Three copies, two kinds of storage, one somewhere else. The rule is simple and incomplete: copies only protect an archive if something checks, regularly, that they still match.
The 3-2-1 rule says: keep three copies of every file, on two different types of storage, with one copy somewhere else. It was popularised for photographers, and is generally traced to Peter Krogh's The DAM Book: Digital Asset Management for Photographers, first published in the 2000s. It has since become the standard advice of the storage industry because it is easy to remember and addresses the ways files are usually destroyed.
For an image archive it is the right place to start and the wrong place to stop. Three copies of a file protect you only if they are still three copies of the same file. Hard drives corrupt data quietly, copy operations fail part way through, and a damaged master synchronised to two backups gives you three damaged files. The missing piece is a way to check, and that is what checksums are for.
What does 3-2-1 mean for an image archive?
Each number answers a different kind of failure.
| Part of the rule | What it guards against | In practice |
|---|---|---|
| 3 copies | A single device failing | A working copy and two backups of every master |
| 2 types of storage | A fault common to one technology, maker or batch | For example, disk storage plus tape or cloud |
| 1 off site | Fire, flood, theft or a power surge at one address | A copy in another building, ideally another town |
Professional guidance says the same thing in more careful language. The Digital Preservation Coalition's Handbook, the standard UK reference, advises making multiple independent copies of digital material and storing them in different geographic locations, and using different types of storage technology to spread risk. It sets out graded levels, derived from the US National Digital Stewardship Alliance: two complete copies that are not in the same place as the minimum, then at least three complete copies with at least one in a different geographic location, rising to three copies in locations with different disaster threats.
The word that matters is independent. A mirrored pair of disks in one enclosure is one copy with a spare part. A folder synchronised between two machines is one copy in two places, because a deletion or a corrupted file is faithfully reproduced on both. An independent copy is one that does not change when the first one does.
The Handbook also recommends a combination of online and offline copies, and describes the offline one as a fire break: a copy on a disconnected disk or a tape on a shelf cannot be reached by malware, a mistaken script or a bad synchronisation. Variations on the rule popular with backup vendors, such as 3-2-1-1-0, add this explicitly: one copy offline or immutable, and zero errors after verification.
Why are three copies not enough without checksums?
Because nothing in the rule tells you whether the copies are good.
Preservation specialists call the property you want fixity: in the DPC's words, the assurance that a digital file has remained unchanged. The instrument is the checksum, which the Handbook describes as a digital fingerprint. A program reads every byte of a file and computes a short string from it. Run it again on an identical file and the string is identical. Change one pixel in a 400 MB TIFF, or let one bit decay on a disk, and the string is completely different.
With a checksum recorded for every master on the day it enters the archive, three things become possible:
- Verified copying. After copying 20,000 files to a new drive, recompute and compare. Any file that did not arrive intact is identified by name.
- Detection of silent damage. A file that still opens may no longer be the file you stored. Its checksum will say so.
- Repair. When two copies disagree, the recorded value tells you which is right. The DPC Handbook's instruction is direct: if corruption is found, use one of the other copies to create a replacement.
A checksum tells you that a file has changed. It does not tell you where, or why. And because it identifies content and ignores the file name, it has a second use: finding exact duplicates scattered across old drives.
How do you create and store checksums?
Any modern operating system can compute them from the command line, and archival tools do it in bulk. For a handful of files, or to see what one looks like, the image hash generator calculates the SHA-256 of a file in the browser without uploading it.
On the choice of algorithm, the DPC Handbook draws a practical line. Where checksums are only there to catch accidental loss or damage, MD5 is sufficient, quick to calculate and well supported. Where it matters that nobody could alter a file undetected, for evidential weight or legal admissibility, a stronger algorithm such as SHA-256 is the better choice. An archive that may one day need to show a buyer or a court exactly what it delivered has reason to use the stronger one from the start.
Then store the values where they will survive. The Handbook's advice is to keep fixity information alongside the digital materials and also in separate databases or systems. In practice:
- A manifest with the files. A plain text file in each folder listing every file name and its checksum.
- A column in the catalogue. The checksum sits in the record for each item, next to its file name and location.
- A copy elsewhere. A manifest that is destroyed with the files it describes proves nothing.
How often should copies be checked and replaced?
The DPC Handbook suggests, as a general guideline, checking data tapes annually and hard-drive-based systems about every six months, while noting that more frequent checks find problems sooner at the cost of more load on the storage. After any deliberate change, such as migrating to new storage or converting a format, new checksums should be established for the new files.
Storage also wears out and goes out of support. The Handbook observes that storage systems and media often last only three to five years and that material should be migrated between systems on a plan, before a failure forces the issue. A backup is therefore less a thing you own than a routine you keep:
- every six months, verify each copy against its manifest;
- every year, restore a sample from each copy to prove that restoration works;
- every three to five years, move each copy to new media, verifying as you go;
- after every incident, record what failed and what was replaced.
How much storage does 3-2-1 need?
Three times the archive, plus room to grow, and for film scans the base figure is larger than most people expect. Uncompressed 16-bit scans of medium-format film run to hundreds of megabytes each, and each master usually has an access JPEG beside it. The archive storage calculator takes the number of items, format, resolution and bit depth and returns the total for master TIFFs and JPEG derivatives, with the 3-2-1 multiplier applied. It is worth running before buying anything, because the answer decides whether disk, tape or cloud is the sensible second medium. The choice of master format matters to the sum as well, as the comparison of TIFF, DNG and JPEG 2000 explains.
A worked example: one archive, two kinds of risk
The Robert Harper Photograph Library holds its digital photographs as Nikon NEF raw files, TIFFs and JPEGs, and its inventory, published in full, counts 95,487 unique photographs and 177,143 frames including duplicates and variants. The 503 inventory lines add up exactly to those totals, though they have not been checked frame by frame against the physical material.
The born-digital files are exactly the material 3-2-1 and fixity checking were designed for: there is no negative to go back to. The other side of the archive is the reverse case. The analogue holdings, mainly 6 x 7 colour transparency film, are largely unscanned, which means that for most of those photographs no digital copy yet exists. Digitisation will create the first digital copy of most of them, and every scan will need a checksum from the moment it is written.
That is the point at which storage design stops being abstract. Scanning and preservation specialists who build fixity into their delivery, with manifests supplied alongside the files, will find the scope of the work on the digitisation partners page.
Frequently asked questions
What is the 3-2-1 backup rule?
Keep three copies of every file, on two different types of storage, with one copy held off site. The rule is generally traced to the photographer Peter Krogh's The DAM Book. It protects against the three commonest causes of loss: a device failing, a fault that affects one type of storage, and a fire, flood or theft at one address.
Is cloud storage enough to back up a photo archive?
On its own, no. A cloud service can be one of the three copies and is a convenient off-site one, but an account can be closed, compromised or synchronised with a mistake. The Digital Preservation Coalition recommends combining online and offline copies, and describes an offline copy as a fire break against problems that spread automatically between connected systems.
What is a checksum and why does an image archive need one?
A checksum is a short string calculated from the contents of a file. The same file always gives the same value, and any change, however small, gives a different one. Recording a checksum for every master file lets you prove, years later, that a copy is identical to the original, and tells you which copy to trust when two differ.
Which checksum algorithm should I use, MD5 or SHA-256?
The Digital Preservation Coalition Handbook says MD5 is sufficient where the aim is to detect accidental loss or damage, and is quick and widely supported. It recommends stronger algorithms such as SHA-256 where resistance to deliberate alteration matters, for example as evidence. For a new archive, SHA-256 is a sensible default, and the cost in time is small.
How often should backups be checked?
Regularly, and on a schedule. As a general guideline the Digital Preservation Coalition Handbook suggests checking data tapes once a year and hard-drive-based systems about every six months. Check again after any event that moves or copies files, such as a migration to new storage, and test that files can actually be restored.
Sources
- Digital Preservation Coalition Handbook: Storage
- Digital Preservation Coalition Handbook: Fixity and checksums
- Backblaze: 3-2-1 vs 3-2-1-1-0 vs 4-3-2 (attribution of the rule to Peter Krogh)
Published by Archive.London. Corrections are welcome under our editorial policy.

