Netflix - the case study

The whole set on one board; the map and the seven problems are linked below.

Boards

Before you start. You need to know that a video file is enormous and a web page is tiny, and that somebody pays for every mile a file travels. That is all; the arithmetic in the story half is one subtraction and one percentage.

The idea in 30 seconds

Netflix’s scarce resource is egress, bytes going out, not storage and not computing power.
The answer is to do everything expensive while nobody is waiting, and to cache the finished film inside your internet provider’s building, one hop away.
That network of boxes is Open Connect, the boxes are OCAs, and prediction fills them overnight rather than demand filling them on the day.
Pressing play is three cheap parallel lookups plus your device’s own key exchange. The deciding half is the control plane, the carrying half is the data plane.
The player, not the server, picks the quality chunk by chunk against a buffer. That is adaptive bitrate, and it is why the picture sharpens a few seconds in.
Availability compounds downward, so 30 services at three nines is a 97% page. The fix is a circuit breaker and a fallback on every call, not better components.

Read the story · Skip to the numbers