Where backups should go: the question of what you are protecting against

The question is usually asked as "how do I take backups", when the real question is: what are you protecting against. Different failures want different destinations, and if you do not know which failure you have in mind, the choice is random.

Three classes of failure

First: something broke inside the machine. The wrong file was deleted, an update half finished, a configuration got mangled. Here a backup helps wherever it lives, even a copy on the same pool.

Second: the disk or the pool is gone. Now the backup sitting on that same pool is gone too. A backup arrangement that worked perfectly a minute ago is absent at exactly the moment it was needed.

Third: the whole machine is gone. Fire, theft, water, a mistake that wiped everything, or ransomware that reached the target while it was attached. Only a copy that lives somewhere else, and preferably is not permanently attached, survives that.

Most setups are protected against the first class and believe they are protected against the third. The difference is invisible until a bad day arrives.

The uncomfortable question

Is your backup target permanently attached and writable by the machine it protects? If the answer is yes, anything that can destroy that machine can destroy the backup too. The sentence is simple, and on its own it eliminates most arrangements.

Destinations, compared honestly

Another storage on the same machine: fast, easy to set up, protects only against the first class. Reasonable as a start, not enough as the only protection.

A second machine on the network: covers the second class and partly the third. For most home and small office setups this is the right balance.

A dedicated backup server: adds things like verifying backups and storing repeated data once, so it saves space and tells you the backup is readable.

An offsite or removable copy: this is the real protection against the third class. Its cost is discipline: it has to be attached and detached, or sent away on a schedule.

The rule that survives

Keep at least one copy where the machine it protects cannot reach it. This matters more than how many copies you keep, because all the other copies can disappear at the same moment.

Space and retention

An unlimited retention policy on a limited disk is an outage with a date on it. If you do not decide how many backups to keep, one day the disk decides for you, and that day is usually the day the backups stop. When choosing a destination, work out how many copies will fit.

Speed and safety do not face the same way

The fastest destination is usually the least protective, because the reason it is fast is that it is close. That is not a coincidence, it is two faces of the same property. When you complain that backups are slow, pay attention to what you are speeding up.

What Atlas does

Atlas lists only the storages that can genuinely hold backups as targets, so it does not offer you a destination you cannot use. If the target you pick sits in the same place as the disk of the machine you are backing up, it says so at the moment you pick it. As a note rather than an error, because a backup in the same place has a legitimate use and is not something to block; but you need to know that it protects you only against the first of the three failure classes above.

When taking a backup it offers three modes, without stopping the machine, suspending it briefly, or stopping it fully, and explains what each one does next to it, because this choice is about the consistency of a running machine and the name alone is not enough.