DEV Community

Jatin Gupta
Jatin Gupta

Posted on

Have you ever wondered how Zoom works?

#systemdesign #techtalks #coding #interview

If only two people are talking, does Zoom send the video?
And if 100 people join a meeting, how can everyone see each other without your laptop sending 99 separate videos?
Let's understand the complete architecture (Peer-to-Peer (P2P), WebRTC, and SFU) in a simple way.

Zoom architecture explained in detail

The Basic Idea

Whenever you join a Zoom meeting, there are actually multiple types of data traveling over the internet:

  • 🎤 Audio

  • 🎥 Video

  • 💬 Chat messages

  • 📄 Live transcript

  • 🖥️ Screen sharing

These are called Real-Time Media Streams because they need to reach everyone almost instantly.

You
   │
   ├── Audio
   ├── Video
   ├── Chat
   ├── Screen Share
   └── Transcript
        │
        ▼
      Zoom
        │
        ▼
Other Participants
Enter fullscreen mode Exit fullscreen mode

What happens in a 1-on-1 call?

Suppose:

You
Your friend
Enter fullscreen mode Exit fullscreen mode

Only two participants exist.
In this case, Zoom can establish a Peer-to-Peer (P2P) connection.

You  ------------------> Friend
        Direct Connection
Enter fullscreen mode Exit fullscreen mode

Your laptop sends one video stream directly to your friend's laptop.
Similarly,

Friend -------------> You
Enter fullscreen mode Exit fullscreen mode

One stream each.
No complicated routing is required.

How do two devices connect?
They usually use WebRTC (Web Real-Time Communication).

WebRTC allows browsers and applications to exchange:

  • Video
  • Audio
  • Data in real time. However, before sending media, both devices need to discover each other.

This process is called Signaling.

Example:

You
   │
Signaling
   │
STUN Server
   │
Signaling
   │
Friend
Enter fullscreen mode Exit fullscreen mode

After exchanging connection information, media can flow directly.

You  ==========> Friend
        Video

Friend ========> You
        Video
Enter fullscreen mode Exit fullscreen mode

What is a STUN Server?
A STUN Server helps devices discover their public IP address and determine how they can communicate through NAT (Network Address Translation).
Think of it like this:

You ask,

"What address does the outside world see for me?"
Enter fullscreen mode Exit fullscreen mode

The STUN server replies with that information so the two peers can attempt a direct connection.

Why P2P is great.

Advantages:

  • Very low latency
  • Less server cost
  • Fast communication
  • Direct transmission 
You ---> Friend
Enter fullscreen mode Exit fullscreen mode

Simple and efficient.

But what happens when more people join?
Imagine:

You
Person B
Person C
Enter fullscreen mode Exit fullscreen mode

Now you must send your video to both people.

      You
     /   \
    /     \
   B       C
Enter fullscreen mode Exit fullscreen mode

You now upload 2 streams.

Suppose there are 10 participants.

        You
      / / | \ \
     A B C D E F G H I
Enter fullscreen mode Exit fullscreen mode

Now your laptop must upload 9 different video streams.
That is a huge problem.

Why is this impossible?

Because your:

  • Internet upload bandwidth is limited
  • CPU is limited
  • Memory is limited
  • Battery is limited If each stream is 2 Mbps,

For 10 people:

2 Mbps × 9

= 18 Mbps upload
Enter fullscreen mode Exit fullscreen mode

Most users cannot continuously upload this much.
If 100 participants exist:

2 Mbps × 99

≈198 Mbps upload
Enter fullscreen mode Exit fullscreen mode

Practically impossible.

So how does Zoom solve this?
Instead of sending separate streams to everyone,
You upload only one stream.

You
   │
   │
   ▼
 Zoom Server
Enter fullscreen mode Exit fullscreen mode

The Zoom server then distributes that stream to everyone else.

             You
              │
              ▼
        Zoom Server
        /    |     \
       /     |      \
      B      C       D
Enter fullscreen mode Exit fullscreen mode

Your laptop uploads only once.
The server does the rest.

This server is called SFU
SFU stands for
Selective Forwarding Unit

It is a media server.
Its job is simple:

  • Receive your stream
  • Forward it to other participants It does not create new videos. It simply forwards them intelligently.

How SFU works
Step 1

You upload one stream.

You
   │
   ▼
 SFU
Enter fullscreen mode Exit fullscreen mode

Step 2

SFU receives it.

You
   │
   ▼
 SFU
Enter fullscreen mode Exit fullscreen mode

Step 3

SFU forwards it.

        SFU
      /  |  \
     /   |   \
    B    C    D
Enter fullscreen mode Exit fullscreen mode

Now each participant does the same.

        A
         \
          \
           SFU
         / |  \
        B  C   D
Enter fullscreen mode Exit fullscreen mode

Everyone uploads only one stream.
The SFU distributes everything.

Why is it called "Selective" Forwarding?
Because it forwards only what is needed.
Example:

If your screen shows only 9 participants,
The SFU may send only those visible videos in high quality.
If someone is hidden or inactive,
The server can reduce quality or stop forwarding temporarily.

This saves:

  • Bandwidth
  • CPU
  • Battery

Does SFU mix videos together?
No.
It simply forwards.

You ---> SFU ---> B

You ---> SFU ---> C

You ---> SFU ---> D
Enter fullscreen mode Exit fullscreen mode

No video mixing happens.

Then what is MCU?
Another architecture is MCU (Multipoint Control Unit).
MCU receives all streams,

A
 \
  \
B ---> MCU
  /
 /
C
Enter fullscreen mode Exit fullscreen mode

It decodes everything,
mixes videos,
creates one combined video,
then sends that back.

        MCU
          │
          ▼
     Combined Video
          │
          ▼
     All Participants
Enter fullscreen mode Exit fullscreen mode

SFU vs MCU

| SFU                              | MCU                              |
| -------------------------------- | -------------------------------- |
| Simply forwards streams          | Mixes streams                    |
| Faster                           | More processing                  |
| Lower latency                    | Higher latency                   |
| Less CPU on server               | Heavy server computation         |
| Scales well                      | More expensive                   |
| Used by many modern meeting apps | Used in some specialized systems |
Enter fullscreen mode Exit fullscreen mode

Why Zoom uses SFU
Because it allows:

  • Millions of users
  • Lower latency
  • Better scalability
  • Lower upload requirements
  • Efficient bandwidth usage Instead of every laptop sending many streams, 
❌ You → B
❌ You → C
❌ You → D
❌ You → E
Enter fullscreen mode Exit fullscreen mode

it becomes

✅ You
      │
      ▼
     SFU
   / / | \ \
  B C D E F
Enter fullscreen mode Exit fullscreen mode

Only one upload from your device.

Complete Flow

User

   │

Capture Camera

   │

Encode Video

   │

Upload One Stream

   │

──────────────

       SFU

──────────────

Receives Stream

Forwards Stream

──────────────

Participants

B

C

D

E

F
Enter fullscreen mode Exit fullscreen mode

Final Takeaway

  • 1-on-1 calls can often use Peer-to-Peer (P2P) with WebRTC, where devices communicate directly after signaling and STUN-based connection setup.
  • Group calls cannot rely on P2P because each participant would need to upload many separate streams, quickly exhausting bandwidth and device resources.
  • To solve this, Zoom uses an SFU (Selective Forwarding Unit). Each participant uploads one media stream to the SFU, and the SFU intelligently forwards that stream to the other participants.
  • This architecture keeps latency low, reduces upload bandwidth requirements, and allows Zoom to scale to meetings with many participants efficiently.

In one line:
"P2P works well for 1-to-1 calls, but for group meetings Zoom scales by using an SFU, where you upload one stream and the server forwards it to everyone else."

Top comments (0)