Design Netflix's Global Video Streaming & Zuul API Architecture

1. Problem Statement & Scope Clarification

System Mission

Design the global video streaming and API gateway infrastructure for Netflix, serving $250\text{M}+$ paid subscribers across $190+$ countries. The platform must handle continuous high-throughput video delivery ($250+\text{ Tbps}$ peak egress), millisecond-tier API routing across multi-region AWS infrastructure, automated regional evacuation during AWS catastrophic outages, and resilient distributed caching.

Functional Requirements

1. User Profile & Content Discovery API: Serve personalized movie recommendations, playback state bookmarks, and search queries with $P99 < 20\text{ ms}$.
2. Dynamic Ingress API Gateway (Zuul 2): Route, filter, rate-limit, and decrypt client requests at the network edge with async non-blocking I/O.
3. Multi-Region Active-Active Traffic Evacuation: Automatically evacuate $100\%$ of user traffic from an impaired AWS region (e.g., us-east-1 to us-west-2 or eu-west-1) within $< 7\text{ minutes}$ without user disruption.
4. Global Video Delivery via Open Connect (CDN): Deliver chunked adaptive bitrate video streams (HLS / DASH) from custom Open Connect Appliances (OCAs) deployed directly inside ISP networks.

Non-Functional Requirements (SLAs & SLOs)

• High Availability: $99.999\%$ uptime for video playback initialization.
• Ultra-Low Edge Latency: $P50 < 5\text{ ms}$, $P99 < 25\text{ ms}$ for API routing.
• Zero Single Point of Failure (SPOF): Fully distributed microservice mesh with circuit breakers (Hystrix / Resilience4j).

2. Capacity & Scale Estimation (Back-of-the-Envelope Math)

Traffic Scale

• Global Paid Subscribers: $250,000,000$.
• Peak Concurrent Video Streams: $15,000,000\text{ streams}$.
• Average Stream Bitrate: $4.5\text{ Mbps}$ ($1080\text{p}$ / $4\text{K}$ blend).
• Peak CDN Video Egress Bandwidth: $\text{Peak Bandwidth} = 15,000,000 \times 4.5\text{ Mbps} = 67,500,000\text{ Mbps} = \mathbf{67.5\text{ Tbps}}$

Control Plane API Throughput

• API Requests per Active Session: $\approx 2\text{ QPS}$ (playback heartbeats, discovery telemetry, telemetry logs).
• Peak API Gateway Volume: $\text{Peak API QPS} = 15,000,000 \times 2 = \mathbf{30,000,000\text{ QPS (Global)}} \approx \mathbf{10,000,000\text{ QPS per Region}}$

Distributed Cache Memory Footprint (EVCache / Memcached)

• Active User Session Context: $250\text{M users} \times 4\text{ KB state} = \mathbf{1\text{ TB RAM}}$.
• Personalized Recommendation Graph Cache: $50\text{M active profiles} \times 50\text{ KB} = \mathbf{2.5\text{ TB RAM}}$.
• Total In-Memory Cache Provisioning: $\approx \mathbf{10\text{ TB RAM}}$ per region across multi-AZ EVCache clusters.

3. High-Level Architecture & Component Mapping

Interactive Architecture Diagram
100%Fit100%
Synthesizing vector architecture diagram...