Regions
The cloud is in real places: a region is a location you choose
A region is a geographic area where a provider runs data centres. Everything you create lives in the region you pick.
The cloud sounds like it is everywhere. It is not. It is in specific places, and each dot here is one of them. Amazon has thirty-eight of these regions, Google forty-two, and Microsoft more than sixty. The first one opened in northern Virginia in 2006, and it is still the largest. When you create anything, a server, a database, a file store, you say which region it goes in, and that is where it physically is. Each region is a complete, separate copy of the cloud. That one choice decides how fast your users are served, which laws apply to your data, and what you pay.
Availability zones
A region is three or more zones; a zone is one or more data centres
An availability zone is an isolated group of data centres. A region has several, a few milliseconds apart.
Zoom in on one region. A region is not one building. It is at least three zones, and each zone is one or more data centres with its own power supply, its own cooling and its own network connections. They are placed kilometres apart, so one fire, flood or power cut cannot reach two of them. But they are close enough, and joined by fast enough private fibre, that data can be copied from one zone to another in a millisecond or two. Far enough to fail separately, near enough to work together: that is the whole design.
Availability zones · Going deeper
Reading region and zone names on the three providers
us-east-1a on AWS, East US zone 1 on Azure and us-east1-b on Google Cloud all name one zone of a region in the eastern United States.
Why zones exist
Zones exist so that one failure cannot take everything
Run copies in two or more zones behind a load balancer and a failed data centre goes unnoticed.
Here is what zones are for. Users arrive at a load balancer, which hands each request to a server. The servers are spread over three zones. Now zone A loses power. Both of its servers are gone. The load balancer notices within seconds and sends everything to zones B and C. Users see nothing. You have lost a third of your capacity and none of your service. This only works if you asked for it. A system you placed in a single zone goes down with that zone, and the provider’s promise for it is much weaker.
Why zones exist · Going deeper
What the provider promises: one zone or two
An SLA states the uptime a provider commits to. Across zones it is 99.99%; for one machine in one zone on AWS, 99.5%.
Latency
Distance is delay: about 1 ms for every 100 km, there and back
Latency is the time a request takes to travel. Physics sets a floor, so serve users from a region near them.
Why not put everything in one cheap region? Because of distance. A signal in fibre covers about two hundred kilometres per millisecond, and every request has to go there and come back. A user in London talking to a region in London or Frankfurt waits a few milliseconds. To Virginia, Mumbai or São Paulo it is sixty to a hundred, before any server has done any work. To Sydney it is a hundred and seventy, on a perfect line that no real cable follows. A page that needs twenty round trips is now seconds slower. No amount of computing power fixes that. Only being closer does.
Choosing a region
Four questions pick a region: law, distance, services, price
Choose a region by data residency rules first, then closeness to users, then service availability, then price.
So how do you choose? Four questions, in order. First, law. Some data must stay in a particular country. If that applies, most of the map is ruled out before you start. Second, distance. Of the regions the law allows, which is closest to most of your users? Third, services. Providers launch new services in a few regions first. Check that what you need exists where you want to be. Fourth, price. The same machine costs different amounts in different regions, sometimes much more. Price comes last because it is the one you can bargain with: work that no user is waiting on can go wherever it is cheapest.
Choosing a region · Going deeper · 1 of 2
The same machine, five prices
On-demand prices vary by region. In 2025 an identical AWS machine costs about 60% more in São Paulo than in N. Virginia.
Choosing a region · Going deeper · 2 of 2
Data residency, sovereignty and the special regions
Where data sits decides whose law reaches it. Providers offer ordinary regions, and a few special ones for the strictest rules.
Where data sits decides whose law reaches it. Providers offer ordinary regions, and a few special ones for the strictest rules. Three words that are easy to mix up. Data residency Where the data is physically stored. You control it by choosing the region. Data sovereignty Which country’s law applies to the data. It usually follows where the data is stored, and sometimes also where the provider is based. Compliance The rules you must prove you follow, such as the EU’s GDPR for personal data or national rules for health and government data. The provider’s promise: data you put in a region stays in that region unless you copy or move it. Replication to a second region, a backup elsewhere, or a CDN are all things you switch on. Regions built for particular rules. Microsoft also groups regions into geographies, such as Europe or the United Kingdom, and promises that data stays within the geography even when it replicates between a pair of regions.
Three words that are easy to mix up
- Data residency Where the data is physically stored. You control it by choosing the region.
- Data sovereignty Which country’s law applies to the data. It usually follows where the data is stored, and sometimes also where the provider is based.
- Compliance The rules you must prove you follow, such as the EU’s GDPR for personal data or national rules for health and government data.
Regions built for particular rules
Microsoft also groups regions into geographies, such as Europe or the United Kingdom, and promises that data stays within the geography even when it replicates between a pair of regions.
Resource scope
Every resource has a scope: a zone, a region, or the world
A resource is zonal, regional or global, and its scope tells you what a failure will do to it.
Not everything you create sits at the same level. Some things live in one zone. A virtual machine is in one building, and so is the disk attached to it. If that zone fails, they are unavailable, which is why you run copies elsewhere. Some things are regional. Object storage keeps copies of every file in three zones without being asked, and a load balancer spans the zones of its region. A zone failure does not stop them. And a few are global. Your users and permissions, your domain names and your content delivery network exist once for the whole account, with no region to choose.
Resource scope · Going deeper
Scopes on the three providers
Machines are zonal and identity is global everywhere. Networks differ: regional on AWS and Azure, global on Google Cloud.
Edge locations
Hundreds of small sites bring content close to users
A CDN serves copies of your content from the edge site nearest each user, cutting the round trip to a few milliseconds.
There are only a few dozen regions, and users are everywhere. The edge closes the gap. Your files sit in one region, here northern Virginia. Around the world, the provider runs hundreds of small sites in major cities. The first time someone in Sydney asks for a file, the Sydney site fetches it from the region and keeps a copy. Everyone in Sydney after that is served locally, in about five milliseconds, not a hundred and seventy. This is a content delivery network. It suits anything that is the same for every user: images, video, scripts, downloads. Your servers and databases stay where they were.
Edge locations · Going deeper
Four ways to get closer, and what each provider calls them
CDNs and DNS form the edge network. Local zones and on-premises racks move real compute closer.
Data residency
Regions are isolated: nothing leaves one unless you send it
A second region is for disaster recovery or for distant users, and replicating to it is your decision.
One last property of regions matters more than it first appears. Regions do not share anything by default. A system you build in Frankfurt, and its data, stay in Frankfurt. That is what lets you promise a customer or a regulator where their data lives. If you want a second region, you build it. You switch on replication to copy the data across, and you decide whether a standby system waits there. People do this for two reasons: to survive the loss of an entire region, which is rare but has happened, and to serve users on another continent from nearby.
Data residency · Going deeper
Paired regions and multi-regions: the provider’s help with two regions
Building across regions is your job, and each provider offers ready-made pieces that make it easier.
Building across regions is your job, and each provider offers ready-made pieces that make it easier. Azure region pairs. Most Azure regions have a fixed partner in the same geography, at least about 300 miles (480 km) away where possible: East US with West US, North Europe (Ireland) with West Europe (Netherlands). Platform updates reach one region of a pair at a time, and if both fail, one is restored first. Geo-redundant storage copies your data to the paired region automatically. Google Cloud multi-regions. Some Google services accept a location wider than a region. A storage bucket created in the multi-region EU or US, or in a named dual-region, keeps copies in more than one region and stays readable if one is lost. AWS: you choose both ends. AWS has no fixed pairs. You pick any two regions and switch on a feature per service: cross-region replication for S3 storage, read replicas in another region for databases, global tables for DynamoDB. Exam shorthand: high availability means multiple zones. Disaster recovery means another region.
Azure region pairs
Most Azure regions have a fixed partner in the same geography, at least about 300 miles (480 km) away where possible: East US with West US, North Europe (Ireland) with West Europe (Netherlands). Platform updates reach one region of a pair at a time, and if both fail, one is restored first. Geo-redundant storage copies your data to the paired region automatically.
Google Cloud multi-regions
Some Google services accept a location wider than a region. A storage bucket created in the multi-region EU or US, or in a named dual-region, keeps copies in more than one region and stays readable if one is lost.
AWS: you choose both ends
AWS has no fixed pairs. You pick any two regions and switch on a feature per service: cross-region replication for S3 storage, read replicas in another region for databases, global tables for DynamoDB.
Lesson complete
You finished “Regions, zones and the edge”
Nicely done.