When normally (without using the name: option) creating a GenServer, the only way to refer to it, is by its Process ID (PID).
In most simple cases, this is good enough. However, in some cases it is annoying that there isn’t a different alias you can refer to, which means you’ll end up passing a list of PIDs around all over your application.
So instead, it is possible to register a process under a name, which is then stored globally (for the current node!) in Erlang. This means that one can refer to the process by using that name (as well as by its PID). This is often used when you only ever need a single process doing a certain kind. It is very common to have constructs like name: __MODULE__, which means that you only want a single GenServer to run using the code in the current module.
There are a few ways to specify a name other than an atom. @msambarino wrote a great explanation here that goes into more detail.
EDIT: Heh, @hubertlepicki was just a second quicker than I. Great blog post!
