When a cable is cut, what matters are the milliseconds.

It makes switching and routing, and it does it in two places at once: the cabinet of a normal building and the cabinet that is on a pole, in a tunnel or in a site hut. That second half is what brings it into our security projects, because in video half the switches are not in a room with air conditioning: they are where the cameras are. And unlike a purely industrial house, the same family also covers the switch that gathers the fibre and the routing that separates the networks.

It also has one thing that in security is worth real money: centralised management that keeps the configuration of every switch on the network. When one dies, you unhook it, you fit the spare and the network hands its configuration back without anyone typing it. Anyone who has changed a switch in a street cabinet on a Saturday afternoon, with the torch in their mouth, understands why that matters.

Instead of hanging the cabinets off a single fibre run, the path is closed and each cabinet has two routes to the recorder: if somebody cuts the fibre with a digger, the traffic goes the other way. The question is not whether it puts itself back together, it is how long it takes.

The generic mechanism that almost every switch carries takes seconds. Seconds on a camera network are not a blink: they are a gap in the recording that does not come back —a video packet that gets thrown away is not asked for again—, they are all the cameras reconnecting to the recorder at once and they are fifty «camera lost» alerts at a stroke on the control desk. The next thing that happens is that somebody starts ignoring those alerts, and that is the real damage. With recovery in milliseconds none of that happens: the gap fits inside the buffers the camera and the recorder already have, nobody gets an alert and the recording stays whole.

Two things that are not in the brochure. The time is not a constant of the device: it depends on how many switches there are in the ring, and the published number is measured in a laboratory, with the ring empty. And the ring shares out paths, not capacity: when one side goes down, everything goes through the other, so it has to be sized for the bad day.

The cabinet of a video installation is usually the worst place in the installation: closed because it is in plain view, full of dust because nobody opens it, and in August, with the sun on it all afternoon, a good deal hotter inside than out. An office switch put in there does not break: it restarts in the small hours when the load drops, works perfectly in the morning and leaves the engineer chasing a ghost for three visits.

Equipment made ready for that differs in unshowy ways. It has no fan, which is the part that clogs with dust and the first to die. It is powered with direct current from the same board as the rest, and it takes two inputs, so one supply can die without taking the network with it. And it mounts on a rail, which is what you have where there is no rack and there is not going to be one.

There is one detail we always use and almost nobody asks for: these boxes carry a volt-free contact, a pair of wires that close when something happens to the switch. It is taken to an input on the alarm panel and the network raises the alert through the same place a forced door does. Nobody has to look at a screen to find out.

By command line, which is where it is really configured and where the configuration ends up written in a file you can save, compare and load back. Anyone coming from another big house recognises the way of working in half an hour. And by browser for day to day: see which port is up, how much power it is giving, switch a port off and on to restart a camera without climbing the pole.

The third is the centralised management, and it is the one that changes the work: the network drawn out, the updates in batches instead of one by one, and the automatic replacement of failed equipment. With a single cabinet it does not add much. With fifteen spread across an industrial estate it is the difference between a network that is maintained and one that is left to run.

Where the video is spread out: an industrial estate, a campus, a port, a plant with several industrial units, a municipal installation with cameras in several streets. There the fibre is going to close a ring because cutting one section cannot leave half the installation blind. And where the maintenance is done by people who do not live there: equipment that stands up to the cabinet, an alert over a volt-free contact and a spare that configures itself when you plug it in is a combination designed for installations that are rarely visited.

In an installation of one cabinet and fifteen cameras in a building with a socket and no heat. It would work more than well enough, but you are paying for a ring, hardened equipment and central management for a network that fits in a normal managed switch, and the administration will feel heavy to you because it is designed for fifteen cabinets.

In the core of a very large office network, where what is asked for is aggregate throughput, access control for the staff's laptops and dense Wi-Fi. That is not what this family does best, and there are houses whose whole life is that.

And where the client's IT team already has an established house, with their people trained and their spares in the store. Bringing a second brand in there adds one more console and one more excuse the day something fails. If what is there copes with the video, it is built on what is there; we only insist when the existing network cannot cope with the part that is out in the weather, and then what is proposed is a piece and not the whole network.

The day to day, yes, and they should. A team that runs managed switching from any other house handles this one: look at ports, see the traffic, restart a camera by switching its port off, add a camera to the network it belongs on, export the configuration to save it. We hand them their administrator account with their password, not ours.

The boundary is in the design of the ring and of who talks to whom, because a small change badly made leaves the network working and lying —traffic still gets through, but there is no second route any more, and that does not show until the fault—; in the sharing out of power when cameras are added, which is a calculation and not an adjustment; and in version upgrades, which are done in a window and with a way back prepared.

And if your team already has another brand, that is not an obstacle: it is the starting point. We arrive with the sum done of how much traffic we are going to put on their network and what we are never going to ask of it, and it is built on what is there for as long as it copes. When it does not cope —almost always on the part that is out in the weather— we say why, with the measurement on the table, and we say it before signing and not in the third month.