Ever been in a conference call where someone dropped out, the audio got muddy, or you couldn’t hear the person who just spoke? It’s not just bad internet. It’s the architecture behind the call that’s failing you. Modern conference calling isn’t just about connecting people-it’s about managing how audio and video flow between dozens, even hundreds, of participants without turning the call into chaos. And the secret lies in two main systems: MCU and SFU.
How Conference Calls Actually Work (Beyond the Dial-In Number)
Think of a conference call like a group conversation in a crowded room. Everyone talks, but not everyone can be heard at once. In old-school phone systems, a central switchboard would mix all voices into one stream and send it back to everyone. That’s the MCU model-Multipoint Control Unit. It’s like a sound engineer in a studio, taking every input, blending it into one clean output, and broadcasting it to all listeners.
But here’s the catch: every time you add a new person, the server has to process another audio and video stream, mix it with all the others, and re-send the new combined version. For 10 people, that’s 10 inputs and 10 outputs. For 30 people? That’s 30 inputs and 30 outputs-and the server is doing 900 mixing operations per second. That eats up CPU, memory, and bandwidth. Most MCU-based systems max out around 20-30 participants before quality drops or calls crash.
That’s why companies like TrueConf and older video conferencing bridges struggle with larger groups. Even high-end hardware can’t keep up with the math. And if one person’s connection glitches? The whole mixed stream gets distorted for everyone.
The SFU Revolution: Why Zoom and Google Meet Handle Hundreds of People
Enter SFU-Selective Forwarding Unit. Instead of mixing everything, the server acts like a traffic cop. Each participant sends one stream to the server. The server doesn’t touch the audio or video. It just decides who gets what. If you’re in a call with 50 people, your device receives only the streams from the people the system thinks you should hear-maybe the active speaker, maybe a few others you’ve pinned.
This changes everything. No more heavy mixing. No more server overload. Zoom, Google Meet, and Wire all use SFU. And that’s why Zoom can handle 1,000 video participants in enterprise meetings. Each participant only needs to send one stream up. The server doesn’t do heavy lifting-it just routes. The heavy work? That’s on your device, decoding a few streams instead of one mixed one.
But SFU isn’t perfect. If you’re on a weak internet connection and 20 people are sharing video, your phone or laptop has to decode 20 separate video streams. That can drain battery, slow down your device, or cause lag. That’s why Zoom and others automatically reduce video quality for inactive participants or switch to audio-only when bandwidth drops.
MCU vs SFU: The Real Trade-Offs
Here’s the breakdown:
| Feature | MCU (Multipoint Control Unit) | SFU (Selective Forwarding Unit) |
|---|---|---|
| How streams are handled | All streams mixed into one on server | Streams routed individually, no mixing |
| Max participants per server | 20-50 | 50-500+ |
| Server load | Very high (CPU-intensive) | Low to moderate |
| Client device load | Low (only one stream to decode) | High (multiple streams to decode) |
| Video layout control | Full control (grid, spotlight, etc.) | Limited (depends on client) |
| Best for | Telemedicine, boardrooms, regulated environments | Large meetings, remote teams, webinars |
MCU still has its place. In healthcare, for example, a doctor might need every specialist’s video visible in a fixed grid layout during a consultation. Mixing streams into one video frame makes that consistent across all devices. But for most business use cases? SFU wins. It’s cheaper to scale, more reliable, and handles modern workstyles better.
Real-World Limits: Why Your RingCentral Call Drops at 8 People
RingCentral claims 10 participants per call. But users on Reddit and support forums report calls starting to glitch at 7 or 8. Why? Because even though RingCentral uses SFU internally, their system still has hard limits built into the API and session management. Each participant needs a unique telephony session ID. If the system can’t assign or track those IDs reliably, it drops connections.
And then there’s AWS Connect. It caps conference calls at six participants total-agent, caller, and four others. That’s not a technical limit. It’s a design choice. Contact centers prioritize agent control: muting, holding, transferring. More participants means more complexity in managing who’s speaking, who’s on hold, and who’s being transferred. For a customer service agent juggling three calls at once, simplicity beats scale.
Genesys is catching on. After hearing from hospital IT teams that they’re forced to use personal phones for multi-clinician consults (creating compliance risks), Genesys moved its 6-participant upgrade from "target" to "committed"-with a release expected in Q3 2023.
Security in Multi-Party Calls: Encryption Isn’t Optional
Encryption in a two-person call is straightforward. But in a group call with 50 people? Each person needs a unique key. If one key is compromised, the whole call is at risk. That’s why Wire’s SFT (Secure Federation Technology) stands out. It doesn’t just encrypt the connection-it encrypts each stream individually, with keys managed so that even the server can’t listen in. And it scales to hundreds of participants without breaking encryption.
Most platforms use server-side encryption, meaning the provider can technically access your call data. That’s fine for most businesses. But for legal, medical, or financial teams? End-to-end encryption isn’t a luxury-it’s a requirement. And SFU makes it possible. With MCU, you’d have to decrypt, mix, then re-encrypt everything on the server. That’s a security nightmare. SFU skips that step entirely.
What Happens Behind the Scenes When You Hit "Join Conference"
When you click "Join," your device connects to a media server via WebRTC. The server checks your token, verifies your permissions, and assigns you a stream ID. Then it starts routing:
- Your mic and camera send one stream to the server.
- The server identifies who’s speaking (using voice activity detection).
- It sends your stream to participants who are likely to want to hear you.
- If someone else speaks, the server switches your video feed to show them.
- Audio levels are adjusted automatically to prevent feedback or echo.
- Firewalls? Handled by STUN/TURN servers that find the best path through your network.
That’s why you need a decent internet connection-even if you’re not sharing video. The server still has to send you multiple audio streams. If your upload speed is below 1 Mbps, you’ll start dropping out or causing lag for others.
What’s Next? AI, 5G, and Adaptive Streams
The next wave isn’t about more participants-it’s about smarter ones. AI is now being used to detect who’s speaking, mute background noise, highlight the active speaker, and even translate speech in real time-all without adding latency.
5G networks will help. With sub-100ms latency predicted by 2025, mobile devices will handle more video streams without dropping. But the real innovation? Adaptive stream quality. Instead of forcing everyone to 720p or 1080p, systems will now dynamically adjust based on your device, connection, and role in the call. If you’re a silent observer, you get 240p. If you’re presenting? You get full HD.
And the future? Hybrid architectures. Some systems will use SFU for routing, but switch to MCU-style mixing for specific views-like a gallery view where everyone’s face is shown in a grid. The server doesn’t mix everything. It only mixes the final output for that one view. Everyone else still gets individual streams.
Choosing the Right System for Your Needs
So what should you use?
- Small teams, frequent calls, mobile users? Zoom or Google Meet. SFU handles it. Easy to use. Reliable.
- Healthcare, legal, or compliance-heavy work? Look for HIPAA/GDPR-compliant platforms with end-to-end encryption. Wire or specialized VoIP providers with SFT.
- Customer service teams? AWS Connect or Genesys. They’re built for agent control, not participant count.
- Large webinars, public events? Zoom’s large meeting mode or Microsoft Teams with SFU clusters.
- On-premise or private cloud? TrueConf or Jitsi (open-source SFU). You control the server, but you also manage the hardware.
Don’t just pick the platform with the highest participant number. Pick the one that matches your use case. A 1,000-person conference is useless if your team can’t hear the person who just asked a question.
Common Pitfalls and How to Avoid Them
- Too many video streams? Turn off video for non-presenters. Use audio-only mode.
- Echo or feedback? Use headphones. Mute when not speaking. Check echo cancellation settings.
- Call drops? Test your upload speed. Aim for at least 1.5 Mbps per person sharing video.
- Can’t add more people? Check your plan limits. RingCentral’s 10-person cap is real. Zoom’s free tier only allows 100 minutes.
- Security concerns? Don’t use public links. Use password protection and waiting rooms.
Most problems aren’t the architecture. They’re the setup. And the setup is only as good as the understanding behind it.
What’s the difference between MCU and SFU in conference calling?
MCU mixes all audio and video streams into one single stream on the server, which is sent to everyone. This uses a lot of server power and limits calls to 20-50 participants. SFU doesn’t mix streams. Instead, it routes individual streams from each participant to others, only sending the ones that are needed. This allows for hundreds of participants with less server load, but puts more strain on the user’s device to decode multiple streams.
Why does my conference call drop when more than 8 people join?
Many platforms, like RingCentral, claim a 10-participant limit, but real-world performance often drops after 7-8 due to session ID management issues, bandwidth constraints, or API limitations. The system may struggle to assign or track unique identifiers for each participant, especially on mobile networks. Switching to a platform built for scale, like Zoom or Google Meet, often solves this.
Can I have a secure conference call with 50+ people?
Yes, but only with platforms that support end-to-end encryption and SFU architecture. Wire’s SFT technology is one example-it maintains encryption for every participant without requiring the server to decrypt or re-encrypt streams. Most mainstream platforms like Zoom or Teams use server-side encryption, meaning the provider can access call data. For legal, medical, or financial use, choose platforms explicitly certified for compliance.
Why do some platforms limit conference calls to 6 participants?
Platforms like AWS Connect and older Genesys systems limit participants to 6 because they’re designed for customer service, not large meetings. Their focus is on agent control-muting, transferring, holding-rather than scale. Adding more participants increases complexity in managing roles and actions during a call. For healthcare or legal teams needing more, newer updates are expanding this limit, but legacy systems still restrict it.
Do I need high-speed internet for a conference call with 10 people?
For audio-only, 1 Mbps upload is enough. But if 5-10 people are sharing video, you need at least 1.5-2 Mbps upload per person sharing video. That means 10 people on video could require 15-20 Mbps upload speed. Most home internet plans have lower upload speeds than download-check yours. If you’re the host, your upload speed is what matters most to everyone else.
Is SFU better than MCU for most businesses?
Yes, for the vast majority. SFU scales better, costs less to run, and supports modern remote work needs. MCU is only better in rare cases where a fixed video layout is required-like a medical consultation where every participant’s face must be visible in a grid. But even then, hybrid systems are emerging that use SFU for routing and MCU-style mixing only for specific views, giving you the best of both.