Choosing the Right WebRTC Architecture: Mesh, MCU, or SFU?

In real‑time multiparty audio/video communication, 1‑to‑1 is just one scenario; education and meetings often require 1‑to‑many or many‑to‑many. The three main architectures are Mesh, MCU (Multi‑Point Control Unit), and SFU (Selective Forwarding Unit).

1. Mesh Architecture

As illustrated, each of five browsers establishes peer‑to‑peer connections with the other four, creating a mesh topology with ten connections. If each connection uses 1 M bandwidth, each client needs 4 M up‑ and down‑link, totaling 20 M. No media server is required; browsers handle encoding/decoding. Mesh is simple to implement and works for 4‑6 participants.

Advantages: Minimal technical difficulty, no central server (only an ICE server for NAT traversal), fully decentralized, leverages client compute, reduces server cost.

Disadvantages: High upstream bandwidth demand on each client, limited participant count, handling NAT failures requires extra fallback strategies.

2. MCU (Multi‑Point Control Unit)

Each browser connects only to a central MCU service, which handles all video encoding, transcoding, decoding, and mixing.

Receive audio/video streams from senders.

Decode the streams.

For video, re‑layout and mix; for audio, mix and resample.

Re‑encode the mixed streams.

Send the resulting stream to receivers. Each client maintains a single connection, so total bandwidth for five participants is 10 M.

Advantages: Minimal bandwidth per client (≈2 M), supports more participants, central processing hides device differences, mature and widely used technology.

Disadvantages: High server CPU consumption, added latency from encoding/decoding, limited flexibility for client‑side custom layouts or effects.

3. SFU (Selective Forwarding Unit)

Each browser connects to a central SFU service, which only forwards RTP packets without decoding or mixing. Clients send their own stream upstream and receive n‑1 streams downstream, resulting in higher total bandwidth (e.g., 25 M for five participants).

Advantages: Server only forwards packets, low CPU load, can adapt bitrate per stream, flexible layout handling.

Disadvantages: Clients must synchronize multiple streams, higher client bandwidth demand, less consistency across devices.

Which Architecture to Choose?

For small groups (≤5 participants), Mesh is sufficient due to its simplicity. For up to 50 participants with limited bandwidth, MCU offers a good balance. For larger scale and ample bandwidth—especially in the 5G era—SFU is generally the most suitable choice.

References: WebRTC Multiparty Video Alternatives, and Why SFU is the Winning Model; WebGlossarySFU; WebRTC Media Servers – SFUs vs MCUs.