SEC-04 / Data Foundations

Unified Namespace (UNS) explained for mid-market manufacturers

A Unified Namespace is a single, structured, real-time view of plant data that every system publishes to and reads from, usually over MQTT. It works when topic structure and write access are governed; without that, it becomes a fast way to send the wrong command to the wrong machine.

7 min read Updated OT engineersCIOsPlant managersCISOs

Key takeaways

  • A UNS is an architecture pattern, not a product: one shared, structured hub of current plant state that systems publish to and subscribe from.
  • Most UNS designs use an ISA-95 style hierarchy (enterprise, site, area, line, cell) and MQTT, often with Sparkplug B, now published as ISO/IEC 20237.
  • The governance question that matters most is who can publish, especially command and setpoint topics.
  • Topic structure needs an owner and a change process, the same way a tag database or PLC program does.
  • For a single plant with one historian and a few consumers, a UNS can be overkill. Start with a governed historian and grow into it.

The Unified Namespace, or UNS, has become one of the most talked-about ideas in industrial data. Strip away the conference slides and it is a simple pattern: instead of wiring every system to every other system, you give the plant one structured, real-time hub. Every producer publishes there. Every consumer reads from there. The structure of that hub mirrors how you think about the plant.

That simplicity is also the risk. A hub that everyone can read from is useful. A hub that everyone can write to is dangerous. This page explains the pattern in plain terms, then spends most of its time on governance, because that is where mid-market UNS projects succeed or fail.

What a UNS is, and what it is not

A Unified Namespace is:

  • A single source of current state. The latest value of each data point, plus events, published in one place.
  • Structured. Topics follow a hierarchy that people recognize, usually based on ISA-95.
  • Event-driven. Systems publish on change. Consumers subscribe. No polling of every system by every other system.
  • Open to many consumers. MES, historian, dashboards, analytics and AI models all read from the same hub.

A UNS is not:

  • A product. It is an architecture. You build it from a broker, edge connectors and conventions.
  • A historian. It holds current state. Something still has to store history.
  • A security boundary. The broker is a critical asset inside your zones, not a firewall.

Point-to-point vs hub

  Point-to-point            Unified Namespace

  ERP----MES                ERP   MES   AI
   | \  / |                   \   |   /
   |  \/  |                  +--------+
   |  /\  |                  | broker |
   | /  \ |                  +--------+
  SCADA--Historian             /   |   \
                           SCADA  Hist  Edge

With five systems, point-to-point gives you up to ten integrations to maintain. With a hub, you have five connections and one structure to govern.

The ISA-95 hierarchy

ISA-95, published internationally as IEC 62264, is the standard for integrating enterprise and control systems. It describes a role-based equipment hierarchy that most UNS designs borrow for their topic structure:

Level Example Example topic segment
Enterprise Your company acme
Site A plant acme/leeds
Area A department or hall acme/leeds/packing
Line or process cell A production line acme/leeds/packing/line3
Work cell or unit A machine or station acme/leeds/packing/line3/filler

ISA-95 also defines the functional levels (Level 0 to Level 4) that the Purdue model popularized, with the standard mainly addressing the interface between manufacturing operations (Level 3) and business systems (Level 4). Using its vocabulary means MES, ERP and OT people can read the same topic tree without translation.

Below the equipment level, teams typically add namespaces for data type, for example .../filler/telemetry, .../filler/state, .../filler/oee, and, critically, .../filler/cmd.

MQTT and Sparkplug B

Most UNS implementations run on MQTT, a lightweight publish/subscribe protocol standardized by OASIS (MQTT 5.0 is the current OASIS Standard). MQTT moves messages. It deliberately does not define what topics should look like or what payloads contain.

Sparkplug fills part of that gap. It is an Eclipse Foundation specification, and Sparkplug 3.0 was published as ISO/IEC 20237 in November 2023. It defines three things on top of MQTT:

  • A topic namespace for edge nodes and devices.
  • A payload format for industrial process variables, with metrics, data types and timestamps.
  • Session state management, so consumers know when an edge node or device goes offline (birth and death certificates), instead of trusting stale values.

Sparkplug also defines command message types (NCMD and DCMD) that let a host application write to edge nodes and devices. That is useful. It is also exactly the thing you need to govern.

Note: Sparkplug's own topic structure is organized by group, edge node and device rather than the full ISA-95 tree. Many teams therefore use Sparkplug for edge data and a separate ISA-95 structured layer for contextualized data. That is a design choice, not a defect. Write down which one you chose and why.

OPC UA often sits underneath: edge connectors read from OPC UA servers on PLCs or SCADA, then publish into the UNS.

Governance: topic structure

A topic tree is a data model. It needs the same treatment as a tag database or a PLC program.

  • Named owner. One person or small group approves new branches and naming changes.
  • Written convention. Levels, separators, casing, units, timestamp rules, payload schema.
  • Change control. Renaming a topic breaks every subscriber. Treat it like a PLC change: review, test, schedule.
  • Schema validation. Reject or quarantine payloads that do not match the agreed format.
  • Context at the right level. Product, batch and shift should appear in a consistent place, not improvised per line.

Governance: write access, the part that matters most

In a typical UNS, the same broker carries telemetry from machines and commands to machines. If access control is loose, a dashboard, an analytics script or an AI agent can publish a setpoint, start a sequence or acknowledge an alarm. That is a safety issue, not just a data issue.

Rules worth adopting:

  1. Default deny. No client publishes anywhere unless an ACL allows it.
  2. One identity per system. No shared credentials across gateways, dashboards and scripts.
  3. Publishers own their branch. The filler's edge node publishes only under the filler's topics.
  4. Commands are a separate class. Command and setpoint topics have their own ACLs, a very short list of approved publishers, and logging.
  5. AI does not write directly. AI outputs go to an advisory topic. A control system or an operator decides whether to act. Closed-loop AI control is a separate, risk-assessed change.
  6. The PLC still decides. Interlocks, limits and safety functions stay in the controller and the SIS, regardless of what arrives over MQTT.
  7. Broker in the right zone. Place the broker deliberately, document it as a conduit endpoint in your zones and conduits model, and bridge only selected topics to IT or cloud.

Watch out: Many brokers ship with permissive defaults for convenience. Proof-of-concept UNS builds often go live with those defaults still in place. Audit the ACLs before the pilot becomes production.

Who does what

Task Typical owner
Topic convention and schema OT data owner or controls lead
Broker operation and patching OT infrastructure (with IT support)
ACLs and credentials OT security lead
Approving command publishers Process owner plus OT security, via MOC
Data meaning and quality per area Area or process owner

Map these to your AI roles and RACI so AI use cases that consume or publish to the UNS have clear accountability.

When a UNS is overkill

A UNS pays off when there are many producers, many consumers and frequent change. For a mid-market plant, be honest about whether that is you.

Situation Reasonable choice
One plant, one historian, two or three consumers Governed historian with an asset model. Skip the UNS for now.
One plant, MES rollout, several new consumers coming Start a small broker-based UNS for one area.
Several plants, cross-site analytics and AI UNS per site with a common convention, bridged selectively.
OEM skids that only speak their own protocol Edge connectors into a historian first; UNS later if needed.

Plant-floor reality: A UNS does not fix bad data. If tags are mislabeled, clocks drift and nobody owns the meaning of a value, a UNS distributes that confusion faster. Fix the data quality basics first.

Common mistakes

  • Treating the broker as "just middleware" and leaving it unpatched and unmonitored.
  • Copying a topic structure from a demo and changing it after twenty subscribers depend on it.
  • Letting every client publish to # because it was easier during commissioning.
  • Assuming the UNS is the historian and losing history during an outage.
  • Building a UNS before there is a use case that needs it.

Where to start

  1. Write down the use cases that need real-time, multi-consumer data. If there are fewer than three, start with a governed historian path (see moving OT data to a cloud historian).
  2. Draft a one-page topic convention based on ISA-95 levels for one area.
  3. Stand up a broker in the right zone with default-deny ACLs.
  4. Connect one line. Publish telemetry and state only. No command topics yet.
  5. Add a historian subscriber so nothing depends on the broker for history.
  6. Only then decide, through MOC, which systems may publish commands and on which topics.

Effort: a single-area pilot is often a few weeks of controls engineering time. Most of the ongoing effort is governance: owning the convention, reviewing ACLs and handling change requests. Score yourself with the industrial data readiness checklist before scaling to more lines.

What good looks like

A short written topic convention that new integrators follow without asking. A broker that is patched, monitored and documented in the zone model. ACLs that a reviewer can read in ten minutes. Command topics with a named list of publishers. And AI use cases that read from the UNS freely but write only to advisory topics until someone has signed off on more.

Sources and further reading

  1. ISA-95 Standard (ISA)
  2. Eclipse Foundation announces Sparkplug as an international standard (ISO/IEC 20237)
  3. Sparkplug Specification (Eclipse Foundation)
  4. MQTT Version 5.0, OASIS Standard
  5. NIST SP 800-82 Rev. 3, Guide to Operational Technology (OT) Security

This guide is general information for manufacturers, not legal advice. Confirm regulatory obligations with qualified counsel.

FAQ

Frequently asked questions

What is a Unified Namespace in simple terms?

It is a single place where every system in the plant publishes its current data, organized in a shared hierarchy such as enterprise, site, area, line and machine. Instead of point-to-point integrations between MES, ERP, SCADA and analytics, each system talks to the hub. Consumers subscribe to what they need. It is usually built on an MQTT broker.

Is Sparkplug B required for a Unified Namespace?

No. Sparkplug B is an Eclipse Foundation specification, published as ISO/IEC 20237 in 2023, that defines a topic namespace, payload format and session state for MQTT in industrial use. Many UNS builds use it for device and edge data, and plain MQTT with JSON for higher-level, ISA-95 structured topics. Pick based on your tools and team.

Can a Unified Namespace replace a historian?

Not by itself. A UNS shows current state and events. A historian stores time-series history for trends, investigations and model training. In most designs a historian subscribes to the UNS and records it. Treat them as complementary, and keep the historian as the record you rely on for quality and incident investigations.

Who should be allowed to publish to the UNS?

Only named systems, each with credentials scoped to its own branch of the topic tree. Command and setpoint topics should be restricted to a very small set of approved publishers, documented in your zone model and changed only through MOC. People and dashboards should not publish commands directly.

Related guides

Know where your plants stand

Score your AI governance, OT security and data readiness in about 10 minutes each. Results stay in your browser.