Netflix Content Delivery Architecture

Unlike companies relying on Akamai or Cloudfront, Netflix designs and deploys its own CDN hardware called Open Connect Appliances (OCAs). These are high-density servers with up to 1 petabyte of storage, placed physically inside ISP datacenters worldwide. Netflix offers them to ISPs for free — ISPs save on peering costs because Netflix traffic stays local, and Netflix pays zero egress fees to transit providers. The OCA runs a custom FreeBSD-based OS optimized for high-throughput file serving.

Every night during off-peak hours (2–4am local time), Netflix's cache-filling system pushes the next day's most-likely-watched content to nearby OCAs. The algorithm uses regional viewing patterns, day-of-week trends, new release schedules, and content popularity decay curves. Popular content is cached on thousands of OCAs globally; niche content lives on fewer. This pre-positioning means when you press play, the first byte arrives in under 50ms from a server potentially a few miles away.

Netflix's steering service uses consistent hashing to pick which OCA serves your request. A virtual ring maps 'content ID + client IP prefix' tuples to server clusters. The same video chunk always maps to the same OCA cluster, maximizing cache hit rates. When an OCA fails, consistent hashing minimizes remapping: only 1/N of keys move — compared to 100% remapping in a simple modulo approach.

Netflix encodes every title at 20+ quality levels — 235 Kbps mobile to 16 Mbps 4K HDR. The client player monitors bandwidth every 10 seconds and switches quality mid-stream. If your connection drops, you get a lower-quality segment within 2 chunks (~4 seconds) rather than buffering. The AV1 codec achieves 30% better compression than H.264 at the same quality, reducing bandwidth costs significantly.

Netflix invented Chaos Engineering with Chaos Monkey — a tool that randomly terminates production instances during business hours. The broader Simian Army includes Chaos Gorilla (terminates entire AWS AZs) and Latency Monkey (introduces artificial delays). This culture means Netflix engineers routinely prove their services handle failure — so when AWS has a real region outage, Netflix users see zero downtime.