System Design: Image CDN Architecture

Designing an image CDN that can handle millions of requests requires careful consideration of caching, storage, and transformation strategies. Let's walk through the architecture.

Requirements

Functional Requirements

• Upload images via API
• Transform images on-the-fly (resize, crop, format conversion)
• Serve images globally with low latency
• Support for multiple resolutions and formats

Non-Functional Requirements

• Handle 10M+ requests per day
• P99 latency under 100ms
• 99.9% availability
• Cost-effective storage

High-Level Architecture

The system consists of several key components:

1. **Upload Service**: Handles image uploads and storage
2. **Transform Service**: Processes image transformations
3. **CDN Layer**: Global edge caching
4. **Origin Storage**: Durable storage for original images

Storage Strategy

For storage, we use a tiered approach:

• **Hot storage**: Recently accessed images on SSDs
• **Warm storage**: Standard object storage (S3)
• **Cold storage**: Glacier for rarely accessed archives

Transformation Pipeline

Images are transformed on-demand using a URL-based API. The transform service parses the requested transformations, checks if a cached version exists, and if not, fetches the original and transforms it.

Caching Strategy

Multi-layer caching is essential:

• **Browser cache**: 1 year for immutable URLs
• **CDN edge cache**: 30 days
• **Transform cache**: 7 days for processed images

Scaling Considerations

To handle peak traffic:

• Horizontal scaling of transform workers
• Pre-warming popular images
• Graceful degradation for non-critical transforms

This architecture can scale to handle billions of image requests while maintaining low latency and reasonable costs.