Netflix loads your show onto a box near you the night before

transcript

The show you'll watch tonight is probably already near you. Netflix predicted it, and moved it there overnight. Before you asked.

A normal cache fetches a title the first time someone asks. 8 PM, Friday. Millions press play. The first viewer of every title misses. Millions of misses at once. It all lands on the links above. Exactly when they're busiest.

So Netflix predicts what each region will watch. Then it fills the boxes off-peak, while the links are idle. At 8 PM, your play is a local hit.

In this design, the top 20 percent sits on every regional server. So hot load spreads across all of them. The long tail? Hashed. One copy per region. Still local. Still off the origin.

A huge season drops at midnight. The day before, it goes onto every box. Midnight. The first viewer and the millionth both hit locally. The playbook a launch like Squid Game season 2 needs. 68 million views in four days, in December 2024.

Don't wait for the first miss. Predict it away. Move bytes before anyone asks.

Full breakdown in the lecture.