What happens when you open a web page

Your device

Everything that happens on your own computer or phone: the browser, and what it remembers from earlier visits.

The internet

The shared network in between: the directory that turns names into addresses, and the edge servers that sit close to you.

The provider's network

The website's own machines, usually in a data centre or a cloud region. Most requests never get this far, because the edge answers them first.

Browser

Where it starts and ends. You type example.com; the browser works out where that is, fetches the page and everything it needs, and draws it.

Steps

1, 2, 4 and the last

Speaks

HTTP, encrypted with TLS

Finally

Builds the page from HTML, CSS and JavaScript

Browser cache

Copies of files the browser has fetched before. If a copy is still fresh, nothing needs to cross the network at all - the fastest request is the one never made.

Step

1

Holds

Images, styles, scripts, and remembered DNS answers

Decided by

Cache headers the site sends

DNS resolver

The internet's directory assistance. Usually run by your internet provider or a public service, it finds the numeric address for a name and remembers the answer for a while.

Step

2

Answers with

An IP address

Usually takes

A few to a few tens of milliseconds

Name servers

When the resolver does not already know, it asks down a chain: a root server points to the .com servers, which point to the site's own authoritative name server, which gives the answer.

Step

3

Asked in order

Root, top-level domain, authoritative

Answer cached for

As long as the site's TTL allows

CDN edge server

One of many servers spread round the world. Your browser connects to the nearest, which answers straight away with any file it has cached and passes everything else on.

Steps

4, 5 and 10

Serves

Images, scripts and styles from its own cache

Also

Ends the encrypted connection, and filters attacks

Load balancer

Spreads requests across many identical servers, and stops sending any to one that has failed its health check - so one machine going down takes nothing down with it.

Step

6

Chooses by

Load, health, or simple rotation

Makes possible

Adding servers without downtime

Application servers

Where the site's own code runs: it reads who you are, gathers what the page needs from the cache and the database, and builds the response.

Steps

7, 8 and 9

Usually

Several identical copies

Hold no data

So any copy can answer any request

Background workers

Jobs that need not hold up the page - sending an email, resizing a photo - are put on a queue and done here, a moment later.

Step

9

Take jobs from

A queue

Keep

The page fast

In-memory cache

Recent answers kept in memory, which is far faster than asking the database. A hit here saves the slowest part of the trip.

Step

7

Reads in

Well under a millisecond

Forgets

Entries after a set time

Database

The permanent record: accounts, orders, posts. A primary takes every write, and replicas share the reading.

Step

8

Primary

Takes every write

Replicas

Share the reads, and stand by

Object storage

Large files kept apart from the database and handed to the CDN, so they are served from the edge rather than through the application.

Step

10

Holds

Images, video, downloads

Served through

The CDN