WhatsApp - the case study

The whole WhatsApp architecture on one board: a server that carries billions of messages a day and is built so it can never read a single one of them. This lesson follows one message across that architecture, from a phone with no fixed address to a phone on the other side of the planet, and the constraints that shaped every choice along the way.

Boards

Before you start. You need to know that a phone app talking to the internet is really talking to a computer elsewhere called a server, and that anything you send has to physically travel there and back. Nothing else here needs any programming, database or maths background.

The idea in 30 seconds

WhatsApp’s design starts from one rule: the courier - the server - must never be able to read your messages, a promise its founder carried from a childhood spent assuming a phone line was tapped.
A phone has no fixed address, so instead of being called, it holds one line open to the server - a persistent connection - and the server pushes down it the instant something arrives.
A message waits in a mailbox until the recipient’s phone sends back an acknowledgment, and only then does the copy get deleted; a retried send is caught by its ID and ignored, not delivered twice for real.
Sending one message to a hundred people, fan-out, is cheap only because every member holds a personal way to unlock it, handed out once instead of relocked per person per message.
End-to-end encryption means only the two phones ever hold the key, and a fresh key locks every message so stealing one buys an attacker nothing else - forward secrecy.
Thirty-two engineers ran this for 450 million people because the software holding a connection open was built to cost almost nothing per connection.

Read the story · Skip to the numbers