* Always back up in _some way_. While a copy of the data on the same array won't protect you against all, or even very many, failure modes, it will protect you against some, and those are usually the most common. Remember that a _backup on the same server_ is still susceptable to some of the _same failures as the original data_ set, but having 2 copies is still better than 1.
* A good rule of thumb is _three copies_: the _original_ (RAID or otherwise); one _onsite copy_ on a different, preferrably offline, medium; and one _offsite copy_. Store the offsite copy in the cloud, a data vault, or at a friend's house; just keep it somewhere else. This is often called the "3-2-1" rule: 3 copies, 2 different media types, 1 offsite.
* _Make backups regularly_, at least once a week, preferrably more, and automate it! Forgetting to back something up and then needing just that backup is a common scenario and is never fun. The more frequently you back up, the better your recovery resolution, and back up regularly-changed files more often.
* _Test_ your backups regularly, at least once a month; _a backup is worthless if you can't restore from it_. Just because you have a backup doesn't mean you're protected: always test them and fix any problems. If you never test your backup, you will almost certainly find it doesn't work, right when you need it.
There are dozens of backup utilities out there which work well; I'm not going to prosthelytize for any one of them, but I personally use [BackupPC](http://backuppc.sourceforge.net/) and plain old `rsync` for my server and workstation backups.
Only you can determine what you need to back up, and that differs for everyone. But if you can't replace a set of data, you should definitely back it up: Murphy's Law applies here as much as anywhere.
