RAID Is Not a Backup

A backup is a copy that does not change when the original changes. Everything else in this area follows from that one sentence, and RAID fails it by design — a mirrored array mirrors a deletion as faithfully as it mirrors a file.

A note on this guide: this rests on what redundancy is specified to do rather than on arrays failed here. See how we test.

The problem, as it appears

Someone builds a NAS, mirrors two drives, and the storage question feels resolved. There are two copies. A drive can die. What else is there?

Then a folder is deleted, or a file is overwritten by a bad save, or something encrypts the share — and the second copy has already agreed.

Why redundancy feels like safety

Because the word "copy" is doing two different jobs and nobody separates them.

RAID makes a copy that is identical at every instant. That is precisely what makes it useful for availability — the array can lose a member and carry on, because the survivor is current.

It is also precisely what makes it useless as a backup. Being always current means never having an older version of anything, and almost everything people actually lose is lost by their own action a moment before they wanted it back.

So it is not that RAID is a weak backup. It is a strong solution to a different problem.

What actually matters

Name the threats separately

The reason this gets muddled is that "losing data" is treated as one risk. It is at least five, and they are defended against differently.

• A drive dies. RAID handles this. So does any backup, more slowly.
• You delete or overwrite something. RAID does not help at all. Versioned backup does.
• Corruption or malware writes through the file system. RAID faithfully stores the damage.
• The whole machine dies — power supply, controller, fire, theft. RAID is in the box.
• The backup itself is bad. Only testing a restore finds this.

Written out, the shape is obvious: redundancy covers exactly one row.

The rule that settles every case

Does an event on the live copy immediately reach the other one?

If yes, it is redundancy or synchronisation. If no — if the second copy is a point in time that survives what happens next — it is a backup.

That is why a permanently connected external drive is weaker than people assume. It is a second copy, but it is reachable by anything that can reach the first, so it defends against a drive dying and nothing else. A drive that is connected only while it is being written to is a materially different product with the same specification.

Sync is not backup, but its version history might be

Cloud sync propagates changes, deletions included, and does so quickly. That is the design.

What rescues most people is not the sync but the version history sitting behind it, and that has a retention window. So the real questions are how far back it goes, whether it covers the folders you care about, and whether you have ever retrieved something from it.

Three copies, two kinds, one elsewhere

The old rule of thumb survives because it addresses the list above rather than any single item.

Three copies covers a failure during a restore. Two different kinds of storage covers a fault that affects one technology or one batch. One somewhere else covers the events that take out a room — fire, theft, flood, and a power supply that fails destructively into everything connected to it.

That last one is why a UPS is adjacent to this topic rather than part of it: it reduces the odds of a bad shutdown corrupting what you have, and it is not a copy of anything.

An untested backup is a belief

The failure that is genuinely embarrassing is the one where the arrangement was correct and nobody had ever restored from it — wrong folders included, the job silently failing for months, an archive that needs a password nobody wrote down.

Restore something small, on purpose, twice a year. It takes ten minutes and it converts an assumption into a fact.

The options, and who each suits

This page contains affiliate links. If you buy through them, Ugly Fixes may earn a commission at no extra cost to you. Products are selected independently.

A drive that is only connected while being written to suits nearly everyone, and it is the cheapest genuine backup there is. Its strength is not the medium — it is being unreachable the rest of the time.

A NAS suits a household with several machines, and it is worth having for availability. Treat it as the thing being backed up rather than as the backup.

A fast external SSD suits someone whose backup is skipped because it is slow, which is the real reason most backups stop happening.

A UPS suits protecting the machine from the shutdown that corrupts things in the first place. It belongs alongside a backup, never instead of one.

If you would rather not evaluate that yourself, the APC Back-UPS Pro BX1500M is the one we would point at for a machine holding data: it is line-interactive with automatic voltage regulation, so a sag is corrected without touching the battery, and it transfers in a typical 8 ms. Note the waveform is a stepped approximation to a sine wave rather than a true sine — fine for most PSUs, worth checking against yours if it is an older active-PFC unit.

The mistake to avoid

Letting redundancy occupy the place in your head where the backup should be. This is the whole failure, and it happens to careful people — the array is real, the second copy is real, and the sense of having addressed it is entirely reasonable right up to the moment something needs an older version.

The second mistake is having one backup and never opening it. The moment you find out whether a backup works should not also be the moment you need it, which is the same argument as a drive that has started clicking: everything gets easier if the decision was made months before the symptom.