# Content delivery networks

> Edge caches close to users: what they cache, how they invalidate, and what they protect.

Concept · Networking & APIs · https://levelcode.org/concepts/cdn

A content delivery network is a fleet of caching servers in hundreds of locations ("edges" or points of presence). Users connect to a nearby edge; the edge serves cached responses or fetches them from your servers (the **origin**) on a miss.

## What a CDN buys you

- **Latency:** a user in São Paulo is served from São Paulo, not Virginia — tens of milliseconds instead of a couple of hundred, and TLS handshakes complete nearby.
- **Origin offload:** a 95% cache hit rate means your servers see one request in twenty.
- **Protection:** CDNs absorb traffic spikes and volumetric DDoS attacks, and often provide a web application firewall and bot management.

## What to cache, and for how long

| Content | Caching |
|---|---|
| Versioned static assets (`app.3f9a2c.js`) | `Cache-Control: public, max-age=31536000, immutable` — a new version gets a new URL |
| Images and video | Long TTLs; often resized or transcoded at the edge |
| Public HTML or API responses | Short TTLs (seconds to minutes), plus `stale-while-revalidate` |
| Personalised responses | `Cache-Control: private` or `no-store` — never shared at the edge |

The cache key is the URL plus whatever the `Vary` header names. Vary on as little as possible: varying on `Cookie` effectively disables caching.

## Invalidation

- **Versioned URLs** are the most reliable: change the URL and the old object simply stops being requested.
- **Purge APIs** remove a path (or everything tagged with a surrogate key) from all edges, typically within seconds.
- **Short TTLs** bound staleness when purges are missed.

## Pitfalls

- Caching responses that set cookies or contain personal data — a classic way to show one user another user's account page.
- Forgetting that a 404 or 500 may be cached too; set short TTLs for errors.
- Assuming a purge is instant everywhere; design for a few seconds of overlap.

See [caching patterns](https://levelcode.org/concepts/caching-patterns) and [cache invalidation](https://levelcode.org/concepts/cache-invalidation).