Comparison
Connect a PLC to the plant network without exposing drives
Short answer
Keep the PLCs, HMIs and drives on one dedicated machine subnet and connect it upward through exactly one controlled boundary. A routed firewall, a second Ethernet interface such as the CP 1243-1, or a 1:1 NAT switch of the S615 class all do the job; the deciding factors are which CPU you have and which side initiates each connection.
On this page
Keep the PLCs, HMIs and drives on one dedicated machine subnet, and give that subnet exactly one controlled path up to the plant network: a routed firewall, a second interface in the PLC, or a 1:1 NAT switch. Which one you build comes down to two facts: what your CPU can carry (an S7-1200 needs a CP 1243-1 for a second LAN; an ET200SP needs the 1514SP-class CPU or a CP/CM module), and which side initiates each connection. This article compares the three boundaries and commits to one per situation; if you are still deciding what belongs on the plant network at all, start with segregating machine drive traffic from the plant network.
What must talk to what #
Write the flows down before you pick hardware, because the whole design hangs on the initiator column:
- Plant to PLC: SCADA or historian polling, plus engineering access. The plant initiates.
- PLC to HMI and PLC to drives: local traffic that stays on the machine switch and never crosses the boundary.
- Machine to plant, if it exists: the PLC pushing values to a broker or database. The machine initiates.
Drives get no rule, no translation and no plant-side address. If a plant system claims it must poll a drive directly, it gets the data through the CPU instead; the PLC is the only data path out of the machine subnet. IEC 62443 gives this arrangement its vocabulary: assets with the same protection requirement form a zone, and every communication leaving the zone crosses a conduit, meaning one controlled, documented channel. Your machine cell is the zone; the three conduits are compared below.
The three designs side by side #
| Design | Use it when | What the PLC needs | Rule behavior | Main pitfall |
|---|---|---|---|---|
| Routed machine subnet with plant firewall rules | Addressing is distinct and you can put rules on the plant firewall | One interface, nothing extra | Stateful rules on source, destination and port; return traffic tracked automatically | A Layer 2 mispatch bridges past the firewall, and machine DHCP can hand out plant addresses |
| Second interface in the PLC (CP 1243-1 or CP/CM) | The machine subnet must not be routed and plant access is S7 to the CPU only | S7-1200: CP 1243-1; ET200SP: 1514SP-class CPU or a CP/CM module | No firewall in the module; the CPU's access settings are the policy | Two switched ports are one network, and the interface carries the CPU's own connections, not the HMI's |
| 1:1 NAT switch (S615 class) | Address ranges overlap or policy requires translated addresses | Nothing; the switch sits in front of the subnet | Each visible device needs a mapping plus a rule; unmapped devices are unreachable | Rules read backwards: initiator and return path must be modeled before configuring |
Routed machine subnet with firewall rules #
This is the default and the design to build when the plant runs a firewall you can configure. The machine subnet stays switched locally: PLC-to-HMI and PLC-to-drive frames never leave the machine switch. One VLAN carries the subnet to a firewall interface, and the rule set expresses only the flows you listed. Stateful filtering tracks each permitted session, so return traffic needs no extra rule, and everything else defaults to deny in both directions. Rules like these are one layer of why layered OT security controls protect a network path.
The failure mode is physical, not logical. If someone patches a machine switch port into a plant switch port, you have a Layer 2 bridge that bypasses the firewall entirely, and the machine's DHCP server, if it runs one, starts offering addresses to plant devices. Static addressing on the machine subnet, or a tightly scoped DHCP with a labeled uplink port, closes that hole.
Second interface in the PLC #
Availability first. No S7-1200 CPU ships two interfaces; several CPUs carry a two-port switch on one interface, which is one network with one IP address, not two LANs. The CP 1243-1 gives the S7-1200 a genuine additional Ethernet interface on the CPU bus, and its documented uses match your pattern exactly: S7 communication across two separated LANs, plus telecontrol links to a control center with TELECONTROL SERVER BASIC V3. On the ET200SP side, the 1514SP is the CPU that carries two onboard interfaces; every other route to a second LAN there is a CP/CM module, and both platforms offer at least one.
Behavior: the CPU's onboard interface faces the drives and HMI, and the second interface faces the plant carrying the CPU's own connections. Plant systems poll the CPU through it; for S7 read access, enable the CPU's PUT/GET access setting in TIA Portal and compile and download the change. What this design does not do is route for anything else, so the HMI cannot reach the plant through the CP. If plant systems must reach more than the CPU, this design is out; use one of the other two.
1:1 NAT switch at the boundary #
A NAT switch of the S615 class sits between the machine subnet and the plant and combines address translation with a firewall, so only chosen devices talk upward. In 1:1 mode, each machine address that must be visible gets its own plant-side translated address; devices without a mapping are unreachable, which is exactly the drive requirement. You expose the two PLCs and nothing else.
NAT's cost is that the rules read backwards. You must know which device initiates: a plant-initiated poll targets the translated address, while a machine-initiated push needs the outbound flow and its return permitted through the translation. Model the initiator and return path per flow before touching the configuration, and expect diagnostics to show translated addresses rather than the real machine ones. The initiator question is the same one behind remote SCADA access through firewalls and proxies.
Worked example: two PLCs, sixteen drives, one poller #
Example addressing plan for one cell: an S7-1200 at 192.168.30.1, an ET200SP at 192.168.30.2, HMIs at 192.168.30.11 and 192.168.30.12, drives at 192.168.30.21 through 192.168.30.36. Plant side: 10.44.0.0/16 with the poller at 10.44.7.22.
The arithmetic of the exposure:
- Mask /24 = 255.255.255.0 leaves 32 - 24 = 8 host bits; 2^8 = 256 addresses, minus network and broadcast = 254 usable.
- Drives: 36 - 21 + 1 = 16 addresses that must appear in no rule and no mapping.
- Design 1 rule set: permit 10.44.7.22 to 192.168.30.1 on TCP/102; permit 10.44.7.22 to 192.168.30.2 on TCP/102; deny everything else to 192.168.30.0/24. Exposure: 2 of 254 addresses, about 0.8 percent.
- Design 3 variant: map 10.44.60.11 to 192.168.30.1 and 10.44.60.12 to 192.168.30.2, so the poller targets 10.44.60.11:102 and 10.44.60.12:102. Unmapped addresses: 254 - 2 = 252, including all 16 drives; a plant-side sweep of the translated range finds two devices.
Both designs pass the requirement. Design 2 does it with no firewall rule at all, because the plant session terminates on the second interface at the CPU.
Commissioning and the final decision #
Cut over in a planned window with the machine stopped; dropping drive communication mid-cycle faults the cell. Then work the list:
- Address the machine subnet, statically or with a scoped DHCP, and label the uplink port.
- Confirm PLC-to-HMI and PLC-to-drive traffic works locally before any uplink exists.
- Build the boundary with default deny: VLAN plus rules, CP connections and access settings, or NAT mappings.
- Connect the uplink, test each permitted flow from its real source, and negative-test a plant-side connection to a drive address: it must fail.
The decision, one line per case:
- You can put rules on the plant firewall and the ranges do not overlap: routed subnet with firewall rules. Default choice.
- Plant access is S7 to the CPU only and the machine subnet must not be routed: second interface, CP 1243-1 on the S7-1200, 1514SP-class CPU or CP/CM on the ET200SP.
- Machine and plant ranges overlap, or policy requires translated addresses: S615-class 1:1 NAT switch in front of the subnet.
Questions asked
Does a two-port S7-1200 CPU give you two networks?
No. The two ports behind several S7-1200 CPUs are an internal switch on one interface, sharing one IP address. The test is in the interface properties: if both ports answer to a single address, it is one LAN. Real separation needs a CP 1243-1 on the S7-1200, or on the ET200SP the 1514SP-class CPU or a CP/CM module.
Can the plant read drive data without a route to the drives?
Yes. Move the values into the PLC program and let the plant read the CPU: the drive data lands in a data block and the poller reads that block over the S7 connection. Data flows upward through one endpoint while the drives keep no plant-side address, no rule and no translation.
Which part of IEC 62443 covers this segmentation?
IEC 62443-3-2 defines the zone and conduit workflow: assess the system, group assets with the same security requirements into zones, and let conduits carry the communication between them, with each channel's controls documented. A machine cell with a single filtered uplink is a textbook zone-and-conduit result.
Why do NAT switch rules fail when the initiator is wrong?
Because translation is direction-sensitive. A plant-initiated poll must target the translated address, and a machine-initiated push must be permitted outbound with its return session tracked through the translation. A rule built for the wrong direction drops the first packet or the reply even though the mapping itself is correct.
How do I keep the machine DHCP server off the plant network?
Prefer static addresses on the machine subnet; an addressing plan like the worked example needs no DHCP at all. If the cell requires it, scope the server to the machine range, put the uplink on a labeled port, and trace every cable before energizing the uplink. A single mispatched switch port is the failure mode that puts machine DHCP on the plant network.