Pokémon GO planned for 5x. It got 50x.

transcript

Pokémon GO planned for a worst case of five times its target. Launch brought fifty. Ten times the worst case.

Niantic and Google picked a traffic target. They called it 1x. Then planned for about 5x. Plenty of headroom, right? What came? 50x. Next to it, the worst case barely shows.

The game opens first in Australia and New Zealand. 15 minutes in, traffic is far past what Niantic expected. It quickly surges to 50 times the target. And the US launch is due the next day.

Google's own post admits it. Not everything was smooth sailing. The game got unstable. Two teams fixed one problem after another.

Players hit a load balancer, then containers on Google's cloud. Japan is next. They need over a thousand more machines. To fit them, they upgrade the cluster. While millions of players join. Google compared it to swapping a plane's engine in flight. They swap the load balancer too. For a newer one, built for HTTPS.

Two weeks after the US, Japan opens. New sign-ups: three times the US launch. And Google reports it launched without incident.

An estimate is a guess. A hit can beat your worst case tenfold. What saved them wasn't a better guess. It was a platform that could grow. And a team that changed it live.

Full breakdown in the lecture.