Regions, region pairs, availability zones and datacentres

How Azure divides the world into geographies, regions, availability zones and datacentres, which failure each level isolates, and why a paired region and an availability set are answers to two completely different questions.

Lesson 1 of 11 in objective architecture-and-services. Azure architecture and services, part of Microsoft Certified: Azure Fundamentals (AZ-900).

The two placement options candidates confuse, and the failure each one actually covers. Availability set — What it separates: Fault domains and update domains; Where those sit: Inside one datacentre; The failure it survives: A rack, a host, a maintenance reboot; The failure it does not: The loss of the datacentre. Availability zone — What it separates: Whole datacentres; Where those sit: Spread across the region; The failure it survives: The loss of a datacentre; The failure it does not: The loss of the whole region Availability set Availability zone What it separates Fault domains and update domains Whole datacentres Where those sit Inside one datacentre Spread across the region The failure it survives A rack, a host, a maintenance reboot The loss of a datacentre The failure it does not The loss of the datacentre The loss of the whole region
The two placement options candidates confuse, and the failure each one actually covers.

Geography, region, zone, datacentre

A geography is a market with its own data-residency and compliance boundary — the sort of line a regulator draws — and it contains at least two regions. A region is a group of datacentres close enough together to be treated as one place: Azure picks the boundary so that traffic between them stays inside a tight latency budget, and joins them with a network of its own. When you create almost anything in Azure, the region is the first question you are asked, and it is a decision that is awkward to reverse.

Inside a region, an availability zone is one or more datacentres with independent power, cooling and networking, far enough apart that the same flood or the same substation failure is unlikely to reach two of them. A region that offers zones offers at least three, which is what makes a quorum-based design possible without leaving the region at all. Two spellings are worth keeping apart: a ZONAL resource is pinned to one named zone, and a ZONE-REDUNDANT one is spread across several by the platform without you placing it.

Below that, and easy to miss, is the availability set — which is not a geography level but a placement instruction inside a single datacentre. Azure spreads a set's members across fault domains, which are groups sharing a power supply and a top-of-rack switch, and across update domains, which are rebooted in turn during planned host maintenance. That covers a broken rack and a maintenance window. It does not cover the building, which is exactly the distinction the exam is built on.

The four levels of Azure geography, from the widest down to the building. Geography contains Region (datacentres near enough to act as one), Availability zone (own power, cooling, network), Datacentre (one or more per zone). Geography Region datacentres near enough to act as one Availability zone own power, cooling, network Datacentre one or more per zone
The four levels of Azure geography, from the widest down to the building.

Region pairs: distance, and updates one side at a time

Every region is paired with another by Microsoft, and you do not choose the partner — you choose only whether to use an option that replicates to it. The pair buys two things. The first is distance: the two are several hundred miles apart wherever the geography allows, so a single flood, fire or grid failure is unlikely to reach both. The second is staged maintenance: planned platform updates go to one region of a pair at a time, so a faulty rollout cannot land on your primary and your recovery target in the same window.

A pair normally sits inside the same geography, and that is not trivia — it is what keeps a data-residency commitment intact after a failover. Brazil South is the exception everybody is taught, because its partner lies in another geography. The other property worth remembering is that cross-region replication is asynchronous, because the distance that makes the pair useful is the same distance that makes a synchronous write impossible.

Sovereign regions are a different cloud, not a special label

Azure Government in the United States and Azure operated by 21Vianet in China are separate instances of Azure with their own regions, their own portals, their own service endpoints and their own operations staff. A subscription in the commercial cloud cannot simply reach across into one of them. When a requirement says physically and logically isolated from commercial tenants, or names a government accreditation, the answer is a sovereign cloud — not a region, not a zone, and certainly not a resource group.

Region choice is a design input, not a formality

Regions are not interchangeable. Availability zones, individual services, individual virtual machine sizes and prices all vary between them, which is why "we will pick the region later" produces a design that cannot be built. The four inputs to weigh are data residency and sovereignty, latency to the people and systems that will use it, whether every service and size the design needs is actually offered there, and cost.

The consequence worth internalising: when a region turns out not to offer a capability the design assumed, there is no switch to flip. Zones in particular are a property of the region, so the only honest options are to deploy somewhere that has them or to knowingly drop the requirement — and a question offering to enable zones in a region that lacks them is offering something that does not exist.

Worth carrying in

Geography
A residency and compliance market holding two or more regions.
Region
A set of datacentres treated as one location, joined by a low-latency network.
Region pair
The partner region Microsoft fixed for this one. Distance, plus staged platform updates.
Availability zone
Physically separate datacentres inside a region, with independent power, cooling and network.
Zonal
Pinned to one named zone. You chose which.
Zone-redundant
Spread across zones by the platform, without you placing it.
Availability set
Placement across fault and update domains INSIDE one datacentre.
Sovereign cloud
A separate instance of Azure — Azure Government, or Azure operated by 21Vianet.

What the exam does with this

Objective
architecture-and-services. Azure architecture and services
Share of the exam
38.46% (the whole objective)
Questions in this lesson
4
Signed for by a person
0

Partly checked. None of the 4 questions here has been read against the cited source by a person. 4 questions have been checked against their cited clause by an automated pass — which is not the same thing, and is not a signature.

Only questions a person has signed for are used in mock exams here. That is the whole difference between the two kinds of checking above.

Questions in this lesson

Drill this lesson

A lesson is one sitting: the trainer draws a short run from these questions alone and spaces the ones you get wrong.

Practise Regions, region pairs, availability zones and datacentres

The rest of objective architecture-and-services