The drive from the home computer is no use here. And it is not marketing.

An office computer drive is made to sit idle most of the day and work in bursts. A recorder's drive works without stopping for the three thousand six hundred and fifty days it will be fitted: the cameras write at all hours, every day, without pauses. They are two different workloads and the manufacturers declare them: there is a number of terabytes a year that each drive is designed to take, and the figure for a desktop drive is a fraction of the one for a surveillance drive.

But the most important difference is another one and almost nobody tells you about it. When a drive finds a piece it cannot read properly, it has two options: keep trying for a long time until it gets it out, or give up quickly and raise an alert. A desktop drive keeps trying, because on a computer what matters is not losing the file. A surveillance drive gives up sooner, because on a recorder what matters is not stopping writing: while the drive keeps trying, the video coming in is not being saved. That behaviour is the technical reason this range exists, and also the reason a desktop drive inside a redundant set can make the system throw it out for being slow.

There is a third difference, more physical: a drive designed to sit alongside seven others copes with their vibration. Manufacturers declare how many units per enclosure each model has been tested for, and putting a single-bay drive in a sixteen-bay enclosure shortens its life without anyone understanding why.

The declared annual workload, which has to be greater than what the cameras will write in a year. It is a simple sum and it is almost never done. The number of bays that model has been tested for. And that it is on the list of validated drives for the equipment it is going into, because that list exists precisely for continuous recording.

The fourth one is trade practice: do not buy all eight drives from the same batch. Identical drives, made in the same week and put through the same work, tend to wear out around the same dates, and that is exactly what you do not want: the second failure arriving while the first is being rebuilt is the one scenario in which everything is lost.

And a decision about size that runs against intuition: bigger drives are not always better. The bigger the drive, the longer it takes to rebuild when it is replaced, and during that rebuild the system has no protection and the other drives are working flat out. With very large drives, the sensible decision is to tolerate two failures instead of one.

The calculation with example figures, so you can repeat it: a camera at three megabits per second writes a little over thirty gigabytes a day. Twenty cameras like that, over six hundred a day. It looks easy, and then come the subtractions: what is lost to formatting, the drives given over to redundancy, the free margin it is wise to leave, and the fact that those three megabits are an average that goes up a fair amount in scenes with movement.

That is why the retention period promised has two parts: the calculation, done beforehand, and the measurement, done two weeks into recording. The second is what turns an estimate into a commitment, and it is half an hour of work that almost nobody puts in.

And one thing that holds for any drive of any brand: one that has been writing for five years is at the end of its life even if it works. Replacing them by the calendar costs money; replacing them when they fail costs the recording of those days, which is what nobody can get back afterwards.

A surveillance drive is not a drive for everything. On the volume where the operating system or the video software database lives, what you need is something else: a solid-state drive, and if possible two mirrored. That is the volume that gets corrupted when the power cuts out, and the one that makes a search by date and time fast or slow.

Nor would we fit them in equipment that is not on their compatibility list, or in an enclosure with more bays than the model declares it supports, or mixed with desktop drives in the same set. The last one happens more than you would think: somebody replaces a broken drive with whatever is to hand and the whole set starts misbehaving.

And we would not fit them in a hot cabinet assuming they will last. Heat is what kills drives, and not all at once: they last two years instead of six and nobody connects it with the temperature of the room.

Changing a drive in a bay is a five-minute operation that requires nothing to be switched off, and dealing with an alert is reading an email. With a one-page procedure —what model to order, where the spare is, how to check the rebuild has finished— the client's own staff can perfectly well handle it, and it is better that way, because the drive gets changed the same day and not when somebody happens to come by.

What is worth making clear is what does NOT happen by itself: after changing a drive you have to check that the rebuild finished properly, and you have to replace the spare you have used. An enclosure with the spare slot empty is without a safety net again and nobody has noticed.

Our part is the part that needs the calculation: when the drives are due for replacement by age, whether the recording period is still being met after adding cameras, and when the redundancy chosen has stopped being enough because today's drives are far bigger than the ones from back then.