The branch near your house (CDNs and object storage)

Netflix’s boxes sitting inside your internet provider’s building, filled at night and serving you by day. This lesson takes that picture and asks the plainer question behind it: when should any stall open branches, and what goes on their shelves?

Before you start. From The chai stall that got famous (load balancing) you need the famous stall that became many stalls, and from The slate next to the stove (caching) the slate: a small board of copied answers, whose hit rate is the share of asks it answers without a walk to the back. The Slate, Properly (caching to real depth) chalked a lifetime beside each line, the TTL (time to live): after that many seconds the line goes blank and the next ask fetches it fresh. One address, a thousand doors showed that one address can be answered from many cities, the nearest one answering. Object Storage and the CDN calls the one real copy of a file the origin, and it owns the mechanics this lesson only points at: cache keys, purging, signed links.

The idea in 30 seconds

A CDN is a network of servers near users that keeps copies of content, so most requests never reach the origin that made it.
The hit rate is the design number: at 95%, one request in twenty reaches the origin, so losing the CDN hands the origin twenty times its usual load.
A pull-through cache fetches an item the first time someone asks and serves every later ask; push stocks items ahead of demand, which pays only when you can predict what will be asked.
The TTL is chosen per item: long for things that never change, short for things that do, and it caps how often each copy asks the origin.
Object storage keeps whole files under names, cheaply, while the database keeps small facts; a pre-signed URL lets a user upload straight to object storage with a signed, expiring permission, so the file never passes through your servers.

Read the story · Skip to the numbers