Regions, AZs & the Backbone

Opens the Cloud & VPC arc. Before anything can be called “the cloud,” it has to sit in a real building, on real land, and that placement decides almost everything that follows.

Before you start. You don’t need any cloud experience for this one - every idea here maps onto a building, a city, and the roads between them. If you’ve read The Two Planes - why AWS feels like 100 random things, you already have the distinction this lesson leans on hardest: the machine doing the work and the machine deciding what work to do fail on different schedules, and on different days.

The idea in 30 seconds

An availability zone is a building with its own power, cooling and network - the biggest thing that’s allowed to die alone.
A region is several AZs in one metro area, 10–100 km apart: close enough for a live second copy, far enough that one disaster can’t reach both.
A shared control plane is what let one AZ’s storm reach its neighbours in 2011 - the buildings were fine, but the dispatcher everyone shared wasn’t.
Distance sets a hard latency floor, about one millisecond per 100 km of fibre, round trip, and nothing you build recovers it.
Edge locations shorten only the public-internet leg of a trip; the rest rides the provider’s own backbone between its buildings.
Multi-region is a much bigger step than multi-AZ, because the link between regions is too slow to keep two live copies in perfect sync.

Read the story · Skip to the numbers