What to watch
1500 requests a second of video, and 240 of everything else. The video almost never reaches Netflix: the edge cache answers 96 percent of it from inside your ISP. Encoding is slow and that is fine. Jobs wait in a queue and nobody is watching a spinner, so seconds here cost nothing. The bottleneck is the licence service, not the video. It has 5 slots at 18ms, so it runs out at about 280 requests a second. Raise the load and watch it fill before anything else does. The breaker in front of it is what stops one slow service from holding every request open. Trip it and playback fails fast instead of hanging. Recommendations sit behind a bulkhead, so if they get slow the play path is untouched. Browsing breaking is survivable; playback breaking is not.What it is made of
Under load
Measured by running this design through the simulator at multiples of its own offered rate of 1742 requests a second. Twenty seconds of simulated time, one fixed seed, so the same numbers come out every time.This design holds 3484 requests a second with 0.3% errors. At 4355 it is losing 2% of them.
Open this example
Load Netflix: streaming at scale from the Examples menu and drag the traffic slider yourself.