πŸ’¬
βœ‰οΈ
πŸ“Έ
▢️

Loading…

Design Netflix: adaptive streaming and CDN delivery

NetflixGuided Mode

🎬

260M subscribers. 15% of global internet traffic.

Play β†’ Entitlement β†’ CDN manifest β†’ ABR stream

The Netflix Scale Problem

Request Flow

1

260 million subscribers

Netflix serves 260 million subscribers across 190 countries. At peak β€” Friday evening in the US after a big series drop β€” it handles over 15% of all downstream internet traffic on the planet. That's not a metaphor. ISPs plan their capacity around Netflix's usage patterns.

2

The core tension

Video streaming has three competing constraints: quality (4K HDR requires 25 Mbps), availability (99.99% uptime = <53 min downtime/year), and scale (millions of concurrent streams). You cannot optimise all three naively β€” the architecture must be partitioned by responsibility.

3

Three distinct systems

We'll build Netflix as three interlocking systems: (1) the Playback Service + Adaptive Streaming Engine that gets video to the screen; (2) the upload and transcoding pipeline that prepares every title before it can stream; and (3) the Recommendation Service that drives 80% of what viewers watch.

Key Insight

Netflix isn't one system β€” it's three systems wired together: playback (low-latency per-session), upload pipeline (high-throughput async), and recommendations (ML-driven personalisation). Each scales independently and fails without taking the others down.

Overview