In most plants, the boundary between IT and OT is protected by a firewall with a DMZ behind it. The firewall’s security depends on the correctness of the rule base; a single faulty “any” rule makes the boundary two-way. Some plants do not accept this risk and want the boundary to be impossible to open through a configuration error. A data diode (unidirectional gateway) is the hardware-based boundary used for this need, physically carrying data in one direction only. Below, we describe how it works, which data is carried and how, how two-way needs are solved, and when it is a better choice than a firewall.
What a data diode is and how it works
The essence is simple: the sending side has only an optical transmitter, and the receiving side only an optical receiver. There is no fiber or transmitter/receiver pair to carry light in the reverse direction. Not a single bit can physically be transmitted from the receiving side to the sending side; neither a configuration error nor a software vulnerability can open this direction. There is no reverse direction.
The direct consequence of this constraint: protocols that require a handshake and acknowledgements (ACK), such as TCP, cannot pass through the diode directly. Commercial products therefore use a three-part structure:
- Sending agent (OT side): Talks to sources in OT with normal protocols; reads data from the historian, receives syslog, collects files from a designated folder.
- One-way hardware: Transfers the data produced by the agent over the one-way link, using a product-specific transport format.
- Receiving agent (IT or DMZ side): Reconstructs the incoming data and serves it like a server to applications in IT, as a historian replica, OPC server or file share.
The application in IT connects not to the system in OT but to the copy on the receiving side. This is the “no direct sessions” principle of OT DMZ design, enforced in hardware.
Protocol replication in one-way transfer
What can be carried through the diode is determined by the “connector” software the product supports. The first question in product selection is therefore not hardware speed but whether the required protocols are supported. A common mistake is buying the diode and looking at the connector list afterwards.
| Data type | On the sending side | On the receiving side |
|---|---|---|
| Historian data | Reading from the historian | Historian replica |
| OPC DA/UA tags | Reading as an OPC client | Publishing as an OPC server |
| Syslog, SNMP traps | Receiving as UDP | Forwarding to the SIEM |
| Files (reports, logs) | Folder monitoring | Writing to a target folder |
| Screen capture/monitoring | Screen capture | Read-only viewing |
Two approaches are seen in historian replication. In the first, the same vendor’s historian runs on the receiving side, and the diode uses a connector that carries the vendor’s own replication stream. In the second, the sending agent transfers tag values in a generic format and the receiving side writes them to a different database. The first better preserves tag metadata, quality flags and backfilling; the second offers flexibility. Historian selection is covered in the article choosing a historian.
Because there is no return channel, loss management sits at the centre of the design. Products generally use forward error correction (FEC), packet retransmission and sequence numbers. Loss and sequence error counters on the receiving side are monitored; when a gap is detected in replication, a scheduled job is defined for the sending side to resend a specific time range. This resend is not triggered by a request from the receiver. It cannot be.
Publish–subscribe models such as MQTT or OPC UA Pub/Sub are conceptually closer to one-way flow. However, because MQTT runs over TCP, it cannot pass through a diode directly; and although OPC UA Pub/Sub’s UDP-based transport does not wait for acknowledgements, in practice these flows are also carried via connectors supported by the product, with agents on both sides of the diode. For the general criteria for choosing a protocol, see the article OPC UA or MQTT.
Two-way needs: patching, remote support, recipes
In data diode projects, the flows that really need to go towards OT take up the most time. If a second path is opened for these flows after the diode is installed, that path now determines the security of the design.
- Patches and antivirus signatures: Options include bringing scanned updates in via a controlled portable media station, transferring only specific file types with a second diode in the reverse direction, or a temporary connection physically connected only during the maintenance window. A reverse diode does not validate the content of incoming data; file validation and scanning are done separately.
- Remote support: Remote access by the machine builder or central engineering does not go through the diode. A separate connection is usually designed: closed by default, requiring physical or procedural approval from site to open, time-limited and logged. Its design is in the guide machine builder remote access.
- Recipes and production orders: Transferring work orders or recipes from ERP or MES to OT is a need the diode cannot solve directly. Options include transferring production orders to the MES in OT manually or via controlled media, a second channel in the reverse direction that allows only a specific message format and validates its content, or a tightly configured firewall instead of a diode for this flow.
Comparison table with a firewall
| Criterion | Firewall + DMZ | Data diode |
|---|---|---|
| Direction guarantee | Rule-dependent, provided in software | Physical, no reverse direction |
| Effect of a configuration error | Can open the boundary | Cannot open the reverse direction |
| Supported flows | Almost any protocol | Protocols with a connector |
| Request–response, writes | Possible | Not possible |
| Operating burden | Rule management, auditing | Connector management, second path |
| Cost items | Device, licence, rule maintenance | Hardware, connector licences, agent servers on both sides |
The most important row in the table is the second. A correctly configured firewall is a strong boundary; but its security depends on the rule base being kept correct for years. Shifts change, suppliers change, “temporary” rules become permanent. For one-way flows, a data diode removes this dependency; in return, it restricts flexibility and pushes two-way needs onto separate paths that must be designed carefully.
Which plants is it preferred for?
A data diode is not the right investment for every plant’s industrial network and OT infrastructure. Situations that support choosing one:
- Weight of consequences: Plants where an attack on OT could result in loss of life, environmental damage or large-scale service disruption: power generation, water and wastewater, chemicals, critical infrastructure.
- Regulatory expectation: If industry regulation or a customer specification calls for a hardware-based one-way boundary for the most critical zones.
- Flow profile: If the flows going to IT are mainly monitoring and reporting, and the flows needed towards OT are few and predictable.
- Target security level: If a high target security level has been set for a zone under IEC 62443 and only one-way data is needed on the conduit. How targets are set is covered in the article IEC 62443-3-3 security levels.
In discrete manufacturing plants with frequent recipe changes and heavy MES–ERP interaction, where many two-way flows towards OT are needed every day, a well-designed firewall + DMZ structure is often more balanced. There is also a middle way: the diode is placed only at the boundary of the most critical zone (for example a safety instrumented system or the main control system), and the rest of the plant is separated with a firewall.
Design and operating considerations
- Flow inventory first: Before choosing a diode, list all flows going to IT and coming into OT. For every flow coming into OT, write down the solution without the diode; if there is a flow that cannot be solved, it is decided before the project starts.
- Managing the agent servers: The sending agent sits in OT and the receiving agent in the DMZ or IT; both need patching, backups and monitoring. The sending agent is managed by OT, from inside OT.
- Time synchronisation: Because NTP requires request–response, it does not pass through the diode. The OT side cannot take time from servers in IT; it needs its own source inside OT (for example a GPS-based time server). Timestamps are generated on the sending side, the receiving side synchronises from an independent source, and the difference between the two clocks is monitored. This item is usually forgotten during commissioning.
- Monitoring: Connector status, loss counters, last successful transfer time and replication delay are connected as alarms to the monitoring system in IT. Replication that stops silently leads to missing data in reports and wrong decisions. Nobody notices.
- Physical integrity: A single cable running around the diode invalidates the whole design. It is checked periodically on site that there are no other connections at the boundary.
- Capacity: In addition to bandwidth, the number of tags the agents can process per second and file size limits are checked.
NIST’s OT security guide SP 800-82 Rev. 3 covers unidirectional gateways among the options for IT/OT boundary protection and is a good starting point for an evaluation. If you want to decide between a firewall and a data diode for your critical boundary based on a flow inventory, see our OT cybersecurity and segmentation and IT/OT integration consulting pages.
Frequently asked questions
Is a firewall still needed once a data diode is installed?
Yes. A data diode only provides the one-way boundary between IT and OT. Separation between zones inside OT, cell firewalls and controlled access paths outside the diode, such as a time-limited remote support connection, are still managed with firewalls. The diode does not replace the firewall; it takes over its role at the most critical boundary.
How do request–response protocols such as OPC UA or Modbus pass through a diode?
The protocol itself does not pass. Intermediary software on the OT side reads the data like a client and sends it over the one-way channel; the intermediary on the IT side publishes the same data like a server. The application in IT connects to this copy. That is why writing to or sending commands to a device in OT from the IT side is not possible via this path.
What happens if data is lost in a data diode?
Because there is no return channel, the receiver cannot request a lost packet again. Products compensate for this with methods such as forward error correction, sending packets multiple times and detecting loss with sequence numbers. Loss counters on the receiving side are monitored; if a gap appears in historian replication, a resend is triggered by a scheduled job on the sending side.
Can software-based one-way solutions replace a data diode?
They do not provide the same assurance. In solutions that achieve one-way behaviour with a firewall rule or software, the reverse direction is physically possible; it can be opened by a configuration error or vulnerability. What sets a data diode apart is that the reverse direction is impossible at hardware level, for example because there is no optical receiver at all. When evaluating a product, it should be asked explicitly how this separation is achieved.