An Availability Zone Means a Different Building
In one line
A region is a geographically separate area, and an availability zone (AZ) is a group of data centers within it that have independent power, cooling and network. The moment you tell these two words apart precisely, half of availability design is solved.
Why it was needed
If you open an architecture diagram that says "we made it redundant," it is common to find that the two servers are in the same rack. On-premises this is visible, but in the cloud it is not — when you launch two instances, they can land in the same place unless we specify where they go.
So the cloud exposes its isolation units by name. That is what an AZ is.
| Unit | What is independent | What is shared |
|---|---|---|
| Availability zone (AZ) | Power, cooling, physical network, building | Low-latency backbone within the region |
| Region | Geography, legal jurisdiction, usually the power grid | Nothing (fully separated) |
| Edge location | Cache point | The origin is in the region |
Latency between AZs is around 1–2 ms, so synchronous replication is practical. Between regions it is tens to hundreds of ms, so synchronous replication is practically impossible. This one number produces the practical rule "multi-AZ is the default, multi-region is a big decision."
Criteria for comparing regions
Names by vendor
| Concept | AWS | Azure | GCP |
|---|---|---|---|
| Region | Region | Region | Region |
| Availability zone | Availability Zone | Availability Zone | Zone |
| Group within a region | — | Region Pair (a separate concept) | — |
Azure has no AZs in some regions (it offers only availability sets). This means the decision to "go multi-AZ" can constrain the choice of region.
What do you choose a region by
- Law and regulation — if there is a requirement that personal data cannot leave the country, it ends here. In Korean public-sector and financial projects, this is the first condition you hit.
- Latency — the physically closer to users, the better. The Seoul and Tokyo regions differ by about 30–40 ms.
- Service availability — not every region offers every service. New services usually appear first in Virginia (us-east-1).
- Price — it differs by region. The same instance can differ by more than 20%.
AZ names differ per account
AWS's ap-northeast-2a points to a different physical AZ for each account. It is a device to spread load evenly, and because of it, even if you say "we're in 2a, and they're in 2a," it is not the same place. If you need to match physical locations across accounts, use the AZ ID (apne2-az1).
What it looks like in the field
- Launched as multi-AZ, but the service stopped when one AZ failed → the DB was in a single AZ.
- A whole-region outage → multi-AZ cannot prevent it. Region outages are rare, but they happen.
- Had to change the design because the service was not in that region → it should have been checked before choosing the region.
What to look at next
The shared responsibility model — the picture that determines how far our responsibility extends in this infrastructure.