SFU (Selective Forwarding Unit): Low-Latency Architecture for Small-to-Medium Video Conferences

1. Core Definition

SFU (Selective Forwarding Unit) is a widely used network architecture in video conferences, with its core function centered on forwarding audio and video streams without any mixing processing. Its workflow is streamlined and efficient:

1. Audio and video streams from each participant are transmitted to the SFU server;
2. The server does not encode, decode, or merge these streams (unlike the MCU (Multipoint Control Unit) architecture, which mixes streams into a single unified stream);
3. Instead, the server directly forwards each participant’s independent stream to all other attendees.

This "no-mixing" design gives SFU two key advantages: ultra-low Latency (typically ≤ 100ms) and low server resource consumption, making it the first choice for most enterprises’ daily small-to-medium-sized meetings (≤ 200 participants).

2. Core Value

SFU addresses the core needs of small-to-medium meetings—real-time interaction and efficient resource use—by delivering three irreplaceable benefits:

• Minimal Latency: Without time-consuming mixing processing, streams are forwarded almost instantly, ensuring natural, face-to-face-like interaction (no delays like "hearing a response seconds after speaking");
• Efficient Resource Use: The server avoids the high computing pressure of mixing multiple streams (common in MCU), reducing hardware costs and operational load;
• Flexible Stream Control: Each participant receives independent streams from others, making it easy to switch between individual views (e.g., focusing on a speaker’s video or a shared Screen Sharing stream) without compromising quality.

3. Key Application Scenarios & Practical Examples

SFU’s low-Latency and efficient design make it ideal for scenarios requiring real-time collaboration or small-scale interaction. Below are its most impactful use cases:

3.1 Small-to-Medium Enterprise Department Meetings

For departmental discussions (e.g., R&D progress reviews, marketing strategy meetings) with ≤ 10 participants across multiple offices, SFU ensures real-time communication and smooth Screen Sharing.

• Why SFU Works: Traditional MCU mixing would add unnecessary Latency for small groups. SFU’s direct forwarding keeps Latency ≤ 100ms, ensuring shared content (e.g., code, PPTs) and speech stay synchronized.
• Practical Example: A software company’s R&D team holds a code review meeting with 10 programmers across 3 offices. Using SFU:
  • When a programmer shares their screen to display code snippets, the stream is forwarded to others instantly—no lag between their mouse clicks and the team’s view;
  • Discussions about code logic are real-time: if a teammate points out a bug, the presenter can modify the code and show the fix immediately. The experience is as seamless as being in the same conference room.

3.2 Remote Interview Scenarios

Remote job interviews rely on real-time interaction to assess candidates’ communication skills and responsiveness—SFU’s ultra-low Latency (50–80ms) eliminates interaction barriers.

• Why SFU Works: High Latency (common in low-end architectures) would cause awkward pauses between a candidate’s answer and the interviewer’s follow-up, distorting the evaluation. SFU’s direct forwarding ensures natural dialogue.
• Practical Example: A tech company uses SFU for campus recruitment interviews, connecting interviewers (headquarters) with candidates (universities in 5 cities):
  • Candidates’ facial expressions (e.g., confidence when explaining projects) and gestures (e.g., pointing to a portfolio on screen) are transmitted to interviewers in real time;
  • Interviewers’ questions are heard instantly by candidates, avoiding disruptions to their train of thought. Post-interview feedback shows 90% of interviewers rated the "interaction naturalness" as "equal to in-person interviews."

3.3 Small-Scale Training Meetings

For training sessions with ≤ 20 participants (e.g., new employee onboarding, product knowledge training), SFU ensures synchronous delivery of course content and real-time Q&A.

• Why SFU Works: It forwards the trainer’s audio/video stream and Auxiliary Stream (e.g., courseware) simultaneously, with no time lag between the trainer’s explanation and the content trainees see. It also supports quick stream synchronization for late joiners.
• Practical Example: An e-commerce company conducts new employee training for 20 hires. Using SFU:
  • The trainer shares product knowledge courseware via Auxiliary Stream, and the stream is forwarded to all trainees in real time—no trainee sees the courseware later than the trainer’s explanation;
  • A trainee who joins 10 minutes late connects instantly: the SFU synchronizes the current meeting streams (trainer’s video + courseware) within 2 seconds, allowing them to catch up without waiting;
  • Trainees can raise hands to ask questions, and their voices are forwarded to the trainer immediately—ensuring active participation and efficient knowledge transfer.

4. Key Takeaway

SFU is not designed for ultra-large-scale meetings (e.g., 500+ participants) where MCU’s stream mixing reduces Bandwidth load. Instead, it excels in small-to-medium, interaction-intensive scenarios:

• Daily department meetings (≤ 10 participants);
• Remote interviews (2–5 participants);
• Small-scale training (≤ 20 participants).

By prioritizing low Latency and efficient resource use, SFU delivers a smooth, natural meeting experience that aligns with the needs of most enterprises’ daily communication.