The 3-2-1 backup rule states: keep three copies of data, on two different media types, with one copy off-site. This is the baseline. What makes backups actually reliable is scheduled restore testing — verifying that the data can be recovered, not just that the backup job completed successfully.
The 3-2-1 rule explained
Three copies means: the production data, a local backup, and an off-site or cloud copy. Two media types means not relying on identical hardware that could fail for the same reason — for example, a local NAS and a cloud object store.
One off-site copy protects against site-level events: fire, flood, theft. Off-site can mean a physically separate location or a cloud storage service.
Backup frequency and retention
Backup frequency should match the acceptable data loss window — the RPO (Recovery Point Objective). If losing four hours of data is acceptable, a four-hour backup interval is the maximum.
Retention determines how far back you can go. A ransomware attack discovered three weeks after infection requires backups that pre-date the infection. A retention policy of seven days would leave no clean restore point.
- Daily full backups + incremental between fulls
- 30-day minimum retention for most production environments
- 90-day retention where compliance requires it
- Immutable backup storage to protect against ransomware deletion
The restore test: the only proof that backups work
Backup software reporting a successful job is not evidence that the data is recoverable. Backup files can be corrupt, incomplete or stored in a format that requires software that is no longer available.
A restore test should verify that a specific file, database or system image can be fully recovered in the target environment. For databases, this means attaching the restored database and running a consistency check.
Cloud backup: what to watch for
Cloud backup services are convenient but introduce two risks that are easy to overlook: egress cost (restoring large datasets from cloud storage can be expensive) and restore time (restoring a 10 TB dataset over a typical internet connection takes days).
For large datasets, a hybrid approach — local backup for fast restore, cloud backup for off-site protection — typically gives the best balance of cost, speed and resilience.
Backup monitoring
Backup jobs fail silently. A monitoring system that alerts on job failure — not just reports success — is essential. Alerts should go to someone who will act on them, not an inbox that is checked weekly.
