Content Delivery Network

Push static assets to edge locations so users worldwide get millisecond load times.

IntermediateTheory11 min read

Read these first

Proxy vs Reverse Proxy

How a CDN works

CloudFront edge delivery

Video segments cached at PoPs; S3 origin fetch only on cache miss.

StreamHub production architecture (AWS)

End-to-end path from user to data — reference this when placing any new service.

CDN flow

User requests cdn.streamhub.com/segments/v_9182/720p/004.ts.
DNS resolves to the nearest edge PoP (point of presence).
Cache HIT — edge serves file from local SSD; origin never contacted.
Cache MISS — edge fetches from origin (S3 or origin shield), caches per TTL, serves user.

What to put on a CDN

Asset type	Cache strategy	StreamHub example
Video segments (.ts, .m4s)	Long TTL, immutable URLs	HLS/DASH chunks
Thumbnails / posters	Medium TTL, purge on update	creator avatars, video posters
JS / CSS bundles	Long TTL + content hash in filename	app.v3f2a1.js
API responses	Short TTL or no cache	Only public, read-heavy endpoints

Video segments (.ts, .m4s)
Cache strategyLong TTL, immutable URLs
StreamHub exampleHLS/DASH chunks
Thumbnails / posters
Cache strategyMedium TTL, purge on update
StreamHub examplecreator avatars, video posters
JS / CSS bundles
Cache strategyLong TTL + content hash in filename
StreamHub exampleapp.v3f2a1.js
API responses
Cache strategyShort TTL or no cache
StreamHub exampleOnly public, read-heavy endpoints

Never cache personalised or authenticated API responses at the CDN without careful Vary headers and cache-key design.

Cache control headers

Java

GET /static/app.v3f2a1.js HTTP/1.1
Host: cdn.streamhub.com

HTTP/1.1 200 OK
Content-Type: application/javascript
Cache-Control: public, max-age=31536000, immutable
ETag: "3f2a1b9c"

Java

GET /segments/v_9182/720p/004.ts HTTP/1.1

HTTP/1.1 200 OK
Cache-Control: public, max-age=86400
Content-Type: video/mp2t

Immutable tells browsers and CDNs the URL will never change content — safe to cache forever. Versioned filenames (app.v3f2a1.js) enable this pattern.

CDN vs origin architecture

StreamHub stores master files in S3. CloudFront (or equivalent) sits in front:

Java

origin:
  domain: streamhub-media-prod.s3.amazonaws.com
  origin_access_control: enabled   # S3 not public; CDN has OAC role

behaviors:
  - path_pattern: /segments/*
    ttl_min: 3600
    ttl_max: 86400
    compress: false                # video already compressed
  - path_pattern: /thumbs/*
    ttl_default: 3600
    compress: true

Aspect	Without CDN	With CDN
Latency (Tokyo user)	200+ ms to US origin	< 20 ms from APAC PoP
Origin load	Every view hits S3	~95% served from edge
Egress cost	High cross-region S3 egress	CDN bulk pricing
DDoS resilience	Origin exposed	CDN absorbs volumetric attack

Latency (Tokyo user)
Without CDN200+ ms to US origin
With CDN< 20 ms from APAC PoP
Origin load
Without CDNEvery view hits S3
With CDN~95% served from edge
Egress cost
Without CDNHigh cross-region S3 egress
With CDNCDN bulk pricing
DDoS resilience
Without CDNOrigin exposed
With CDNCDN absorbs volumetric attack

Cache invalidation

When a creator replaces a thumbnail, stale edge copies must go.

Invalidation strategies

Versioned URLs — thumb_v2.jpg instead of purging thumb.jpg (preferred).
API purge — POST /purge with path list (propagation takes minutes).
Short TTL — accept brief staleness for non-critical assets.
Surrogate keys — tag related objects; purge by tag (Fastly, Cloudflare).

Java

# CloudFront invalidation (expensive — use sparingly)
aws cloudfront create-invalidation \
  --distribution-id E1234567890 \
  --paths "/thumbs/u_42/*"

Dynamic content at the edge

Modern CDNs support edge compute (CloudFront Functions, Workers, Lambda@Edge) for:

A/B routing and geo redirects.
Token validation before serving premium video.
Request header normalisation.

Keep edge logic tiny — cold starts and debugging are harder than in central regions.

Measuring CDN effectiveness

Java

Metrics to track:
  - cache_hit_ratio (target > 90% for static/video)
  - origin_bandwidth (should drop after CDN enable)
  - p95 time_to_first_byte by geography
  - invalidation count (high = wrong URL strategy)

StreamHub's cache hit ratio on video segments exceeds 94% globally after tuning segment URL immutability.

Quick recall

Everything you need if you only revisit this box.

CDNs cache content at edge PoPs close to users, reducing latency and origin load.
Cache immutable assets with long TTL and versioned filenames; purge is a last resort.
StreamHub serves video segments and static assets from CDN; S3 remains origin.
Cache-Control headers drive CDN and browser caching behaviour.
CDNs reduce egress cost and absorb volumetric traffic at the edge.
Target > 90% cache hit ratio for static and video workloads.

Test yourself

Answer these before moving on — recall is what makes it stick.