Unified namespace: the idea, and what it does not solve

The problem a unified namespace solves is visible on the whiteboard of every plant that has been integrated over time: a mesh of point to point connections between systems that were never designed to talk to each other.

The MES reads from three PLCs directly. The historian polls a different set. The reporting tool has its own OPC connection. The maintenance system has a spreadsheet somebody exports on Fridays.

The count that makes the argument

Connecting n systems point to point takes up to n times n minus one, over two connections. Six systems is fifteen. Ten is forty five.

Through a broker it is ten connections, one each. That is the whole architectural argument and it is a good one.

The hierarchy is the actual work

The broker is the easy part. Agreeing what things are called is where the effort is, and where the projects that fail, fail.

Most people start from ISA-95: enterprise, site, area, line, cell. A topic becomes something like acme/leeds/packing/line3/filler/temperature.

The rule that keeps it usable: the hierarchy describes the plant, not the software. The moment a topic path contains the name of a SCADA system or a database table, the namespace has started describing its own implementation and will have to be renamed when that implementation changes.

Report by exception, and the retained state

The second idea, and it is what makes this efficient rather than merely tidy.

Publishers send a value when it changes, not on a poll. A tank level that has not moved produces no traffic at all.

MQTT's retained messages mean a subscriber connecting at three in the morning immediately receives the current value of everything it subscribes to, without waiting for the next change. Without that, a late subscriber sees nothing until something happens.

What Sparkplug B adds

Plain MQTT gives you topics and payloads and nothing else. Every serious deployment then reinvents the same three things, and Sparkplug B is those three things written down.

• Birth and death certificates. A publisher announces its full tag set on connect, and the broker publishes its death message if it disappears. Subscribers know what exists and when it stopped.
• State management. Sequence numbers so a subscriber knows it missed something, and a defined rebirth mechanism to resynchronise.
• A payload format. Typed, with timestamps and metrics, rather than a JSON blob whose shape each publisher invents.

What it does not solve

Worth being blunt, because the enthusiasm sometimes outruns it.

Protocol drivers. Something still has to talk Modbus to the old machine. The namespace does not read a PLC; a gateway does, and publishes.

Data modelling. Publishing a temperature does not say what it is the temperature of, what units it is in, or how it relates to a batch. That is modelling work and it is unavoidable.

Governance. Somebody has to own the naming, or within a year there will be three names for the same measurement, all in use, none wrong.

Security. A broker every system can reach is a broker an attacker who reaches one system can also reach. Authentication per client, topic level authorisation and TLS are not optional, and the default configuration of most brokers has none of them.

Where to start

Not with the whole plant. One line, one gateway, one broker, the ISA-95 hierarchy for that line, and one consumer that genuinely needs the data.

Getting one line right teaches you the naming argument at a scale where renaming everything is still an afternoon.