The difference between Region and Availability Zones is not really clear. If a Region is a set of Data Centres and an AZ is ‘within’ a Region… does an AZ actually be a set of Data Centre? What’s the difference?!
I think of Availability Zones as infrastructure resources on seperate power, network and cooling; they can be areas of a single data center, or a seperate data centers connected by high speed networks. The main goal of an AZ is to eliminate a single point of failure, deploy in different AZs to provide availability in the event that one AZ fails. Microsoft may implement this differently, as it may be different data centers per region, however the end result is the same, deploy in different AZs to provide availability in the event that one AZ fails.
Another way of understanding this is that a Region provides "intra-geo" resiliency (storms, earthquakes, power-grid failure) and an AZ provides "intra-region" resilency (local flooding, electrical sub-station failure, human error, etc). The combination of Regions and AZ’s covers all failure modes.
As an side…why a minimum of three AZ’s and not two – this allows persistent storage services with high-durability to be setup with minimum latency. A quorum (tie-breaker/vote) is required to decide which is the most durable data – hence an odd number of AZ’s are required, three being the minimum.
The Spectrum of Azure High Availability Options
Azure high availability includes a spectrum of options, allowing progressively more confidence that an application will stay alive in the face of failure.
Below are four options for ensuring a Virtual Machine (VM) is highly available. As we explain below, Azure’s high availability mechanisms, specifically Availability Zones, extend beyond VMs to additional Azure services.
1. Single VM – running a Virtual Machine (VM) on Azure with no replication.
2. Availability Sets -running a VM with one or more replicated copies on separate hardware within the same Availability Zone, providing resiliency against machine failure.
3. Availability Zones – running a VM with one or more replicated copies on different Availability Zones, providing resiliency against data center failure.
4. Region Pairs – running a VM with one or more replicated copies on different Azure Regions (but always staying within the same geopolitical boundary, typically meaning the same country), protecting against natural disasters and large-scale outages.
where do you fit Region Pairs into the mix ?
As Serge says, having more that one region provides "Intra-geo" resiliency, that means that if an earthquake takes place in a region, then you have all your data replicated in another one. All the regions have one "Paired regions", it means that Microsoft has decided which is the best "back-up" region for each region probably taking into account the latency between them. Having a pre-stablished paired region also helps at the moment of prioritizing which region to recover in a disaster (It provides order) , always trying to have at least one one of the paired regions alive.