Ah. I understand. For the projects I ever did on consultancy basis, I've setup a replica db for database stuff. And backups every 3-24 hours to s3 using glacier or infrequent access class depending on the nature of data. I keep a limited number of db snapshots usually. Some people want everything. For cases like that it's always glacier. I should learn how to store just the incremental changes but again, I've never worked on data more than a few gigs.

For standard data like images or files uploaded by users, I pray that s3 doesn't ever crap out on me and lose my data. Given AWS's guarantees, Ive always felt comfortable with that. Never worked on anything so critical that 1 file lost out of a hundred billion would be an issue.