Design to Scale
Content Delivery Network (CDN)
Cache content at edge locations close to users, so most requests never cross an ocean or touch your origin — and design pages so they can actually be cached.
What you will learn
- Explain why distance dominates latency for far-away users.
- Describe how edges cache, expire and purge content.
- Design cacheable responses with Cache-Control and stable URLs.
- Recognise when personalisation defeats the CDN.
Before you start
Overview
A publisher in Paris does not ship every book from Paris: it keeps stock in warehouses near its readers. Most orders leave from the nearest warehouse; only unusual titles come all the way from Paris.
Light in fibre needs ~70 ms to cross the Atlantic and back, more to Asia, and every request may need several round trips. A CDN places caching servers (edges) in hundreds of locations. A request goes to the nearest edge; on a hit it is answered locally, on a miss the edge fetches from your origin and keeps a copy for as long as the response's caching headers allow.
Run it
Users in Europe, the US and Asia browse product pages served from Frankfurt. Compare p50/p95 and Origin busy between A and B, purge an edge, then see C, where every page is personalised.
Making responses cacheable
| Header / technique | Effect |
|---|---|
Cache-Control: public, s-maxage=600 | Edges may keep it 10 minutes (browsers follow max-age). |
stale-while-revalidate=60 | Serve the old copy while refreshing in the background. |
Versioned asset URLs (app.3f9c.js) | Cache forever (immutable); a new deploy gets a new URL. |
Vary kept minimal | Every Vary value multiplies the copies the edge must store. |
Cache-Control: private, no-store | Personal data never cached at the edge. |
Example
@GetMapping("/products/{id}")
ResponseEntity<ProductPage> product(@PathVariable long id) {
return ResponseEntity.ok()
.cacheControl(CacheControl.maxAge(Duration.ofMinutes(1)).sMaxAge(Duration.ofMinutes(10)).cachePublic()
.staleWhileRevalidate(Duration.ofMinutes(1)))
.body(pages.render(id)); // same for everyone: no prices per user, no cart
}
@GetMapping("/me/cart-summary") // small, personal, never cached by the CDN
ResponseEntity<CartSummary> cart(Principal user) {
return ResponseEntity.ok().cacheControl(CacheControl.noStore()).body(carts.summary(user));
}Trade-offs
| Benefits | Costs | |
|---|---|---|
| Latency | Hits are answered a few milliseconds from the user. | Misses add a hop; uncacheable content gains nothing. |
| Origin load | The origin serves a small fraction of traffic and survives spikes. | A global purge or expiry can send a wave of misses to the origin. |
| Freshness | TTLs and purges control how long content lives. | Content can be stale up to the TTL; purges take time to propagate. |
Production considerations
- Track hit ratio per path and region; a drop usually means a header or URL change.
- Use origin shielding (a mid-tier cache) so a miss in many edges is one origin request.
- Split pages into a cacheable shell and small personal API calls.
- Rate-limit and protect the origin: the CDN also absorbs attacks, but only for cached paths.