Unified Namespace (UNS) Architecture: Topic Design with the ISA-95 Hierarchy

How a Unified Namespace is built in the field: the ISA-95 topic tree, naming rules, separating raw and contextualised data, broker scaling and limits.
Technical summary
Technology
MQTT 3.1.1/5.0, Sparkplug B, ISA-95, OPC UA, Mosquitto, broker bridging
Topic
Designing a plant-wide, event-driven central data layer
Level
Advanced
Prerequisites
  • MQTT publish/subscribe and retained message concepts
  • Basic knowledge of the ISA-95 equipment hierarchy
In this article7 sections

In a factory, SCADA, the historian, MES, maintenance software, the quality system and reporting tools often want data from the same PLCs. With each new need, a new connection is opened; a few years later, nobody knows exactly which system connects to which PLC from where. A Unified Namespace (UNS) is an architectural approach that replaces this sprawl with a single, plant-wide, event-driven data layer. Below, from a field perspective, we describe the components of a UNS, topic design based on the ISA-95 equipment hierarchy, broker architecture, and the situations where a UNS is the wrong choice.

Where point-to-point integration breaks down

In the classic structure, each consumer pulls data directly from its source: SCADA reads the PLCs, the historian gets data from SCADA or a separate OPC server, the MES queries work order status from the PLC with its own driver, and the reporting tool connects to the historian database. In a structure where N systems talk to each other, the maximum number of possible connections is N×(N−1)/2; for ten systems, the example calculation gives 45 potential integrations.

In the field, this translates to:

  • PLC load: The same variable is queried separately by several systems. On older PLCs with limited communication resources, connection slots fill up and a new client cannot connect.
  • Inconsistent meaning: The same counter lives as “Production_Qty” in SCADA, “GoodCount” in the MES and under another name in reporting; units and reset logic are interpreted differently everywhere. At the end of the shift, two screens show two different numbers.
  • Cost of change: When the PLC program changes, an inventory of all integrations is needed to know which systems will be affected. This inventory usually does not exist.
  • Difficulty adding consumers: A simple data need from the analytics team turns into a connection project lasting weeks.

The Unified Namespace concept and its components

The idea is simple: the current state of the operation is published in one place, in a hierarchy that reflects the business structure. Every system that produces data writes its data to this layer, and every system that needs it subscribes from here. Producer and consumer do not need to know each other.

A UNS is not a product but an architectural pattern. Its typical components:

Component Role Typical technology
Edge connectivity layer Reading and publishing PLC and device protocols Edge gateway, OPC UA client, Sparkplug B edge node
Message broker Publish/subscribe, last-state storage, access control MQTT broker (3.1.1 or 5.0)
Contextualisation services Adding equipment, work order and product information to raw data Stream processing, rule engine, MES
Consumers Visualisation, storage, analytics, business processes SCADA, historian, MES, ERP integration, BI

What sets a UNS apart is the concept of “last known state”. Thanks to retained messages in MQTT, a newly connected consumer immediately gets the last value of every topic. The retained value stays on the broker even when the publisher goes offline; it does not show that the value is current. That is why the value must be read together with its timestamp and the source’s connection state. QoS (delivery between client and broker), the retained last value, session state and deduplicating a transaction in the application are separate topics. Querying history is the historian’s job. A UNS does not store historical data; it distributes the current state.

If you use Sparkplug B for transport from the field to the centre, a conversion layer is needed between its flat topic structure and the hierarchical tree of the UNS; we described this distinction in detail in the MQTT Sparkplug B guide.

A topic tree based on the ISA-95 hierarchy

The common reference point for the tree’s skeleton is the role-based equipment hierarchy of ISA-95 (IEC 62264): enterprise, site, area, work centre (line, process cell, production unit or storage zone) and work unit. Because this hierarchy reflects the physical and organisational structure of the operation, it becomes the common language of IT and OT. ISA-95 does not define an MQTT topic layout; the tree below is an example built by analogy with the standard’s equipment hierarchy and is not a mandatory standard. You can access the standard via ISA.

company/
└── gebze/                                  # site
    ├── filling/                            # area
    │   ├── line3/                          # line (work center)
    │   │   ├── filler01/                   # work unit
    │   │   │   ├── status/machine_state    # "running"
    │   │   │   ├── production/good_count   # 18420
    │   │   │   ├── production/reject_count # 37
    │   │   │   ├── process/product_temp    # {"v": 21.4, "u": "°C", "ts": ...}
    │   │   │   └── event/stoppage          # stoppage event
    │   │   └── line/active_work_order      # "WO-2026-11-0412"
    │   └── ...
    └── energy/
        └── main_panel/active_power

Put the naming rules in writing at the start of the project. If you try to fix them later, you will have to go through every consumer one by one:

  • Lowercase, no non-ASCII characters, no spaces; underscore or hyphen as the separator. Choose one; don’t mix them.
  • Don’t use / at the start of a topic; it creates an empty first level. The characters + and #, and levels starting with $, have special meaning in MQTT.
  • After the equipment levels, use a fixed data classification (such as status, production, process, event, command). That way a consumer gets the status of all machines with a wildcard subscription such as company/gebze/+/+/+/status/#.
  • The equipment name should represent the functional identity, not the physical location. Accept that the topic path will change if a machine is moved to another line, and define a change process for this.
  • Fix the payload format: the value, unit, source timestamp and quality information should arrive in every message with the same field names.

If you build the work unit-level structure to match the type model on the OPC UA side, conversion becomes simpler; details are in the article the OPC UA information model and Companion Specifications.

Raw data, contextualisation and event data

Not everything published to the UNS is of the same nature. Separating them is necessary to keep the tree readable:

Data type Example Publishing method
Raw signal A counter, temperature or bit in the PLC On change; often at the edge
Contextualised value Good count for the active work order, temperature with unit Retained; the main content of the tree
Event Stoppage started/ended, recipe changed, batch completed Retain off; QoS 1
Calculated value Shift OEE, energy per unit Published by the calculating service

Moving raw PLC addresses (such as DB12.DBW40) directly into the UNS is nothing more than moving the complexity of the point-to-point structure into the broker. At the edge layer, raw data must be converted into meaningful names and units; context such as work order and product should be added by the MES or a separate service.

Don’t use retained for events. A newly connected consumer mistakes a “stoppage started” event from hours ago for a new event and raises a false alarm; this is one of the common UNS mistakes. The state resulting from the event (is the machine currently stopped) is kept in a separate retained topic. Transactional data, i.e. work order confirmations and material consumption reports, is considered separately; their exchange with the ERP, which requires guaranteed delivery, is the subject of the article MES–ERP integration with ISA-95.

Broker architecture and scaling

The broker is the central component serving the whole operation. The heaviest decision in the architecture is made here.

Approach Advantage Limitation
Single central broker Simple; one tree, one ACL Single point of failure; if the site–centre link drops, local consumers are affected too
Clustered broker High availability, horizontal scaling Product support and licensing; retained/session consistency needs testing
Site broker + central broker (bridge) Works locally even if the site link drops; DMZ transit is controlled Bridge definitions and loop prevention need care

In organisations with several plants, or with a DMZ between OT and IT, the bridge model is common: each site runs its own broker in the OT infrastructure, and only selected topic subtrees are passed to the broker in the DMZ. Define the direction and scope of the bridge explicitly. In the IT-to-OT direction, only specific topics in the command class should be allowed; preferably none. An example bridge definition in Mosquitto:

connection gebze-to-dmz
address dmz-broker.company.local:8883
bridge_cafile /etc/mosquitto/certs/ca.crt
bridge_certfile /etc/mosquitto/certs/gebze-bridge.crt
bridge_keyfile /etc/mosquitto/certs/gebze-bridge.key
# outbound only, QoS 1
topic company/gebze/+/+/+/status/# out 1
topic company/gebze/+/+/+/production/# out 1
topic company/gebze/+/+/+/event/#   out 1

We covered which component sits where in the DMZ in the article OT DMZ design. When scaling, the broker’s load is determined by the number of topics, messages per second, subscribers and retained messages. Report by exception and suitable deadband values reduce traffic significantly compared with publishing at fixed intervals. An analogue signal without a deadband also publishes its noise. MQTT 5.0’s shared subscription feature distributes load across several copies of a service processing the same data; details are in the OASIS MQTT 5.0 specification.

Tie access control to the tree structure: a line’s edge gateway writes only to its own subtree, the reporting tool only reads, and write access to command topics belongs only to defined services.

When a UNS is the wrong choice

A UNS is not the answer to every integration problem. In the following situations, another approach is more appropriate or is needed in addition to the UNS:

  • Real-time control: Interlocks between PLCs, motion control and safety functions are done not via a broker but with deterministic field protocols such as PROFINET and EtherNet/IP.
  • High-frequency waveform data: Pushing samples at the kHz level, such as vibration analysis, to the broker as individual messages is not right; process them at the edge and publish summary values.
  • Transactions requiring guaranteed delivery: ERP records such as work order confirmations or stock movements need idempotent API calls and an acknowledgement mechanism; relying on the publish/subscribe model alone is risky.
  • Small and stable structures: In an unchanging structure where two or three systems talk, the operating burden of a broker can outweigh the gain.
  • If ownership is not defined: A UNS whose topic tree and naming rules have no owner soon turns into a disorderly pile of topics. Authority to add new branches, change payloads and deprecate topics must be assigned to a team.

The migration usually starts with a single line or area: first the naming rule and payload format are fixed, an edge layer is set up, one consumer (for example a production dashboard) is connected, and then the other consumers are moved from their old connections to the UNS. Old connections are closed only after the new path has been verified. For planning, see our IT/OT integration consulting page, and for setting up the edge and broker layer, our OPC UA, MQTT and data integration page.

Frequently asked questions

Must MQTT be used for a Unified Namespace?

It is not mandatory. MQTT is the most common choice because of its publish/subscribe model, last-state storage with retained messages and hierarchical topic structure. As a concept, a UNS is a single event-driven data layer reflecting the business structure, and it can also be built with other messaging infrastructures.

Does a UNS replace the historian?

No. The UNS distributes the current state of the operation; it does not store historical data. The historian is one of the consumers subscribing to the UNS; it takes on time-series queries, trend analysis and long-term storage.

Does the existing SCADA have to be rebuilt to connect it to the UNS?

No. Many SCADA products can subscribe as an MQTT client or publish their own data. If there is no support, an intermediate converter is used. Do the migration by moving consumers to the UNS one by one, without shutting down existing connections all at once.

What happens in the topic tree if a machine is physically moved to another line?

Because the topic path reflects the equipment’s place in the hierarchy, it changes. Define a change process for the move, apply a transition period that tells consumers the old and new paths, and keep the asset identity in the historian independent of the topic path.

Support

Contact us for support or a discovery session on this topic.

Contact form
[email protected]

Eretek Industrial

PLC, PROFINET/PROFIBUS and SCADA communication, IT/OT network segmentation and industrial field connectivity. We install and monitor without stopping the production line.

Knowledge Center

Other articles