Equipment that is installed once and visited almost never.

It makes switching, routing and industrial connectivity, and little else. There is no office Wi-Fi here, no data centre switch, no domestic range. What there is is equipment for places where a box has to work for years without anybody going in to look at it.

Its natural ground is public service infrastructure: railway —track, stations, level crossings, rolling stock—, the water cycle —abstraction, pumping, treatment— and energy. Three features in common: they are spread out, they are alone, and they are inside a framework where the equipment you install has to be justified.

In security it turns up for exactly that reason. When the cameras are in a pumping hut fifteen kilometres from the next building, in a marshalling yard or inside a substation, the network that joins them does not look like a company's, and the brand that makes it usually comes from this world and not from the IT one.

Vibration first, because it is the one that surprises most. A switch bolted to the structure of a bridge, to the cabinet of a level crossing or to a rack inside a vehicle takes constant knocks for years, and what fails is not the electronics: it is the connections. A connector that in a rack stays where you left it works itself loose there, and out comes the worst fault of all: intermittent, with no pattern, and one that disappears when the technician arrives and moves the cable to look.

Electrical noise next. Beside the switchgear of a substation or with the overhead line above, the air is full of interference, and that shows in a deceiving way: the links do not go down, they start losing packets. And a link that loses packets and does not go down appears on no alarm panel: it appears in the recording, in jumps, two weeks later.

And the third: nobody in front of it. A pumping hut is visited when maintenance is due, and that changes what equipment makes sense. The one that raises the alert on its own that something is wrong. The one that restarts and reconfigures remotely. The one that carries on working with one of its two power supplies dead. And the one that does not need anybody to change a jammed fan, because it has no fan.

With points spread over kilometres, hanging each one off a single fibre run means that any building work, any rodent and any digger leaves a whole branch blind until somebody goes out to splice it. Closing the path into a ring gives two routes, and then the question is how long it takes to rebuild itself.

The generic mechanism takes seconds, and seconds on a camera network are a gap in the recording that is not recovered and a volley of «camera lost» alerts all at once. With recovery in milliseconds, the gap fits inside the buffers the camera and the recorder already have: nobody notices. With two warnings that are not on the datasheet: the time depends on how many nodes the ring has, and the ring gives paths and not capacity, so it is sized for the day one side is cut.

They mean that the list of admissible equipment is part of the specification. There is no arguing about which switch is better: you check which switch is approved for that use and who keeps that list. Equipment that would work without blinking may not be installable, and there is no technical appeal against that. Changing brand there is not an engineering choice: it is a modification somebody approves, with its file and its lead time, and the lead time is almost always longer than the site's. We have seen it turn a two-week extension into a two-month one.

There is a side effect worth knowing: in these sectors it counts for a lot that the same equipment stays available for a long time. A water installation is extended years later and what it wants is to put the same box, with the same configuration, in the cabinet next door. The houses that live off this world keep their products going considerably longer than the consumer ones, and that is not appreciated on day one.

The day to day is run from the browser —which port is up, how much traffic is passing, restarting a camera by switching its port off without sending anybody out to a hut—, and there is a command line and central management for when there are many points.

What makes the difference in installations that are rarely visited is that each unit's configuration can be exported, saved and loaded back exactly as it was. That turns replacing a dead switch into a twenty-minute task the contractor's electrician can do with a spare box and a file, instead of a visit from somebody who knows how to configure a network. In a spread-out installation, that is the difference between a day and a week.

Where the cameras are at distant points that have to be joined: pumping huts, water tanks, treatment plants, substations, marshalling yards, stations, level crossings, tunnels. Where there is vibration or power switchgear nearby, which is where a normal network starts losing packets without anything going down. And where the sector has its list of admissible equipment.

And in a case that is not exactly about security and comes up a lot: when the video network has to live alongside the installation's control network —PLCs, remote control units, whatever governs the pumps— without touching it. There you need equipment that understands both worlds and that lets you separate them properly, because whoever runs the control of a pumping station is not going to accept, rightly, some cameras sharing their network with no boundary.

In a normal office building, with a rack, a socket and a temperature fit for people. There this class of equipment brings nothing you would notice and costs more than an ordinary managed switch with more ports and more power for the cameras.

As the main network of a large campus, where what is asked for is aggregate throughput, dense Wi-Fi and access control for the staff's laptops. That is not what this house does.

And ahead of the network that is already there when the one already there reaches the place and holds up. In these installations there is usually a control systems department with its brand, its spares and its procedure; one more brand means one more store and one more argument on the day of the fault. What we propose almost always is the missing piece —the points the existing network does not reach— and not the whole network.

There is usually corporate IT, a control systems team that handles the plant side, and sometimes a maintenance contractor who is the one who actually goes out to the places. All three can run the day to day; the problem is not capability, it is that if it is not shared out in writing nobody runs it. IT does not go into a substation and control does not touch a network configuration.

With the share-out decided, the day to day is perfectly manageable: looking at ports, restarting a camera, registering another one, exporting the configuration, changing a dead unit with the saved file. We hand over the administrator account in the customer's name, the map of what is at each point and a short replacement procedure.

The boundary is in the ring and in the separation of networks. Touching the ring badly leaves the network passing traffic and with no second route, and that is not noticed until the cut. And moving the boundary between the video network and the control network is not decided from a configuration screen: it is decided with both people in charge present. If the control team already has its brand and its way of working, that is respected and we build on top of it; we adapt to the network that is there as long as it reaches where it has to reach and takes the throughput with margin. When it does not, we say why, with the measurement done and before signing.