Consider a group of hosts attached to an Ethernet segment. A single IP address prefix would be assigned to the entire Ethernet, in this example 10.10.1.32/27. Each device attached to the Ethernet would have a unique IP address matching the prefix. The Ethernet, its address prefix, and its hosts would be collectively refered to as a subnet. An address such as 10.10.1.70, which doesn't match the Ethernet's prefix, would not be a valid IP address on this subnet.

The router on the left would maintain a table listing the Ethernet addresses for every device on this subnet, and therefore be capable of delivering packets directly to these hosts. Routers elsewhere in the network would only maintain a single entry for the entire subnet, indicating that traffic for 10.10.1.32/27 should be delivered to this router.

An Ethernet with a larger number of hosts would require a shorter prefix (matching more addresses), while an Ethernet with only a few hosts could be assigned a longer prefix.

As a more complex example, consider a group of closely interconnected Ethernets, each one a subnet. Then there might be a single address prefix for the entire group. The group's prefix would contain all the address prefixes for the individual subnets.