You find out the real result of a backup job not on the night it runs but on the day you need to restore. Green check marks line up in the console; then a server goes down and it turns out the chain is broken, nobody has the encryption password or the backup was encrypted along with everything else by ransomware. The 3-2-1-1-0 rule is a simple control framework for ruling out these scenarios from the start: each number is a safeguard against a specific type of failure.
The five parts of the rule
| Number | Meaning | What does it protect against? |
|---|---|---|
| 3 | At least three copies of the data (live data + two backups) | A single backup becoming corrupted |
| 2 | Backups on at least two different media or systems | A failure in the same storage product or medium |
| 1 | One copy at a different location | Fire, flood, theft, loss of access to the building |
| 1 | One copy immutable or offline | Ransomware and a compromised admin account |
| 0 | Verified, error-free restore | An unnoticed corrupt or incomplete backup |
Put the table next to your current setup. The two rows most often missing are “immutable copy” and “0.” The first three numbers are met on paper in most places; the last two require discipline.
1. Set the targets first: RPO and RTO
Not every system is equally important. Even a few hours of data loss can be serious for an ERP database, while a daily backup may be enough for an archive file server. List your critical systems and write down the acceptable data loss (RPO) and acceptable downtime (RTO) for each. Backup frequency, retention and technology choices follow from this table.
The IT team should not set these numbers alone. The business unit answers “How many hours can the accounting system be down at most?”; IT then says what infrastructure makes that possible and what it costs. A target that isn’t written down becomes a point of argument in a disaster.
A detail that gets forgotten is dependency order. The application server won’t start without the domain controller and DNS. Calculate RTO not for a single machine but for the chain of systems the business process needs to run.
2. A different location and different media
The first backup is usually on local storage for fast restores. The second copy is sent to another building, a data center or the cloud. It is important that this copy cannot be deleted with the same admin account over the same network as the primary environment.
The “different media” requirement is not met by two separate folders on the same NAS. Same device, same disk pool, same firmware bug: a single failure takes both. Different media means a different device, a different type of storage or a different provider.
The second checkpoint for the off-site copy is bandwidth. If the amount of data changing each day is more than the line can transfer in the overnight window, the off-site copy falls a little further behind every day. A common method is to move the first full copy on a physical disk and then send only incremental data.
3. An immutable or offline copy
In ransomware attacks, attackers often go after the backups first. That is why at least one copy should be kept:
- on storage that cannot be deleted or changed (immutable) for the defined retention period,
- or on media physically disconnected from the network (an offline disk or tape).
The methods that can be used for an immutable copy are compared in our guide to immutable backup options.
The backup system’s admin accounts should also be kept separate and protected with multi-factor authentication.
Immutability isn’t achieved just by ticking a box in the backup software. Verify that the feature is really active on the storage side by trying to delete a backup before its retention period ends. If the deletion succeeds, there is no protection. If you use offline disks, put the rotation on a schedule; an “offline” disk that has been plugged in for weeks is an online disk.
4. “0”: restore testing
A backup job that completes successfully does not guarantee that the data can be restored. The test should include:
- Sample file-level restores
- Fully bringing up critical servers in an isolated environment
- Checking that the database and application open consistently
- Measuring the restore time and comparing it with the RTO target
- Recording the results with the date and the person responsible
An isolated environment is essential. Booting a domain controller or application server connected to the production network causes problems such as conflicting IPs, duplicate computer accounts and unexpected replication. Use a separate virtual switch or VLAN for testing.
The measured time is often longer than estimated. A VM that comes back in minutes from local disk can take hours from the off-site copy. It is better to see this difference during a test than during a disaster. An untested backup is an assumption.
5. Gaps we often see
- Backup failure notifications that nobody reads
- Newly added servers not included in the backup scope
- SaaS data such as Microsoft 365 not backed up because it is “already in the cloud”
- Encryption keys and the backup configuration kept in one place, without a backup
- Restore steps not written down
Notifications often go to the mailbox of someone who has left or a distribution list nobody looks at. Check the recipient list once a year. To keep new servers from falling outside the scope, build job definitions by folder, tag or container instead of machine by machine; a new VM created in the right place is automatically included.
Checklist
- Critical systems and RPO/RTO targets defined
- At least three copies, two different media
- One copy at a different location
- One copy immutable or offline
- Backup admin accounts separate and protected with MFA
- Restore tests planned and recorded
Frequently asked questions
What are RPO and RTO?
RPO (Recovery Point Objective) expresses the maximum acceptable data loss as time, for example the last four hours of data. RTO (Recovery Time Objective) states how quickly a system must be working again. Backup frequency and method are chosen to meet these targets.
Does a synced cloud folder count as a backup?
Not on its own. Sync also carries deleted or encrypted files to the other side. You need a separately managed backup with version history and a retention policy.
How often should restores be tested?
For critical systems they should be tested at regular, planned intervals, and repeated after any major change to the infrastructure. The frequency is set by the system’s importance and its RTO target.