I'm sure many of our longtime users think we're a broken record preaching the importance of backups, but occasionally we do get calls from shops with horribly damaged or unreadable databases (usually by network events like server crashes, electrical storms, random acts of God) and no backup. It's probably the most painful call we ever have to handle at the helpdesk. Telling someone that their database is, in the technical jargon, hosed, hurts us, as well as the unfortunate client. It's an incredibly painful way to learn how critical it is to have a good recovery system.
There are many ways to ensure you're not one of the unfortunate few. It could be as simple as the first person in every morning burning a copy to CD or more complicated involving redundant media, offsites, firesafes, and safety deposit boxes.
Here's the heart of the system at C&P (yeah, it's not much to look at):
We start with a simple Windows XP machine running Retrospect. Every night, all applications which lock databases shut down and the entire backup is performed, copying from both OS X Servers as well as Windows servers. We backup to an Exabyte drive with rotating tapes (stored in a firesafe) as well as an external OWC firewire drive. There are two of these drives, and they trade places going offsite each week.
This entire system was not too expensive though there are cheaper ways to do it. UNIX-savvy folk can write cron jobs to automate their backup, including offsites in some situations. Whatever solution shops go with, make sure it includes the ability to recover from any individual day in the past week (two weeks is better) and that you send something offsite periodically in case a real disaster happens. Overwriting each night's backup and never taking anything offsite is only marginally better than no backup.
