The short answer is to do whatever the customer does on other servers -- consistency is key now that a single person can be responsible for dozens of servers.
The long answer is a long, paranoid rant about having multiple levels of protection against a variety of catastrophes -- some imagined, most real, others unimaginable.
Here are my rules at a high level:
Wherever possible, a full, daily, offline backup to reliable, fault-tolerant local disk is preferred. It means restorations are quick and easy.
If daily offlines aren't possible... Daily onlines, transaction log archiving, plus weekly offlines during the maintenance window.
If daily onlines are troublesome... Daily incrementals/deltas, weekly onlines, transaction log archiving, and monthly offline backups.
The more complex the backup routine, the more complex and less tolerant of failure the restores are. Regular and repeated restores to a test box are absolutely critical. The best customers practice database restores quarterly to find flaws in their configuration and processes, and correct them before they turn a hiccup into a career-altering disaster.
I'd be interested to see what others have to say about this as well.
-JD.
