Designing a Video Streaming Platform

Video streaming requires solving unique challenges in storage, encoding, delivery, and adaptability.

Requirements

• Upload and process video content
• Stream to millions of concurrent viewers
• Adaptive quality based on network speed
• Low latency start time (<2s)

Video Processing Pipeline

Upload → S3 (raw)
    ↓
Transcoding Service (FFmpeg)
    ↓
Multiple resolutions: 360p, 720p, 1080p, 4K
    ↓
HLS segments (10s chunks)
    ↓
S3 (processed) → CDN

Adaptive Bitrate Streaming (ABR)

Client dynamically switches quality based on bandwidth.

Fast network   → 1080p chunks
Slower network → 720p chunks
Slow network   → 360p chunks

Formats: HLS (Apple), DASH (standard)

CDN Strategy

• Store popular content at edge nodes globally
• 95%+ of requests served from CDN, not origin
• Use Netflix-style Open Connect Appliances for ISP peering

Storage

Metadata: PostgreSQL
Video files: S3
Thumbnails: S3 + CDN
View history: Cassandra
Recommendations: ML pipeline

Bandwidth Estimation

1080p = 4 Mbps
10M concurrent users
= 40 Tbps total bandwidth

Conclusion

Video platforms are CDN-first architectures. Transcoding pipelines, ABR, and CDN edge caching are the core pillars.