SFU (Selective Forwarding Unit)

An SFU (Selective Forwarding Unit) is a media server that sits between WebRTC participants and relays each incoming video stream to every other participant, without decoding or re-encoding the media — a deliberate contrast with an MCU (Multipoint Control Unit), which composites all streams into a single mixed output. Because the SFU merely forwards compressed packets, it consumes far less CPU than an MCU and can handle many more concurrent streams on the same hardware. For a live classroom this matters: a mesh of direct peer connections between 30 participants would require each browser to send 29 simultaneous upload streams, crushing most consumer connections, whereas an SFU means each participant uploads once and the server handles redistribution. The SFU gains extra efficiency when senders use simulcast — transmitting the same camera feed at multiple quality levels simultaneously — because the server can then selectively forward only the resolution appropriate for each receiver's bandwidth. SVC (Scalable Video Coding) achieves a similar result within a single encoded stream using dependent layers. The trade-off is that an SFU does not reduce the total download burden on viewers in proportion to participant count: a viewer still receives one stream per visible participant tile. Popular open-source SFU implementations include mediasoup, Janus, and Pion, while managed cloud offerings abstract the infrastructure entirely.

The handbook · 7 volumes

Fora Soft Video Engineering Handbook

An SFU is what lets a live class grow past a handful of people, and its configuration decides who sees a frozen teacher. The seven volumes cover the video stack; for SFU design start with Volume 3 on WebRTC, real-time delivery, and QoE monitoring, then Volume 2 on congestion control and NAT, with codecs in Volume 1. Written by Nikolay Sapunov, CEO at Fora Soft.

Browse the handbook on Amazon →7 volumes · 9,000+ pages · Kindle & paperback