[TimTheTinker]

You can think of all user-facing domains in a system as an interface or API, which abstracts away implementation details about which specific servers are behind them.

This allows flexibility in infrastructure — you can swap machines in and out (e.g. by updating public DNS records, updating the machines’ IP addresses, or adding them to (or removing them from) a load balanced pool behind a reverse proxy). But you still need a way to reference individual machines regardless of whether they’re serving or not.

That’s where domains like 1e100.net come in — a system of concrete (non-abstract) references to specific machines in your infrastructure.