WebRTC is a real-time communication technology based on P2P. (If P2P hole punching fails, the TURN server will be used for data forwarding.) With 2 clients participating, the network topology is shown in the figure:

1. Mesh architecture

With only two clients participating, we can use the topology shown above. But if there are 3 clients participating at the same time (such as multi-person video conference), if still follow the above method, the topology will become:

As we can see from the picture above, in the real-time communication involving 3 people, each client must maintain 4 connections (2 upstream, 2 downstream); the same way, If there are N clients participating, each client must maintainN-1Up,N-1One downstream, which will greatly occupy the upstream bandwidth and downstream bandwidth of the client;

This kind of architecture that uses P2P before each end is called Mesh architecture.

2. SFU (Selective Forwarding Unit) architecture

The core feature of the SFU architecture is to "disguise" itself as a WebRTC client. The other clients of WebRTC do not actually know that they are connected through P2P in the past is a real The client is still a server, we usually call this connection P2S, namely: Peer to Server. In addition to "disguised" as a WebRTC client, the SFU server also has one of the most important capabilities is to haveone-to-manyThe ability to forward data from one client to multiple other clients.

In this network topology, no matter how many people make video calls at the same time, each WebRTC client only needs to connect to one SFU server, and only one uplink data is needed, which can greatly reduce the Mesh model to the client in the multi-person video call scenario. Bringing upward bandwidth pressure.

The biggest difference between the SFU server and the TURN server is that the TURN server is only an auxiliary data forwarding channel provided for WebRTC clients, and transparent data forwarding is performed when P2P fails. SFU is "business-aware", it has an equal relationship with WebRTC client, and even "takes over" the application and control of WebRTC client's data forwarding.

3. MCU (MultiPoint Control Unit) architecture

As can be seen from the definition of SFU above, the network topology model of SFU, through the SFU Server to achieve one-to-many, reduces the upstream bandwidth pressure of each client in the multi-person video call scenario, but the downlink is still multi-path Streaming, as the number of callers increases, the pressure on the downstream bandwidth will still increase proportionally, so can there be only one stream of downstream traffic? —— Yes, it can be solved by consolidating on the server side and then issuing, as shown in the following figure:

At present, with the promotion of 5G technology, it is foreseeable that bandwidth is increasingly not a problem, so SFU may have more advantages in the future. Common open source SFU servers are:Licode，Janus，Jitsi，mediasoup，Medooze and many more.

Article reference:
WebRTC development practice: why you need an SFU server
webrtc notes (3): Multiplayer video communication commonly used architecture Mesh / MCU / SFU