HLDintermediatePremium

Zoom HLD: Why Video Calls Use an SFU, Not a Mesh

A Zoom / video-conferencing system design: mesh vs SFU vs MCU topologies, an SFU that uploads once and fans out, simulcast layer selection, signaling vs media, NAT traversal, and scaling rooms.

By fiveyearsdevJune 14, 202614 min read

You're on a video call. Two people, fine. A third joins, a fourth — and someone's face freezes into a blocky smear while your laptop fan spins up like it's about to take off. That isn't your Wi-Fi having a bad moment; it's arithmetic. If every camera sends its picture straight to every other person, your upload has to carry one copy of your video per participant, and home broadband taps out around five. So "Design Zoom" is really one sentence: who sends what to whom, so a fiftieth person can join without melting the first? The answer real conferencing reaches for is the SFU — every participant…

What’s inside

Let's start nowhere near a computer

Where this exact shape shows up

Step 1 — Functional requirements (sentences first)

Step 2 — Non-functional requirements

Step 3 — The data model (and why so little is stored)

Step 4 — The topology decision

Step 5 — Simulcast: a layer per viewer

Step 6 — You don't forward 49 videos: active-speaker selection

Step 7 — Signaling, media, and getting through NATs

Step 8 — Trade-offs (each one keeping an NFR)

The complete implementation

Step 9 — Only now, the boxes

Step 10 — Scaling the design, one bottleneck at a time

Step 11 — When a piece fails: designing for failure

The interview corner

Where to go from here

Read this one free

HLD