Designing Netflix: Putting the Movie in a Box Near the Viewer

Friday, 9 pm, Kochi

The Menon family in Kochi has waited a year for the new season of their favourite thriller. It drops at 9 pm on Friday on ChalChitra Stream. At 9:00:05, they press play, along with several million other households across the country. The first frame appears in under two seconds. Halfway through episode 2, the cache box at their internet provider fails, and they never notice. (The Menons, ChalChitra, Aditya and every number here are made up for teaching.)

Aditya designs ChalChitra's streaming. He explains it with a city that has one giant film warehouse on the far side of town. Every family drives there for a movie, and the roads jam every evening. Now imagine the warehouse puts a small shelf of films in each neighbourhood, and stocks it at night when the roads are empty. Families walk to the nearest shelf. A head office answers only two questions:

• "Am I allowed to watch this?"
• "Which shelf is nearest to me?"

That's the idea behind large video streaming. This guide designs an on-demand streaming service in the style of Netflix. It is an original interview design, not Netflix's private architecture. Where it describes how Netflix does something, the source is Netflix's own public Open Connect overview. Uploading user videos is covered in our YouTube guide, and live sport in our Hotstar guide.

In 30 seconds, as an architect would frame it: "Split it into a control plane in the cloud and a data plane of cache appliances inside or near ISPs. The control plane checks entitlement and licence, and steers each play to a ranked list of appliances that hold the needed files, using file availability, health, proximity and load. Appliances only report status and serve files, and they hold no member data. Instead of filling on demand, a fill planner uses popularity to stock appliances during off-peak windows, so release night streams from boxes that already have the files. Encodings are chosen per title. Players fail over across ranked URLs without asking the cloud again."

1. The rules and the pieces

Members pick a profile, browse a catalogue, press play, and expect the video to start quickly and keep going, on phones, TVs and browsers. The service remembers where each profile stopped. Titles have licences that differ by country and by date.

Recommendations, billing and studio delivery are separate systems. This guide covers the path from the play button to the screen.

The Appliance record is the unusual one. The control side must know which box holds which file, or it can't choose well.

2. "First frame in two seconds": two planes, kept apart

• The control plane makes decisions.
• The data plane moves the heavy bytes.

Netflix's overview describes exactly this split. Its cache appliances, called Open Connect Appliances (OCAs), do only two things: report their status to the control services in the cloud, and serve files over HTTP or HTTPS. They store no viewing history, no DRM information and no member data.

flowchart TD
    P@{ icon: "algomi:ext-device", label: "Player app", pos: "b", h: 52 }
    A@{ icon: "algomi:svc-cube", label: "Playback service", pos: "b", h: 52 }
    P --> A
    E@{ icon: "algomi:svc-cube", label: "Entitlement and licence check", pos: "b", h: 52 }
    A --> E
    S@{ icon: "algomi:svc-flow", label: "Steering service", pos: "b", h: 52 }
    A --> S
    K@{ icon: "algomi:data-bolt", label: "Cache inventory", pos: "b", h: 52 }
    S --> K
    B@{ icon: "algomi:data-bolt", label: "Cache appliances near viewers", pos: "b", h: 52 }
    B -->|report health and files| K
    P -->|fetch video files| B