Global CDN Media Serving

A distributed CDN architecture for global media delivery.

Last modified on March 1, 2026

Problem Statement & Constraints

Develop a global media serving system that efficiently delivers static and dynamic assets worldwide using a content delivery network, while handling background processing for user uploads. The architecture must optimize for low latency, high availability, and cost-effectiveness, ensuring secure and reliable access to media content across diverse geographic regions.

Functional Requirements

Serve static and media assets globally.
Support background processing and transcoding for uploads.
Provide signed URLs and access control for private content.

Non-Functional Requirements

Scale: 1M requests/sec, global distribution; multi-region deployment.
Availability: 99.99% uptime with multi-CDN failover.
Consistency: Eventual consistency for media updates.
Latency: P99 < 100ms to edge; P99 < 500ms origin.
Workload Profile:
- Read:Write ratio: ~98:2
- Peak throughput: 1M requests/sec
- Retention: indefinite (hot); archive to cold storage after 1y

High-Level Architecture

graph LR Users --> CDN CDN -->|miss| Shield Shield -->|miss| Origin Origin --> Storage Uploader --> Transcoder Transcoder --> Storage Transcoder --> CDN

Users access media via a global CDN edge. Cache misses route through a regional Origin Shield to collapse concurrent requests before hitting the Origin storage. Concurrently, an upload pipeline transcodes raw files into optimized renditions and pre-warms the CDN edges.

Data Design

Object storage manages raw uploads, transcoded media, and access logs. The CDN layer defines cache key patterns and validation lifetimes, using surrogate keys for bulk invalidations.

Object Storage Layout (S3)

Bucket	Prefix / Path	Retention	Description
`raw-uploads`	`user_id/YYYY-MM-DD/`	30 days	Original untouched files.
`media-assets`	`asset_id/rendition/`	Indefinite	Post-transcoding optimal variants.
`static-logs`	`cdn/pop_id/HH_MM/`	90 days	Aggregated edge access logs.

Cache Key & Logic (CDN)

Item	Cache Key Pattern	TTL (Default)	Invalidation Tag
Images	`host/path?w=100&q=80`	30 days	`img:<asset_id>`
Videos	`host/path/playlist.m3u8`	1 year	`vid:<asset_id>`
Manifests	`host/config.json`	60 seconds	`config:global`

Deep Dive & Trade-offs

Deep Dive

Multi-tier caching: A regional Origin Shield (L2 cache) collapses concurrent edge misses, protecting the origin from viral traffic.
Invalidation strategy: Surrogate keys allow bulk purging of related variants, while short TTLs use stale-while-revalidate for metadata.
Security & Access: Time-limited signed URLs and rotatable keys provide zero-downtime secure delivery at the edge.
Content Optimization: Real-time resizing and Accept header negotiation deliver modern, optimal formats (WebP/AVIF).
Multi-CDN Failover: DNS-based anycast routing and health probes ensure automatic failover during provider degradation.

Trade-offs

Origin Shield vs. Direct: Shield reduces load significantly but adds a latency hop for cold-cache requests; direct is faster for misses but risks origin collapse.
On-the-fly vs. Pre-generation: Transforms save storage but increase edge CPU and latency; Pre-generation is faster to serve but increases storage costs.
Multi-CDN vs. Single Provider: Multi-CDN increases resilience but doubles configuration overhead and complicates cache invalidation sync.

Operational Excellence

SLIs / SLOs

SLO: 99.99% of media requests served successfully (2xx/3xx) from edge or origin.
SLO: P99 latency < 100 ms for cached content, < 500 ms for cache misses through origin shield.
SLIs: cache_hit_ratio, origin_request_rate, edge_latency_p99, upload_success_rate, transcoding_duration_p95.

Reliability & Resiliency

Synthetic: Global probes to measure edge latency and multi-region availability.
Failover: Regular multi-CDN failover drills to validate DNS switchover.
Load: Test upload-to-delivery pipeline at 10x normal traffic.

Service Resilience tag
Design patterns for reliable microservice behavior under load; implementing strict request idempotency, non-blocking async I/O, robust circuit breakers, durable background queues, and observability.
Model Serving & Inference tag
Principles for highly available production ML inference; utilizing immutable model registries, dynamic request micro-batching, safe canary rollouts, graceful fallback degradation, and efficient GPU memory management.
Distributed Web Crawler tag category
A highly resilient architectural design for a Google-scale web crawler; heavily focusing on breadth-first search (BFS), extensive DNS resolution caching, and polite handling of malicious domains.
URL Shortener & Pastebin tag category
A robust structural design for a highly available, extremely read-heavy service bridging short aliases to long URLs; implementing Base62 encoding, Snowflake IDs, and strict collision avoidance.