hybrid | Internal staff + external guests | Internal SFU + TURN relay for blocked connections |
See France hosting and sovereign video cloud.
Symptoms of an undersized SFU
| Symptom | Likely cause |
|---|---|
| Stable at 5, unstable at 20 | Insufficient SFU or bandwidth |
| Latency rises with participants | SFU node CPU or network saturation |
| Quality degraded for everyone | No simulcast / SVC |
| Random disconnections | ICE timeouts, no backup TURN |
| OK in pilot, degraded in production | Real load not simulated, cluster not sized |
A real network pilot (corporate workstation + 4G + screen share) remains the best test before general rollout.
FAQ
When is an SFU required?
From 3+ participants in real-time interaction, P2P becomes unsustainable; SFU is the expected architecture.
Does SFU work on corporate networks?
Yes with documented STUN/TURN and firewall rules—part of enterprise rollout.
How does SFU relate to the video API?
The SFU is the media plane; video API and signaling are the control plane for your applications.
Key takeaways
- SFU is the distribution layer of professional WebRTC—not a replacement for WebRTC standards.
- Prefer SFU over MCU for scalable browser-based meetings.
- Size STUN/TURN and SFU capacity with your IT team before production rollout.
Next step
Request a quote to scope SFU capacity, TURN, and sovereign hosting.