| Point | Full | Incremental | Differential |
| What is copied | All data, every time | Changes since the last backup of any type | Changes since the last full backup |
| Backup time / storage | Longest / most | Shortest / least | Medium; grows each day until the next full |
| Restore needs | Only the last full | Last full + every incremental since, in order | Last full + only the latest differential |
| Restore speed | Fastest | Slowest | Medium |
Example with a Sunday full: a Wednesday crash needs Sun + Mon + Tue + Wed incrementals, or Sun full + Wed differential.
Recommended strategy for a bank (aim: RPO ≈ 0 data loss, RTO of minutes):
- Weekly full backup (e.g. Friday night) plus a daily differential — any restore needs only two sets, keeping recovery quick.
- Continuous transaction-log backups every few minutes, for point-in-time recovery up to just before the failure.
- Real-time replication to a standby database at a disaster recovery (DR) site in another region, for fail-over within minutes.
- 3-2-1 rule: 3 copies, on 2 different media, 1 off-site; include an immutable / offline copy against ransomware.
- Encrypt backups, control access, and test restores regularly (DR drills) to confirm RPO and RTO are met. Keep retention as Bangladesh Bank guidelines require.
— rhs