Comparison
Why layered OT security controls protect a network path
Short answer
No single control secures an OT request path. Network reachability, session protection, user identity and host hardening each block a different failure, so commissioning verifies them in that order: route first, then TLS and certificates, then login and role permissions, then the hardened baseline. A host that answers a connection proves none of the other three.
On this page
- Why does one security control never cover an OT request path?
- What do IEC 62443 zones and conduits ask of a plant network?
- Which four layers does a commissioning test verify, and in what order?
- What does each layer prove, and what does it leave open?
- How do you commission all four layers on one real path?
- What does hardening change on a Linux-based OT host?
- Does the same sequence apply when the OT host is a virtual machine?
- Which layer fails first in practice, and how do you isolate it?
- Questions asked
A plant network request crosses several controls between its origin and the service it targets, and no one of them carries the whole load. Reachability, encryption, identity and hardening answer four different questions about the same packet, so a commissioning test that stops at the first one proves almost nothing about the rest. This article compares the four layers on one OT request path: what each prevents, what it cannot prove, and the order in which to verify them — route first, session second, identity third, hardened baseline last. The counts in the worked example are illustrative; the ports and the standard behind the sequence are not.
Why does one security control never cover an OT request path? #
One control cannot secure an OT request path because reachability, encryption, identity and permission are four separate checks: a host that answers a TCP connection may still refuse TLS, accept any account, or let an unauthorized user write a setpoint.
Trace the path before judging it. A data collector, an engineering station, a remote diagnostic session and a controller-to-server exchange each run over their own interfaces and zones; for every exchange, write down the initiating system, the destination service, the zones in between and the return path. Four independent failure points sit along it: the route can drop the packet, the service can refuse to encrypt, the directory can reject the account, and the role can deny the write. A test that confirms the first — a ping or a TCP connect — says nothing about the other three, which is why one firewall rule, one certificate or one successful login is never a completed commissioning. Segmentation limits which networks may attempt a connection at all; the controls behind it decide what that connection may carry and who may drive it.
What do IEC 62443 zones and conduits ask of a plant network? #
IEC 62443 asks you to divide the plant into zones — assets with common security requirements — joined only by named conduits, so every request between areas crosses a boundary that someone has sized, documented and decided to allow.
In the standard's terms, a zone is a set of assets, physical or functional, that share common security requirements; published summaries of IEC 62443-3-3 describe exactly that grouping. A conduit is the controlled connection between two zones. In practice, the zone boundary is where your firewall or router policy lives, and the conduit is the rule pair you can point at: source zone, destination service, allowed protocol. Segmentation and the host firewall are complementary, not alternatives — the zone policy decides which networks may even try to reach a host, while the host's own firewall and address allowlist narrow the same path further to approved peers. An address allowlist admits a machine; it does not authenticate a user or encrypt application data. The standard's defense-in-depth intent is precisely that no single mechanism is trusted to do all three.
Which four layers does a commissioning test verify, and in what order? #
Verify four layers in a fixed order — network reachability, session protection, user identity and permission, host hardening — because each test needs the previous one to pass, and each one proves something the other three cannot.
Run them as a sequence, not a menu. Reachability comes first because every later test needs a working route. Session protection comes second because an unencrypted login would hand credentials to anyone on the wire. Identity and permission come third because authorization is only meaningful inside a protected session. The hardened baseline comes last as the audit that the approved path is the only one left open. Worked in that order, each result also isolates the next failure: if the route passes and the TLS handshake fails, you are fixing certificates, not switch ports. Verify identity in two steps — the account can log in, then an out-of-role operation is denied. A successful login alone is not evidence of correct authorization; the denial test is.
What does each layer prove, and what does it leave open? #
Each layer closes exactly one class of failure, and the comparison below shows the failure each prevents, the question it cannot answer and the evidence commissioning records for it.
| Layer | Failure it prevents | What it cannot prove | Commissioning evidence |
|---|---|---|---|
| Network reachability — zones, conduits, host firewall, address allowlists | Unapproved networks and hosts reaching the service | That the session is encrypted, or that a user is who they claim | TCP connect from the approved source succeeds; a non-approved source is refused |
| Session protection — TLS and certificate trust | Eavesdropping, tampering and impostor endpoints on the wire | Who is connecting, or what that connection may do | Handshake succeeds against the plant PKI certificate; a self-signed test certificate is rejected |
| Identity and permission — directory login, role-based access | Anonymous access and out-of-role operations | That unused services are closed or patched | Login works for the intended role; an out-of-role write is denied |
| Host hardening — baseline, minimal services, patching | Services and defaults no approved path needs | That the traffic left enabled is protected | Only approved listeners remain open; every baseline deviation is documented |
Read the third column as the reason the other three rows exist: every layer's blind spot is another layer's job, which is what defense in depth means on a single request path. Two of the layers — protected sessions and directory-backed access — are also the pair most often configured on the same host, and a walkthrough of configuring encryption and access on a SCADA server shows that separation kept clean in practice.
How do you commission all four layers on one real path? #
Work one path end to end with real addresses and ports — say a collector at 10.20.5.41 reading an OPC UA server at 10.20.8.10 on TCP 4840 — and record a pass or fail at each of the four layers before moving to the next.
The example uses illustrative addresses and counts; the ports are the documented defaults for the two protocols. A data collector at 10.20.5.41 must read an OPC UA server at 10.20.8.10 and publish to an MQTT broker at 10.20.8.30, both across one conduit. The services use their default ports, TCP 4840 for OPC UA and TCP 8883 for MQTT over TLS; both can be reconfigured, so every check runs against the configured port, not the default. The approved path map contains exactly two rules: 10.20.5.41 to 10.20.8.10:4840, and 10.20.5.41 to 10.20.8.30:8883. Before hardening, a scan of the server host lists 17 listening TCP ports; the approved path needs 1 of them, so hardening closes 17 − 1 = 16, and the same scan afterwards lists 1.
Then the four tests, in order:
- Route: a TCP connect to 10.20.8.10:4840 from 10.20.5.41 succeeds; the same connect from a second collector at 10.20.5.42 fails. Record both results — the refusal is part of the pass.
- Session: the OPC UA client completes a TLS handshake with a certificate signed by the plant PKI. Repeated with a self-signed certificate, the server rejects the connection; that rejection proves validation is enforced, not merely available. Resolve a certificate failure by fixing the trust chain, never by disabling validation on a system headed for production.
- Identity: the collector's service account logs in against the directory, a read succeeds, and a write to the same server is denied by role.
- Hardening: the rescan shows the single approved listener, the conduit holds exactly the two rules in the map, and every deviation from the applied baseline is written down with an owner and a reason.
The arithmetic is deliberately small — two rules, one listener left of 17 — but the habit scales: every port or rule that exists without a line in the path map is a finding, not a convenience.
What does hardening change on a Linux-based OT host? #
Hardening removes every port, service and device path the approved map does not name: a published baseline such as the CIS benchmark for SUSE Linux Enterprise Server 15 turns that principle into checkable configuration items you can apply and audit.
A CIS benchmark is prescriptive configuration guidance for one named OS release, and companion benchmarks exist for the components plant hosts commonly run, such as container runtimes, web servers and databases. Applying one means enabling the smallest set of listening services, removing default accounts, enforcing a USB device policy with allow- and block-lists, switching on file integrity monitoring, and keeping the kernel and packages inside the vendor's update cycle. Two habits keep the baseline honest: record every item you deliberately deviate from, with a reason, and re-run the audit with each patch cycle, because updates and added roles can turn services back on. Hardening is verified last in the commissioning sequence, but it is maintained continuously.
Does the same sequence apply when the OT host is a virtual machine? #
The same four layers apply unchanged to a virtualized SCADA or historian host, with the hypervisor added as one more boundary that a conduit rule must cover, and the guest's configuration evidence collected the same way.
The guest still needs its route, its TLS, its directory login and its baseline. The virtual switch and the host's management interface need the same conduit treatment as any physical zone crossing, and snapshot, clone and migration features deserve a line in the path map too, since a restored snapshot can quietly resurrect a service the baseline removed. Performance, though, is a separate discipline from security: a hardened guest on an undersized or misconfigured host still fails the plant, and tuning it follows its own checklist — fixing slow automation VMs on a Windows 11 host walks through that side of the same machine.
Which layer fails first in practice, and how do you isolate it? #
Note the last hop that passed and the first that failed, because the failure's position in the sequence names its layer: a connection that opens but refuses TLS is a certificate problem, never a routing one.
Keep the results in a small table of their own — hop, expected result, actual — and the diagnosis reads straight off it. A TCP connect that fails points at routing, zone policy or the allowlist. A connect that succeeds while the TLS handshake fails points at the certificate chain or at a security mode never enabled for that specific service: a protocol that supports TLS does not mean the connection you built uses it. A handshake that succeeds while login fails points at the directory — account state, bind settings or password. A login that succeeds while the operation is denied is usually not a fault at all but the role boundary doing its job; check the assignment, not the wall. If an unapproved host still connects after apparently successful commissioning, the allowlist rule or its netmask is wrong, and the reachability layer gets re-tested first. Resist the shortcut of opening extra ports to make a test pass: record the change, fix the cause, and close the temporary rule the same day.
Questions asked
Can OPC UA and MQTT share one edge device on this kind of path?
Yes. OPC UA and MQTT can run side by side on the same edge host when the installed software supports both roles and the host has the resources for both. Each protocol still gets its own security decision: the OPC UA session's encryption and certificate trust are configured separately from the MQTT broker connection, so commissioning treats them as two protected sessions from one address, each with its own pass or fail.
Is an IP allowlist the same as authentication?
No. An allowlist decides which machine may open a connection; authentication decides which person or service account may use it afterwards. A machine on the list can still run compromised software, and a valid account on a non-approved network cannot even reach the service. The commissioning evidence differs too: a refused TCP connect proves the allowlist, a denied login proves the directory, which is why the two tests stay separate steps.
What security level does IEC 62443 require for my plant?
The standard sets no single level for every plant. IEC 62443-3-3 defines graduated security levels, and the target for each zone comes from your own risk assessment of that zone's assets and the consequences of losing them. Choose the level per zone and conduit first; it then becomes the acceptance criterion your four-layer commissioning tests are measured against.
How often should the path map and rules be re-verified?
Re-run the four tests whenever a change touches the path: an operating system or application patch, a new client address, a moved server, or a replaced firewall or switch. Tie the cadence to the patch cycle so the checks run at least as often as the software underneath them changes, and keep the expected results from the last run so each re-check is a comparison, not a fresh investigation.