Availability Zone

Large internet services don’t run on computers in an office, but in data centers: halls full of servers that other companies can use for a fee. This rental arrangement is called the cloud. Providers like Amazon, Microsoft, or Google don’t operate their data centers as a single monolithic block. They divide them into separate locations called Availability Zones. Each such zone has its own power supply, its own cooling, and its own connection to the internet. If power fails in one zone, the others remain unaffected.

Why a single data center is a risk

Data centers fail. A digger severs a fiber-optic cable, a transformer catches fire, the air conditioning gives out on a hot day. Such events are rare, but over the years they reliably occur. Anyone who has all their servers in one location is offline at that moment.

This is exactly why Availability Zones exist. An online shop can distribute its servers across three zones. If one fails, the other two take over all the traffic. For customers, the outage is then at most noticeable as a brief delay. Cloud providers contractually promise their customers certain levels of availability, often 99.99 percent uptime per year. Such guarantees usually only apply if the customer actually distributes across multiple zones.

A common misconception: many believe the cloud is automatically fail-safe. That’s not true. The distribution has to be set up and paid for yourself. Anyone who, for cost reasons, puts everything into a single zone ends up with the same weak point as their own server room in the basement.

Zones, regions, and the distance between them

Availability Zones are grouped into larger units called regions. A region is a geographic area, such as Frankfurt, Ireland, or Northern Virginia. A region typically contains three or more zones. The zones within a region are far enough apart that a fire or a flood won’t hit several of them at once. Usually this is a few kilometers to a few dozen kilometers.

At the same time, they must not be too far apart. Data takes time to travel through a cable. At a distance of a few kilometers, this delay is under one millisecond. This is why servers in different zones of a region can work together as closely as if they were standing side by side. A database can mirror every change to a second zone instantly.

Between regions, this doesn’t work as easily. From Frankfurt to Virginia, a signal takes about 90 milliseconds round trip. This gives rise to a rough rule of thumb: multiple zones protect against technical failures, multiple regions protect against disasters and legal problems. Zones are everyday protection, regions are the emergency plan.

Availability Zones in cloud bills and in the news

Anyone who sets up an account with a cloud provider makes this decision right away. When starting a server, you have to select a region and a zone, often with labels like eu-central-1a. The a at the end stands for the zone. Prices also depend on this: some providers charge extra for data traffic between two zones.

In the news, Availability Zones usually come up when something goes wrong. When a major outage at a provider takes down thousands of websites, reports often state which zone or region was affected. Such reports also show how strongly entire industries depend on just a few locations.

For AI applications, this topic is playing an increasingly important role. The graphics chips on which AI models run are scarce and not available in every zone. Anyone who needs computing power for a large model therefore can’t always choose freely. Some companies book their capacity months in advance for a specific zone.