At your scale you should have hourly snapshots at least and a replicated replay log for your DB that is nearline but fairly separated by a firewall or even separate building.