There are drives that write the tracks overlapping. In a recorder, no.

In the catalogue of either of the two big drive houses there are four worlds: the desktop computer drive, the network storage one, the surveillance one and the data centre one. From the outside they are almost identical and the price is closer than it should be. What changes is the load they are designed for: how many terabytes a year they take, how many hours a day they work, how many drives alongside them they have been tested with, and what they do when they find a piece they cannot read properly.

That last one is the difference that decides whether a recorder works well. A desktop drive keeps trying for a long time to recover a doubtful sector, because on a computer what matters is the file. On a recorder what matters is not stopping writing, so the right drive gives up sooner and raises an alert: while it keeps trying, the video coming in is lost.

And there is a second trap that you only know about by having suffered it: some drives, in order to fit more capacity in, write the tracks overlapping like the tiles on a roof. That is fine for storing things and reading them, and bad for rewriting without stopping, because every change forces a whole zone to be redone. In a recorder, which overwrites the oldest material every day, that shows; and during the rebuild of a redundant set it can turn a few hours into several days. You have to check on the datasheet that the model chosen is not one of those.

The video volume: spinning drives designed for continuous recording, of whatever size is needed, with the declared annual load above what the cameras will write in a year and tested for the number of bays in the enclosure.

The volume for the system and the video software database: surveillance drives do not go there. Two mirrored solid-state drives go there, because it is the volume that decides whether a search by date and time answers quickly, and it is the one that gets corrupted when the machine loses power halfway through writing. Being small does not make it less important: without that volume there is no system, even if the video is intact.

And the volume for what has to be kept longer, if there is one: there what matters is cheap capacity and coping with being written and read a little. They are three different decisions and saving on the second is the one that costs most.

A camera sending three megabits per second writes a little over thirty gigabytes a day. Twenty cameras like that, over six hundred gigabytes daily. With forty terabytes that comes to about sixty days, and that is the number that goes in the offer.

Then come the subtractions. The capacity on the box is not what is left after formatting. The drives given over to redundancy are not capacity. A volume filled to the brim behaves badly, so a margin is held back. And the three megabits are an average: in a scene with movement, with summer light and with the car park full, they go a good deal higher. The sixty days end up as thirty-five, and nobody has done anything wrong: the sum was incomplete.

So the honest number comes out in two steps: the sum before buying, with the subtractions in plain sight, and the measurement of the real rate two weeks into recording. The second step is what turns an estimate into a period you can sign for.

A desktop drive in a recorder, never, however tempting the price difference and even if it works for the first year. And a drive with overlapping tracks neither, for the reason explained above: the day of a rebuild it becomes the problem.

Nor would we mix drives from different ranges in the same set, or put a model tested for two bays in a sixteen-bay enclosure: the vibration of its neighbours shortens its life and nobody connects it with that.

And we would not fit them in a hot cabinet taking anything for granted. Heat is what kills drives, and it does it without warning: they last half as long as they were meant to and the fault looks like bad luck.

Changing a drive is opening the bay and putting the new one in, with nothing switched off. Dealing with the alert is reading an email. With a piece of paper that says which exact model to order, where the spare is and how to check the rebuild finished, the client's IT team or maintenance team runs it. And it is as well that they do, because that way the drive gets changed the same day.

What has to be written on that paper is exactly what gets forgotten: the exact model, because fitting whichever drive was to hand is the most common way of degrading a set; and that the spare used has to be replaced, because an enclosure with the slot empty is without protection again.

And there is one thing left that is not daily administration and is worth reviewing with us once a year: how many years the drives have been in, whether the redundancy chosen is still reasonable for the size they are today, and whether the days of recording the contract promises are the ones the system really gives.