System Design: Netflix (High Level)

 

1️⃣ What is Netflix solving?

Netflix is a global video streaming platform that:

• Streams video to millions of users

• Provides personalized recommendations

• Works at global scale

Key challenge: High-throughput video delivery with low latency

---

2️⃣ Core Requirements (Interview Language)

Functional

• Browse movies & shows

• Stream video

• Resume playback

• Personalized recommendations

Non-Functional

• Extremely high availability

• Global low latency

• Scalability

• Fault tolerance

---

3️⃣ High-Level Architecture

Clients

• Mobile apps

• Smart TVs

• Web browsers

Backend Services

• API Gateway

• User Service

• Catalog Service

• Recommendation Service

• Playback Service

Video Delivery

• CDN (Content Delivery Network)

---

4️⃣ Key Design Concepts (Very Important)

🔹 Video Storage & Streaming

• Videos stored in cloud storage

• Delivered via CDN (Netflix Open Connect)

• Users stream from nearest edge server

🔹 Why CDN?

• Avoids sending video from central servers

• Reduces latency

• Handles massive traffic

🔹 Metadata vs Video

• Metadata → APIs (titles, thumbnails)

• Video content → CDN (separate path)

🔹 Recommendations

• Based on:

  • Watch history

  • User behavior

  • ML models

• Offline batch + online serving

---

5️⃣ Data Storage (High Level)

• Cassandra / DynamoDB → user data

• S3 / Blob Storage → video files

• Kafka → user events

• ElasticSearch → search

---

6️⃣ Typical Interview Flow (How You Say It)

“Netflix separates metadata from video streaming.
APIs serve catalog data, while videos are streamed via a CDN from edge servers.
This allows Netflix to scale globally with low latency and high availability.”

---

7️⃣ Common Interview Follow-ups

• Why not stream directly from backend? → Bandwidth & latency

• How to resume video? → Playback position stored

• How recommendations scale? → Offline ML pipelines