CDN Cache

The Layer That Changed the Internet

 without CDN:
   Sydney user ────── 200ms+ ──────► Virginia origin
   every asset, every user, every time

 with CDN:
   Sydney user ── 10ms ──► [Sydney edge] ─(miss only)─► origin
   edge holds copies; origin sees a fraction of traffic

 CDN = geographically distributed reverse-proxy cache.
 your own lesson covers mechanics; this is its caching role.

What Belongs at the Edge

Content	Edge behavior
Static assets (JS/CSS/images/fonts)	Cache months, immutable
Video segments	Cache aggressively; origin sees ~1 fetch per segment per region
Product/category pages	Short TTL (30–300s)
API GET responses	TTL seconds where staleness allows
Personalized/dynamic	Don’t cache — or fragment carefully

The Origin-Shielding Multiplier

 naive edge fill: 500 edges × miss = 500 origin fetches per object

 with an origin shield (middle-tier cache):
   edges ──► [1-2 shields] ──► origin
   first regional miss fills shield; other edges hit the shield
 
 origin load drops from O(edges) to O(objects) —
 this single setting choice often cuts origin traffic 10–50x

Control From Your Side

 CDNs mostly obey YOUR headers:

 Cache-Control: public, max-age=3600     edge caches 1h
 Cache-Control: private                  edge skips (per-user)
 Surrogate-Control: max-age=86400        edge-only override
                                         (browser gets shorter rule)

 plus platform tools:
 - PURGE by URL/tag      instant removal on deploy/scandal
 - cache keys            vary by host/path/query/cookie/header/geo
 - stale-while-revalidate at edge: zero-lag deploys

Cache-Key Discipline

 what's IN the key decides sharing vs fragmentation:

 /api/products?color=red&sort=price&_=1690123456
    ↑ all query params in key → cache-buster param = total misses

 trim keys deliberately:
   ignore tracking params (_, fbclid)
   normalize case/order
   fragment ONLY on real variance (currency, language)
 
 default keys are almost always wrong for APIs; configure them.

The Numbers That Sell It

 typical production outcomes:
 - static assets:   90–99% edge hit rate → origin nearly silent
 - latency:         p50 drops from ~150ms to ~15ms for far users
 - egress cost:     CDN bandwidth usually cheaper than origin cloud egress
 - outage survival: stale-if-error keeps site up during origin incidents

 caching at the edge isn't an optimization anymore;
 for global products it's table stakes architecture.

Interview Framing

“Serve images to global users” = CDN answer with structure: edge placement rationale (latency + offload), headers/surrogate control, cache-key normalization, purge strategy for bad uploads, shield mention as bonus. Connecting to browser-cache lessons (immutable versioned assets) shows layered-cache fluency.