Where should I keep a second copy of my files?

On separate hardware that is not permanently writeable from the machine holding the original, and ideally not in the same building. The point of the copy is to survive whatever reaches the first one.

The question to ask about any proposed second copy is simple: what could reach the original that would not reach this?

The properties that make a copy count

Separate hardware. Different drives in a different enclosure, so an enclosure fault, a power supply failure or a controller problem does not take both.

Not permanently writeable. Software that encrypts a share will encrypt everything it can write to, including a copy mounted as a folder. A copy that is disconnected, pulled rather than pushed, or held where the source machine has no delete permission survives that.

History. A copy that mirrors the original exactly will faithfully mirror a deletion the next time it runs. Versions, snapshots on the destination, or simply a delay before the copy runs are what turn a mirror into something you can recover from.

Somewhere else. A fire, a flood or a burglary is an event that happens to an address rather than to a device.

The convention usually written as three copies, two kinds of media, one of them off site is a compact way of saying all four of those at once.

The arrangements people actually use

Rotating external drives. Two drives, one connected and one elsewhere, swapped on a schedule. Cheap, unplugged when idle, and off site half the time. The weakness is that it depends on somebody doing the swap.

A second network enclosure, at another address, receiving a scheduled copy. This is the arrangement that scales, and most enclosure makers publish a tool for replicating between two of their own machines. The copy should be pulled by the destination or written where the source cannot delete, so that trouble at the source does not propagate.

A remote service. Off site by construction, current by construction, and dependent on the connection for both the copy and the restore. Check what a full restore would actually involve before relying on it, because that is the operation nobody rehearses.

Optical or tape media for archives that never change. Genuinely separate, genuinely offline, and a different kind of media, which is the second leg of the convention.

Most households end up with two of these rather than one, which is the intended outcome.

What the copy is protecting against, in order of likelihood

A deletion or an overwrite, which is by far the most common. Hostile encryption. A failure of the enclosure rather than of a drive. Theft. Fire and water.

Notice that a redundancy level in the original machine addresses none of those. It addresses the failure of a drive, which is the one case where the second copy is not needed.

The part that gets skipped

Test a restore. Take one file, and then one folder, and bring them back from the copy onto a machine. A copy nobody has ever restored from is a copy whose usefulness is a matter of faith, and the failure modes — wrong permissions, a forgotten passphrase, media that will not read — are all quiet until the day they are not.

The manufacturer’s own documentation comes first, and this page is not a substitute for it. Every rate, capacity, class figure, endurance rating, port speed and power figure on this page is reported from the manufacturer's own documentation, with its unit written as published, and named as such — none of it is measured here. A stated capacity is reported as the maker counts it, and is never restated in another unit, because the two are not the same figure. And the second statement repeated rather than assumed: a RAID level is not a backup. It keeps an array readable through the failure of a drive, and it does nothing about a file deleted by mistake, an array encrypted by ransomware, a theft or a fire. A second copy somewhere else is a separate decision, and nothing on this site describes a machine as a place where files are safe.