As we stated previously, each datacenters provide no redundancy, so multiple datacenters will be connected together and made into an availability zone. The connections between datacenters are low-latency layer 2 connections over metro links. An availability zone is thus geographically distributed across multiple facilities that are still close enough to provide low latency across the zone:

The availability zone provides minimal support for redundancy. For instance, an EBS volume in an availability zone is redundantly placed on two different volume arrays in two different facilities. But even with features such as EBS volume redundancy, which traditionally might be considered as high availability, all the datacenters in an availability zone are connected in such a way that the whole availability zone is considered a one-fault isolation environment. Thus, if we want to achieve high availability within a region, we need to deploy our infrastructure to at least two availability zones.

The availability zones are also interconnected via low-latency network connections, although due to a wider geographical distribution of the availability zones themselves, we can expect to see a bit higher latency (usually within 5 ms) than within one availability zone (usually under 2 ms). This is a consideration that needs to be kept in mind when designing any application that is very sensitive to latency. However, as long as the traffic within a region does not traverse from the private network to the public network, it does not count toward our outgoing AWS transfer pricing:

There are services, such as AWS S3, that will be automatically distributed across at least three availability zones to provide regional high availability. There are others where we can simply specify the requirement for increased regional redundancy, such as AWS RDS, where setting up a multi-availability zone RDS database will deploy two instances in two availability zones that are synchronously replicated. Then, there are services such as AWS EC2, where we can select the availability zone when we launch instances. Is it then up to us to correctly deploy and configure services so that they are fully redundant within a region.