Guides · Running a lab

Backing up a lab that cannot afford to lose a year

The 3-2-1 rule, adapted to a lab that buys hardware once and has no IT budget. What to spend, what to skip, and the test that tells you whether any of it works.

Reading time
10 min
Licence
Free to copy into your own lab documentation. No signup.

Most labs do have backups. What they usually do not have is a backup that has ever been restored from, which is a different thing, and the difference only becomes visible on the worst possible day.

This guide assumes the situation most Indian academic labs are actually in: a one-time equipment budget, no recurring IT line, no dedicated support, and a strong and entirely rational preference for hardware you own over a subscription you have to defend every year.

Who this is for

Anyone responsible for a lab's data who has an external drive and a vague sense that it is not enough. No IT knowledge assumed. Most of what follows costs nothing.

3-2-1, in plain terms

The rule survives because it is about failure modes, not products. Three copies, on two different kinds of storage, one of them somewhere else.

  • Three copies — the working copy plus two backups. Two copies means one failure away from one copy, and one copy is not a backup.
  • Two kinds — because identical drives bought together tend to fail together. A disk and a network share; a disk and a cloud account.
  • One elsewhere — because fire, theft, flood and a failed UPS take out everything in one room at once.

“Somewhere else” does not have to mean cloud. Another building on campus counts. A drive in the PI's office counts. It has to be a place that a single bad event in the lab cannot reach.

What actually goes wrong

Worth knowing, because it decides what you defend against.

Drives fail without warning

Mechanical drives typically last three to five years in a warm room, and the failure is usually sudden. An external drive that lives on a bench near an instrument is in a worse environment than a server room by a wide margin.

People delete things

More data is lost to a wrong drag, an overwritten file or a cleared folder than to hardware. A mirrored copy that syncs instantly propagates the deletion instantly. This is why sync is not backup.

Ransomware encrypts what it can reach

Including the mapped network drive and the plugged-in external disk. A backup that is always connected is a backup that can be encrypted with the original.

The two properties that matter most are the ones people skip: a copy that is disconnected, and a copy that keeps a version from before the mistake.

A setup that costs almost nothing

For a typical lab producing a few hundred gigabytes a year:

  1. Working copy — wherever the lab actually works. Shared drive, lab server, instrument PC.
  2. Local backup — a scheduled copy to a drive that stays in the lab. Automatic, daily or weekly. Use the operating system's own tool; it is free and adequate.
  3. Offline rotation — two drives, alternating. One is connected, one is in a different building. Swap monthly. The disconnected one is what survives ransomware and the wrong drag.

Two drives large enough for a year of data, bought once, is the entire hardware cost. The rotation is the part that requires a person, which is why it needs a named owner and a date — the first Monday of the month, by one role.

Before buying anything, ask the computer centre

Most Indian universities and institutes provide storage that departments do not know about, often generous and already backed up. It is worth one email before spending a rupee. Ask specifically: how much, is it backed up, how would we restore, and who do we contact.

The only test that counts

A backup is a claim until you have restored from it. Twice a year, do this — it takes twenty minutes.

The twice-yearly restore test 0 / 7
Pick a file at random from six months agoNot one you chose because you know it is fine. Random, and old.
Restore it from the backup, not the originalTo a different folder, so you cannot confuse the two.
Open it in the software that made itA file that copies but will not open is not a backup. This catches silent corruption.
Time how long the whole thing tookIf a single file takes an hour, a full restore is a week. Better to learn that now.
Check the newest backup is actually recentSilently stopped jobs are the commonest failure. Look at the date, not the schedule.
Confirm somebody other than you could do itWrite the steps down. Backups that only one person can restore are a single point of failure.
Swap the offline drive while you are at itTie the rotation to the test and you will remember both.
Why we care about this

Backups protect the bytes. They do not protect the meaning — a restored folder of files with no record of which run produced which result is data you still cannot use. That link is what woodle.cloud keeps. Do the backups regardless; they are the floor, not the ceiling.

Where your files live →

What not to bother with

Some advice does not survive contact with an academic lab.

  • Backing up everything equally. Raw instrument output and analysis code are irreplaceable. Derived files you can regenerate are not. Protect the first properly rather than everything badly.
  • RAID as a backup. RAID protects against one disk failing. It does not protect against deletion, encryption, theft or fire, and it is a single machine.
  • A consumer sync folder as the only copy. It mirrors mistakes, and free tiers fill silently, at which point it stops and tells nobody.
  • Elaborate schedules nobody maintains. A weekly copy that runs for five years beats a nightly one abandoned in March.

The number worth knowing

Ask the question directly: if the lab's main storage died right now, how much work would we repeat?

Answer it in weeks of somebody's time, not gigabytes. That number is what justifies two drives and half an hour a month, and it is the only version of this argument that works in a budget meeting.

Take it with you

The restore test is free to copy into your own lab documentation. Put a date on it. No attribution needed and no signup. If you do one thing, do the test — most labs discover something in the first twenty minutes.