Most OT networks that look segmented on a diagram are flat in practice: one routed path, one dual-homed historian or one vendor VPN is enough to connect corporate IT to the controllers. The Purdue model tells you where assets sit, but it does not tell you which flows are allowed or how strong each boundary must be; IEC 62443 zones and conduits do. This guide is the design and enforcement manual: how to turn a Purdue diagram into a zone and conduit model, how to specify each conduit, which enforcement technology fits which boundary, how to migrate a live plant without stopping production, and how to prove the segmentation works. It is for OT architects, network engineers, integrators and security teams. The OT Security Guide introduces the Purdue levels and the industrial DMZ, and the IEC 62443 Guide explains the standard; this page goes deeper on the network design itself.
Key takeaways
- Purdue describes, zones and conduits prescribe. Use Purdue levels as a shared vocabulary, then define zones by consequence and trust and specify every conduit: endpoints, protocols, direction, owner and enforcing device.
- Some separations are mandatory. IEC 62443-3-2 says business and control system assets shall be separated and safety-related assets shall be separated; temporarily connected devices, wireless and externally connected devices should be separated.
- Nothing crosses from IT to OT directly. NIST SP 800-82 Rev. 3 calls for a DMZ with firewalls or unidirectional gateways so traffic does not pass directly between corporate and OT networks; every session terminates in the DMZ.
- Segmentation must fail safe. Essential functions must keep working if a boundary firewall fails closed or the site islands, and a denial of service on the control network must not stop safety instrumented functions.
- Monitor before you enforce. Learn the real flows passively, write the conduit rules from evidence, run them in log-only or alert mode, then enforce in a maintenance window with a tested rollback.
Purdue levels and their limits as a design tool
The Purdue Enterprise Reference Architecture places the physical process at Level 0, basic control (PLCs, RTUs, safety logic solvers) at Level 1, supervisory control (HMIs, SCADA, engineering workstations) at Level 2, site operations (historians, OT domain services) at Level 3, an industrial DMZ at Level 3.5 and the enterprise at Levels 4 and 5. It is still the clearest way to explain trust boundaries to engineers and auditors. Its limits as a design tool are practical.
- Levels are not boundaries: nothing in the model says how many zones Level 2 should be split into, or that two production lines at the same level should not talk to each other.
- Modern flows cut across levels: IIoT sensors reporting to cloud platforms, cellular-connected remote sites, vendor remote support built into products and OPC UA servers on controllers do not respect a strict hierarchy.
- Diagrams drift: plants accumulate dual-homed hosts, unmanaged switches and temporary links that the Purdue diagram never shows.
- It has no strength dimension: it does not say how well a boundary must resist an attacker. IEC 62443 target security levels (SL-T, from SL 1 casual violation to SL 4 sophisticated, well-resourced, highly motivated attackers) do.
The practical approach is to keep Purdue levels as labels on the diagram and use zones and conduits as the actual design.
Designing zones
IEC 62443 defines a zone as a grouping of logical or physical assets based on risk or other criteria, such as criticality, operational function, location, required access and responsible organisation. A conduit is a logical grouping of communication channels with shared security requirements that connects two or more zones. Zoning follows the 62443-3-2 risk workflow: identify the system under consideration, perform an initial risk assessment, partition into zones and conduits, perform a detailed assessment where initial risk exceeds tolerable risk, document the requirements and SL-T per zone and conduit in a Cybersecurity Requirements Specification, and obtain asset owner approval. See the IEC 62443-3-2 OT Risk Assessment Guide.
| Zone | Typical contents | Purdue level | Design rule |
|---|---|---|---|
| Enterprise | ERP, email, office IT, corporate identity | 4 and 5 | No route into control zones; reaches only DMZ services |
| Industrial DMZ | Historian replica, remote access broker, jump hosts, patch and anti-malware staging, file transfer | 3.5 | Every IT to OT session terminates here; no service in the DMZ is a pass-through |
| Site operations | Site historian, OT domain controllers, OT monitoring sensors and collectors, backup | 3 | Separate OT identity domain; reachable from DMZ only on defined ports |
| Cell or area control (one zone per line, unit or process area) | HMIs, SCADA servers, PLCs, drives | 1 and 2 | Cells do not talk to each other except through defined conduits |
| Engineering | Engineering workstations, project repositories, licence servers | 2 or 3 | Only zone allowed to send program and configuration changes to controllers |
| Safety | Safety instrumented system logic solvers and their engineering station | 1 | Separate zone (shall); read-only data to the control zone where possible |
| Wireless and remote | Wireless HART and Wi-Fi gateways, cellular RTUs, remote sites | Varies | Separate zones (should), treated as untrusted until the gateway |
| Transient devices | Vendor and contractor laptops, USB scanning kiosks | Varies | Dedicated zone or port (should), never directly on a control VLAN |
- Draw by consequence: assets whose compromise has the same worst-case consequence and the same trust go together. A boiler management system and a packaging line do not share a zone just because they share a VLAN.
- Keep zones small enough to contain an incident: one zone per line or process unit lets you isolate a compromised cell and keep the rest of the plant running.
- Do not put people in the control zone: office PCs, printers and general purpose email or browsing have no place in Level 1 and 2 zones; IEC 62443-3-3 includes a requirement restricting general purpose person-to-person communication.
Specifying conduits
A conduit is not a firewall rule; it is a documented contract that the enforcing device then implements. Write one record per conduit.
| Field | Example |
|---|---|
| Conduit ID and owner | C-07, owned by site OT engineering |
| Zones connected | Cell 3 control to site operations |
| Channels | Historian collector to PLCs on EtherNet/IP TCP 44818 (read services); OPC UA TCP 4840 from historian interface node |
| Direction and initiator | Initiated from site operations only; no inbound sessions to site operations from the cell |
| Enforcing device and mode | Industrial firewall FW-3, default deny, protocol inspection allowing reads only |
| Identity and authentication | Named service account for the interface node; certificates where the protocol supports them |
| SL-T and countermeasures | SL 2; inspection, logging, monitoring sensor on SPAN |
| Failure behaviour | If FW-3 fails closed, cell continues to run; historian buffers data |
| Review date and change reference | Annual, or on any change to the cell or remote access |
Typical channels to name explicitly include Modbus TCP 502, Siemens S7 TCP 102, EtherNet/IP 44818 and 2222, OPC UA 4840, DNP3 20000, historian ports and RDP 3389 for any remote sessions. Anything not listed in a conduit record is denied.
Choosing enforcement technology
| Technology | Strength | Limitations | Best fit |
|---|---|---|---|
| VLANs with switch or router ACLs | Cheap, already present | Stateless or weakly stateful; misconfiguration common; no protocol awareness | Separating devices within a zone, low SL-T boundaries |
| Stateful firewall | Default deny, logging, well understood | Port-level only: cannot tell a Modbus read from a write | Enterprise to DMZ, DMZ to site operations |
| Industrial firewall with deep packet inspection | Can allow or block by protocol function (read, write, program download, mode change) | Must support each protocol; adds latency; needs OT skills to tune | Conduits into control and engineering zones |
| Unidirectional gateway or data diode | Hardware-enforced one-way flow | No remote writes or interactive access; replication software must be supported for each data source | High-consequence sites sending historian or alarm data outward. See the Data Diodes and Unidirectional Gateways Guide |
| Host-based firewalls and microsegmentation agents | Fine-grained per host | Rarely supported on controllers or vendor-validated HMIs | Windows servers in Level 3 and the DMZ. See the Network Segmentation Guide |
| Physical air gap | No network path | Data still moves by USB and laptops; often eroded over time | Safety systems and a few isolated processes, with transient device controls |
Decision guidance
- IT to OT boundary: always a DMZ with two firewall layers or a firewall plus unidirectional gateway; never a single router ACL.
- Control and safety zones: protocol-aware filtering where writes or program changes are possible; diodes where only outbound data is needed.
- Small site: one well-configured industrial firewall with a DMZ leg, a separate safety network and brokered remote access covers most of the risk; spend effort on rules and monitoring rather than more boxes.
- Large or multi-site estate: standardise a reference zone model per site type, manage firewall policy centrally with change control, and treat the wide area network between sites as an external network with its own conduits.
Migrating a flat network without stopping production
- Discover real flows: deploy passive monitoring on SPAN or TAP ports for several weeks across shift patterns, batch runs and maintenance, so infrequent flows (month-end reports, backup jobs, vendor diagnostics) are captured. See the OT Network Security Monitoring Guide.
- Find the bridges: look specifically for dual-homed hosts (historians, engineering workstations, HMIs with a second NIC on the office network), wireless bridges, cellular modems and remote access tools installed by vendors.
- Build the DMZ first: move historian replication, remote access, patch staging and file transfer into the DMZ and remove every direct path from enterprise to Levels 0 to 3. This step gives the largest risk reduction.
- Insert boundaries in bypass or monitor mode: place firewalls inline in a maintenance window with an allow-all, log-everything policy, then compare logs with the conduit records.
- Tighten one conduit at a time: switch rules to default deny for a single conduit, with operations and engineering present and a timed rollback (keep the previous rule set ready to restore in minutes).
- Re-address only where needed: many controllers cannot change IP addresses without a project download; use layer 2 transparent firewalls or routed boundaries that avoid re-addressing where possible.
- Remove the old paths: disconnect second NICs, delete legacy VPN accounts and firewall exceptions, and update drawings.
- Test failure behaviour: confirm the plant keeps running if a boundary firewall fails closed or the site loses its WAN link, and that safety functions are unaffected.
Where identity fits
Segmentation decides which networks can talk; identity decides who and what may use each conduit. NIST SP 800-82 Rev. 3 recommends that OT accounts do not rely on corporate user accounts and that each network has its own authentication. In practice that means an OT domain with no trust to corporate Active Directory, named accounts with MFA for every human crossing the DMZ, and an inventory of the service accounts that conduits exist for: historian interfaces, OPC clients, backup agents and monitoring collectors. Those service accounts often hold broad rights across zones, so scope them to the conduit they serve and rotate their credentials where vendors allow. See the OT and ICS Identity and Access Guide and the OT Remote Access Guide.
Lessons from incidents
The Colonial Pipeline Ransomware Attack 2021 began with a dormant VPN account without MFA, and the confirmed attacker activity was in IT. Yet the operator shut down the pipeline as a precaution from 7 to 13 May 2021. The lesson for segmentation is that a boundary is only useful if you can prove it: operators need documented conduits, monitoring evidence that nothing crossed them, and an OT that can run while IT is isolated. Without that evidence, the safe decision is to stop.
Detection and measurement
- Conduit violations: denied connections at each boundary, especially any attempt from enterprise address space to Levels 0 to 2.
- New flows: monitoring alerts for a talker or protocol not in any conduit record.
- Bridges: hosts seen with addresses in two zones, new wireless or cellular devices, unexpected DHCP servers.
- Remote access: sessions that do not originate from the DMZ broker.
Metrics
- Percentage of observed flows covered by a documented conduit record (target 100%).
- Number of direct enterprise to control zone paths (target zero).
- Number of dual-homed hosts across zone boundaries (target zero).
- Percentage of conduits into control zones enforced by protocol-aware or unidirectional devices.
- Firewall rules reviewed in the last 12 months with a recorded business reason.
- Time to isolate a single cell zone during an exercise without affecting other cells.
Framework mapping
| Design element | IEC 62443 | NIST SP 800-53 Rev. 5 | NIST CSF 2.0 | Threat addressed (ATT&CK for ICS) |
|---|---|---|---|---|
| Zone partitioning and DMZ | 3-2 partitioning requirements; 3-3 FR5 Restricted Data Flow (network segmentation, zone boundary protection) | SC-7 Boundary Protection; SC-7(21) Isolation of System Components; SC-7(29) Separate Subnets to Isolate Functions | PR.IR-01 | T0886 Remote Services; T0866 Exploitation of Remote Services |
| Default deny conduits | 3-3 FR5 | SC-7(5) Deny by Default, Allow by Exception; AC-4 Information Flow Enforcement | PR.IR-01 | T0867 Lateral Tool Transfer |
| Documented flows | 3-2 Cybersecurity Requirements Specification | CA-9 Internal System Connections; CM-2 Baseline Configuration | ID.AM-03 | None (governance) |
| Brokered remote access | 3-3 FR1 and FR5 | AC-17 Remote Access | PR.AA-05 | T0822 External Remote Services; T0859 Valid Accounts |
| Boundary monitoring | 3-3 FR6 Timely Response to Events | SI-4 System Monitoring | DE.CM-01 | T0884 Connection Proxy |
| Fail-safe boundaries | 3-2 essential function constraints; FR7 Resource Availability | SC-3 Security Function Isolation | PR.IR-03 | T0814 Denial of Service |
Other mappings: ISO/IEC 27001:2022 A.8.22 Segregation of networks; CIS Controls v8.1 Control 12 Network Infrastructure Management; MITRE D3FEND D3-NI Network Isolation and D3-NTF Network Traffic Filtering.
Common failure patterns
- Zones drawn from existing VLANs: the diagram reflects how the network was cabled, not consequence or trust, so a single zone holds safety and non-safety assets.
- A DMZ that passes traffic through: firewall rules that let enterprise hosts reach Level 2 directly "via" the DMZ, or a jump host that is also on the corporate domain.
- Dual-homed historians and engineering stations: the most common bypass of every segmentation design, usually added for convenience and never removed.
- Allow-any rules from commissioning: temporary any-any rules added during start-up that become permanent because nobody owns the conduit.
- Vendor remote access outside the DMZ: a cellular router or remote desktop tool in the cabinet that creates its own conduit to the internet.
- No failure testing: a boundary firewall that, when it fails closed, stops HMIs or interlocks that depended on a flow nobody documented.
Practitioner checklist
- Run passive monitoring long enough to capture all operating modes before designing rules.
- Define zones by consequence and trust, with safety and business assets separated as 62443-3-2 requires.
- Write a conduit record for every flow: endpoints, protocol, direction, owner, enforcing device and SL-T.
- Build the industrial DMZ and remove every direct enterprise to OT path first.
- Use protocol-aware filtering or unidirectional gateways on conduits into control and safety zones.
- Remove dual-homed hosts, unmanaged wireless and cabinet modems, or bring them into a documented conduit.
- Enforce one conduit at a time in maintenance windows with a timed rollback.
- Test that essential and safety functions continue if a boundary fails closed or the site islands.
- Keep OT identity separate from corporate identity and scope service accounts to their conduits.
- Review rules and conduit records at least annually and after every change to remote access.
Standards and references
- ISA: ISA/IEC 62443 Series of Standards
- ISA Global Cybersecurity Alliance: Quick Start Guide, An Overview of the ISA/IEC 62443 Standards
- NIST: SP 800-82 Rev. 3 Guide to Operational Technology (OT) Security
- NIST: SP 800-53 Rev. 5
- NIST: Cybersecurity Framework 2.0
- MITRE ATT&CK for ICS: Remote Services
- MITRE D3FEND: Network Isolation
Related NHI Mgmt Group resources: OT Security Guide · IEC 62443 Guide · IEC 62443-3-3 and 4-2 System and Component Security Levels Guide · OT Remote Access Guide · Zero Trust Network Segmentation Playbook for Flat Networks