Growing by bays is fine. Until it is four boxes chained together.

It is network storage for the medium-sized company, administered from the browser, and with a wider catalogue than average: models with more expansion slots, with faster networking as standard, with bays for solid-state drives as well as spinning ones, and expansion boxes to add bays when the first ones fill up.

That gives you room to manoeuvre in a video project. You can put a faster network towards the recording server, or dedicate a few fast drives to the video software's index and the spinning ones to the video, which is the split that really improves how quick it feels when somebody searches by date and time.

And it brings, like all of this class, a large catalogue of applications you can install inside it. On an office machine that is an advantage. On a machine holding up the recording of a security system, what is surplus gets in the way: every thing installed is one more thing to update and one more way in.

The idea is a good one: when the bays fill up, you hook on an expansion box with a cable and add more drives without changing machine. For a system that has grown from twenty to thirty-five cameras, that solves the problem without throwing anything away.

What you have to know is that the cable is now a part of the system. If it comes loose or fails, the volume that lived on both sides is left half there, and depending on how it is configured that can mean recording stopped. The rule we follow is not to spread one volume across the main box and the expansion when it can be avoided: each box with its own volumes, so a fault stays where it is.

And there is a cumulative effect: two boxes get on well, four chained together are already a fragile system and hard to watch, with more cables, more power supplies and more things that can go dark. When the sums come out at four boxes, the honest answer is almost always to change league instead of carrying on chaining.

The sums: the capacity of the box is not the one in the brochure. You have to subtract what is lost in formatting, the drives given over to redundancy, the free margin it is wise not to touch and, above all, you have to use the real bit rate of the cameras and not their catalogue average. A scene with movement uses considerably more than a still one, and summer uses more than winter. We sum it up in a sentence we repeat a lot: a number of terabytes is not a number of days.

The drive that fails: with redundancy, one breaks and the system carries on. What you have to keep in mind is the rebuild, which with large drives is many hours and sometimes days. While it lasts, the other drives run flat out and there is no protection any more: it is the riskiest moment in the life of the system. Two decisions come out of that: with large drives, tolerate two failures instead of one, and leave a spare drive inside so it comes in on its own.

And the sentence that always has to be said, however repetitive it sounds: a set of drives with redundancy is not a backup. It protects you from one drive breaking. It does not protect you from a deletion, from a fire or from somebody carrying the box away.

It fits in a medium-sized installation, just like its direct competition, and it fits particularly well when you need something specific from the hardware: more network bandwidth towards the recorder, fast drives for the video software's index, or expansion by bays because you know the installation is going to grow in stages.

It stops fitting for the same three reasons as always. When the simultaneous write and read rate goes past what a box like this holds comfortably. When the system cannot be stopped, because here there is a single controller and a fault or an update stops the storage. And when the capacity is only reached by chaining several boxes together.

There is a fourth reason that is not technical: if the client has an IT department with a view of its own that prefers enterprise storage because it already knows how to operate it, that preference counts. Administering two different classes of storage costs more than the difference in price.

With access from the internet, never. Network storage reachable from outside has been the target of mass attacks, and one encrypted by an extortionist takes the whole recording with it. This lives on its own network, with no direct way out, updated and with its own passwords. If it has to be looked at from outside, you come in through a tunnel and not by opening a port.

Nor with half the application catalogue installed. On a machine holding up the recording you install what is needed and nothing else, and what comes as standard and is not used gets switched off. It is not fussiness: every thing switched on is a door and an update waiting.

And we would not fit it as the only place where an important recording exists, nor with desktop drives inside, nor in a closed cabinet with no ventilation: a box full of drives is a constant source of heat, and heat is what kills drives.

The day-to-day is simple: the state of the drives from the browser, alerts by email, changing a broken drive, checking that the copies are being made. A person from IT handles that without difficulty, and it is as well that they do: the video storage has to be in the house inventory and in the house alerts, not be a box only the installer looks at.

What adds work compared with a simpler machine is precisely its flexibility: there is more software inside to update, and if there are expansion boxes you also have to watch their power supplies and their cables. So the agreement includes one more thing: a quarterly review of the updates waiting and of what is installed and not used.

Our side is still the sums: what happens to the days of recording when new cameras come in, when it is time to stop chaining boxes and change league, and a check once a year asking for the recording from the oldest day the contract promises.