Having a variable available from anywhere through a function has two advantages over a global. First it gains some protection from being modified anywhere you want (though not enough in this case because anyone can close the connection). Secondly it allows you to put off the decision over whether to initialize the variable as long as possible. If you need it, it will be there, but if you never needed it you didn't want to do the work. The third win is that if the variable has a complex initialization then every place it is used you document where to look for its initialization. That can be nice in a complex system.

The second is the real win here.

Connecting to a database takes a lot of work. For many useful programs the database connection takes more time than the entire program does! This is not the kind of operation that you want to possibly wind up doing repeatedly in a loop.

Furthermore this is an operation that is common to want to do from many places for many reasons. Which is why the appropriate life of a database connection is often longer than the life of your script! In fact that is one of the main wins that make people go to mod_perl, perlex, and friends. That you can have database connections that be reused across multiple web pages.

Incidentally the above point is why I think it is important to know why things are good design decisions. Zigster's advice as abstract advice is along a line that is generally recognized as the right thing to do. Were it practically any other example I would (probably silently) be in agreement with him.

But the cost of this advice is that wind up creating and destroying things a lot. Usually the overhead of creation and destruction is justified from reduced conflicts and easier to follow code. But in this case the excess creations can quickly turn into an..interesting..stress test for your database...