Rolling mill hall with a coil of sheet metal and a control cabinet on a column

MET · Metals & Mining

Field note · Metals & Mining · Manufacturing

Flat networks: why one switch can see the whole plant

Range Lab2026-11-036 min read

Ask someone on a plant floor to describe the network and, on a lot of sites, the honest answer is a shrug and six words: "It's all on the one switch."

Those six words describe a flat network completely. They also say more about a site's exposure than an hour of talk about firewalls.

Everything can reach everything

A network is flat when any device can talk directly to any other, with nothing in between to say no. The controller in the furnace cabinet, the HMI in the control room, the historian, the engineering workstation and the printer in the office all see each other's traffic and can reach each other's ports. No zones, so no conduits, so nothing to inspect, log or block.

Flat doesn't mean cheap. Plenty of flat networks run on good managed switches with VLANs configured. The VLANs were set up to manage traffic, not to keep anyone out, and a router somewhere joins them back together without a single rule. Segmented on paper, flat in practice: it is the most common kind, and the hardest to spot from a drawing.

A quick test. Pick the oldest controller on site and ask what could open a connection to it. On a flat network the answer is anything with a cable, including the laptop a contractor carried in this morning.

Wide view of a metals plant with a furnace, overhead cranes and long conveyors
MIN · 02 / 09 · wide

Nobody designs one

Flat networks grow. They usually arrive three ways, often all three on the same site.

Growth. A line goes in with its own small network. Then a second line, then a packing hall, each separate because each was its own project. One day someone wants production counts from all of them on one screen, and the fastest fix is to join the networks. A switch does it, first time, for very little money, and nobody revisits it.

Convenience. The engineering workstation needs to program controllers on three lines, so it goes where it can reach all three. The historian needs data from everything, so everything gets a path to it. Each decision is sensible on the day. Added up, the easiest path for the engineer is the easiest path for anyone.

Integrator defaults. A machine builder ships a cell with a switch in the cabinet and a few spare ports. The commissioning engineer, on a deadline, plugs it into the nearest uplink, and the switch keeps its factory settings forever. In metals and mining the story is older and starker: the network was built before anyone imagined an office connecting to it, and when the office finally arrived, through a maintenance system or a contractor's link, it landed on a network that had never needed a boundary.

Every connection on a flat network was somebody's good idea.

One laptop, the whole plant

On a flat network, being on the network and being able to reach the controllers are the same thing. A contractor's laptop that has visited fifty other sites, a workstation that also reads email, a vendor's remote-access box: whichever one is compromised first, the person behind it can see every device and try every port.

Plant protocols make it worse than an office would be. Modbus and the programming protocols controllers listen on have no authentication of their own, because they were designed for a wire only trusted equipment could touch. On a flat network, everything touches that wire. The oldest, least patchable controller on site sets the risk level for all the rest.

The office firewall helps, up to a point. It keeps the office out of the plant, which matters. But the laptop, the workstation and the vendor box are already inside. Our Level 3.5 note covers the boundary; this one is about what happens past it.

A weekend won't fix it

Once a plant sees its network this way, the instinct is to fix everything in one long weekend. It rarely ends well. Undocumented devices stop talking to things they depended on, the production team loses faith, and the project closes with half the rules disabled "for now". Segmentation works as a staged project, and the stages have an order.

First, know what's there. Not what the drawing says, or the switch configuration, but what is actually on the network and who it actually talks to. A passive capture at the right switch ports gives you this without sending a packet to a controller, which is the only responsible way to look at the kind of controller you find on a mill. The result is an asset inventory and a map of conversations, and you need both before writing a single rule.

Second, draw zones on paper. Group equipment by shared purpose and trust: a line and its HMIs, the historian and its servers, the engineering workstations, the vendor links. Draw the conduits between them from the conversation map, not from memory. This is the work ISA/IEC 62443-3-2 describes, and it is where the surprises turn up. Better to meet them on paper.

Third, cut the cheapest boundary. Most sites have one boundary that is both the most valuable and the easiest: the vendor remote-access paths, the office-to-plant join, or the engineering workstations. A firewall or brokered jump host goes in at that one point, with rules written from the captured traffic, and the production team watches nothing break. That proof pays for the rest of the project.

Fourth, the rest, one window at a time. Each remaining boundary gets its own small change, timed for a maintenance window when the affected line is down anyway, with a named person on site ready to roll back. In manufacturing that often means one line at a time over several months. In a smelter it means waiting for a furnace reline, because nobody restarts a furnace on a Tuesday afternoon.

Sample · illustrative, not a client site

Stage What changes When Production risk
1 Passive capture; asset list and conversation map Any time; nothing touched None
2 Zones and conduits drawn on paper Any time; nothing touched None
3 Vendor remote-access paths brokered through one point Next planned stop Low; vendors notified
4 Office-to-plant join moved behind a firewall Next shutdown Moderate; rules from capture
5 Lines separated into zones, one per window Over the year Low per line; one at a time
Close view of a control cabinet with a PLC rack, terminal blocks and a bundle of patch cables
MFG · 09 / 09 · detail

Slower, and it holds

This is why the Range method starts with listening rather than architecture. We don't sell firewalls and we don't configure them. We put the plant in a position to specify them. A flat network isn't a failure of the people running the site, and no product fixes it on its own. Knowing, in writing, what is on the network and what talks to what fixes it, followed by cutting boundaries in an order the plant can live with.

Done this way, segmentation is slower than the weekend version, and it holds. The plant ends up with zones that match the equipment really there, named and logged conduits, and a production team that watched every change go in and saw nothing break.

New to the vocabulary? Learn OT covers it from zero, free.