Smart Manufacturing & Industrial IoT

The problem: the plant floor doesn't speak MQTT

Most cloud IoT platforms are built around MQTT and HTTP, because that is what consumer and lightweight sensor devices speak. A factory floor is different. PLCs and drives talk Modbus TCP. Process historians and SCADA supervisors talk OPC UA. Motion and robotics cells increasingly talk DDS/RTPS. Network and asset monitoring runs over SNMP. None of that is optional — it is the installed base, often running equipment with a service life measured in decades.

The usual fix is a protocol gateway: a small appliance or service that translates Modbus or OPC UA into MQTT so it can reach a cloud platform. Every gateway is another thing to patch, another point of failure, and another hop of latency between a sensor reading and the system that acts on it. When a press is about to jam or a motor is drawing current it shouldn't, that extra hop is the difference between an alert and a shutdown.

How Relay fits

Relay parses Modbus, OPC UA, DDS/RTPS, MQTT-SN and SNMP natively, alongside MQTT, AMQP and the rest of its 21 application protocols. A single Relay cluster sits on the plant network and speaks every one of those wire formats directly — there is no gateway sidecar, no separate licence per protocol, and no extra hop between a device and the system reading it. The same cluster can also expose that data over MQTT or a REST/webhook bridge for whatever dashboard or MES system consumes it upstream.

• →10M+ concurrent connections: connect sensors across every line, cell and facility from one control plane.
• →Sub-millisecond latency: anomaly detection and alerting run close to the source, not after a cloud round-trip.
• →Built-in stream processing: SQL-based rules, filtering and enrichment run inside Relay, so a lightweight anomaly check doesn't need a separate stream-processing cluster.
• →Zero downtime: cluster mode gives real failover, and rolling upgrades never drop a message in flight.

A typical architecture

Field devices and PLCs publish directly to a Relay cluster running on the plant network, over whichever protocol they already speak. Relay's routing and filtering rules forward relevant events to a local anomaly-detection model or historian, while raw telemetry is aggregated and forwarded — over MQTT, gRPC or HTTP/2 — to whatever sits above the plant network, whether that is a corporate MES, a cloud data lake, or nothing at all if the site runs air-gapped. Because durability, replay and delivery guarantees are switched on per module rather than bundled into a fixed tier, a line that only needs best-effort telemetry can stay cheap while a safety-relevant cell gets exactly-once delivery and full replay.

Deployment

Relay runs the same way on AWS, Azure, GCP, your own racks, or fully air-gapped — the rate card and the feature set don't change with the target. For a regulated or safety-critical plant, that means the exact platform validated in a lab or a pilot line is the one that ships to production, with no cloud dependency introduced along the way.