The 3-2-1 idea has been around for decades and it survives because it is easy to remember and hard to argue with. It is a guideline for arranging copies, not a product or a compliance requirement.
Three copies
The live data counts as one. So a business with a server and one backup has two copies — which is better than most, and still one bad coincidence away from nothing. The third copy covers the case where a restore fails or a backup turns out to be incomplete.
Two types of storage
Copies on the same kind of device tend to fail the same way. A server and a NAS, or a NAS and cloud storage, are more independent than two volumes inside the same chassis. In practice this usually means: production storage plus backup storage of a different sort.
One copy off site
This is the part people postpone, and the part that matters when something physical happens. Fire, theft, water and electrical damage do not care how well the RAID is configured. Off site can be cloud storage, a backup at another branch, or a rotated disk kept somewhere safe.
Cloud is usually the least trouble, because it needs no one to remember anything. Rotated disks work perfectly well too, as long as somebody genuinely rotates them — in practice that needs a named person and a habit, not an intention.
- ProductionServer, NAS and applications in daily use
- Local backupSeparate storage on site for quick restores
- Off-site copyCloud or another location, for physical disasters
- Restore testProof that the copies can actually be used
How this looks in a real office
- A Windows Server or Proxmox host running ERP, Tally and file shares.
- A nightly backup to a NAS or a Proxmox Backup Server on site, kept for a few weeks.
- A daily or weekly copy of the important data to cloud storage.
- Alerts when a job fails, and a note of when the last restore was tested.
How long should copies be kept?
Retention is the other half of the design, and the half that is usually skipped. Daily copies kept for two to four weeks handle the common case: something broke last night. Monthly copies kept for a year handle the slower discoveries — a ledger that has been wrong since March, a folder nobody opened until the audit. If every copy is overwritten within a week, you are protected against hardware failure but not against mistakes.
Common mistakes
- Counting a RAID mirror as a second copy
- Three copies that all live in the same room
- An off-site copy nobody has verified in a year
- Backing up the files but not the server configuration, so a rebuild takes days
- No retention plan, so last month's clean copy has already been overwritten
Checklist
- Three copies, counting production
- At least two different storage types or platforms
- One copy off site or in the cloud
- Backup failures raise an alert someone reads
- A restore tested within the last few months
If your current setup misses two or three of these, that is normal — and fixable in an afternoon for most small offices.