Book a consultation

HiveMQ

HiveMQ Manufacturing: Plant Placement, Customers and Failure Modes

HiveMQ manufacturing explained: where the broker sits in a plant, published customer stories, integration paths compared and how it fails in the field.

Vladimir Romanov


HiveMQ manufacturing projects put an MQTT broker between the machines that produce data and the systems that use it. HiveMQ GmbH, a private German company, sells that broker, a plant-side gateway called HiveMQ Edge, and extensions and tools around both. This page covers where HiveMQ sits in a plant, what its published customer stories claim, which integrations are documented, how the integration paths compare and how it fails. It is an independent profile, current as of November 2026, built from HiveMQ’s documentation and releases. Everything about HiveMQ’s customers is HiveMQ’s own claim. The page sits under the HiveMQ technology hub, and the company is profiled at HiveMQ the company.

Where HiveMQ sits in a plant

HiveMQ is a layer, not a level. The broker usually runs at Level 3 as the plant broker. HiveMQ Edge runs closer to the line, converts PLC protocols to MQTT and can run an embedded broker for a production area. HiveMQ’s own Unified Namespace article describes the same shape: Edge converts point-to-point protocols to MQTT for an area or line, devices with native MQTT or tools such as Kepware feed a raw namespace, and bridges connect brokers across levels and sites. Siemens’ Architecture Hub page for HiveMQ (a “Partner Solution” page that carries no partnership text) draws one HiveMQ instance per factory, enforcing topic structure, metadata and access control, with selected topics bridged to a central HiveMQ cluster.

Where HiveMQ sits on the ISA-95 and Purdue levels Five bands from Level 0 and 1 at the bottom to Level 4 and 5 at the top. PLCs and devices feed HiveMQ Edge at the area or line level. Edge publishes to a plant broker at Level 3, where Ignition, a historian and MES also sit. An optional IT-facing broker in the DMZ at Level 3.5 receives bridged topics. A central HiveMQ cluster or HiveMQ Cloud broker at Level 4 and 5 serves ERP, MES, analytics and Kafka. Level 4 and 5 Enterprise and cloud Level 3.5 DMZ (optional) Level 3 Site operations Level 2 Area and line Level 0 and 1 Machines and PLCs Central HiveMQ cluster or HiveMQ Cloud ERP, MES, analytics, Kafka IT-facing broker or bridge endpoint (our pattern) Plant broker or cluster (HiveMQ broker) Ignition, historian, MES HiveMQ Edge per line or area (adapters, embedded broker) Line HMI, native MQTT devices PLCs, drives, sensors: OPC UA, Siemens S7, Modbus, EtherNet/IP, MQTT polled or subscribed MQTT publish bridge, plant to DMZ bridge, DMZ to centre
HiveMQ is a layer, not a level: Edge at the line, a broker at the plant, a bridge upward. The DMZ broker is our pattern, not a HiveMQ reference design.

The dashed DMZ broker is our view, not a HiveMQ reference design. We found no HiveMQ guidance on Purdue or ISA-95 level placement, and OT network security explains why we would add it. The Unified Namespace itself is covered on its own page: see HiveMQ Unified Namespace.

Large cutting machine inside a bright industrial hall with a high roof and tall windows
A machine hall: the broker only matters once someone decides which of these machines publish, to which topics, at what rate.

HiveMQ manufacturing customer stories, as HiveMQ publishes them

HiveMQ’s case-study pages are undated vendor claims with no independent verification. The listing page showed 24 stories and a “Load more” button when we read it, so this is a selection. The table is what each page says, and nothing more.

CustomerSectorWhat HiveMQ’s page claims
Mercedes-Benz (page address says daimler)AutomotiveA Vehicle Diagnostics System rolled out in 24 factories, with 10,000 test devices and nearly 470 million messages a month. Each instance connects its test devices to a HiveMQ broker located in the factory. Quote from the chief software architect: “We have been running the VDS using HiveMQ for 4 years and the HiveMQ broker has not gone down.”
Eli LillyPharmaAn equipment connectivity platform for standalone instruments such as pH meters and balances, using Sparkplug B over MQTT, “several hundred instruments” across laboratory and manufacturing sites, with bridge and Kafka extensions to the cloud
CurrentaChemical parksDigital samplers for wastewater testing “that allow remote configuration and online data access”; “more than 10,000 concurrent connections” and throughput “exceeding 10,000 messages per second”; chosen for the clustering feature and the Enterprise Security Extension
FordAutomotiveA standard way to collect machine data across plants, “thousands of machines per plant across dozens of sites”, with no quantified results

How we read these. Only the Mercedes-Benz and Currenta pages give scale numbers a plant engineer can test an architecture against, and the Mercedes-Benz quote claims four years of running at an unknown date. Lilly’s page does not say which HiveMQ edition is used, and its expansion date has passed. The Currenta page gives an availability target as 99.9% in one place and 99% in another. We found no HiveMQ case study tagged to food and beverage or consumer goods in what we read. None of this says HiveMQ is a poor choice. It says the public evidence is thinner than the logo wall suggests, so ask for a reference call with a plant of your size.

HiveMQ manufacturing integrations that matter

IntegrationWhat is documentedStatus
Ignition (Inductive Automation)HiveMQ joined the Ignition Technology Ecosystem Program on 23 July 2025; a HiveMQ guide connects Cirrus Link modules to the brokerDocumented. See Ignition and HiveMQ
HighByte Intelligence HubA partner page, a joint webinar and a HiveMQ knowledge-base guide for TLS and non-TLS listenersDocumented, thin on data flow.
Kepware (PTC)A partner page says the work “focuses on embedding Kepware technology directly into HiveMQ”Thin: no date, version or mechanism
SiemensThe Architecture Hub reference design aboveReference design, not a stated partnership
LitmusNone foundAbsent from what we read
Rockwell AutomationNo Rockwell Automation partner or integration page found. Edge’s EtherNet/IP Logical Addressing adapter addresses CIP attributes by class, instance and attribute and documents slot routing for ControlLogix and CompactLogix racksNo integration found; protocol access documented

Beyond these, the extension catalogue covers Kafka, several databases and the major clouds, and Edge adds protocol adapters. We found no extension for AVEVA PI or Canary. The adapter detail is on protocol adapters.

Integration paths compared

The question is rarely “can HiveMQ take the data”. It is which component turns PLC data into MQTT topics, and who owns that component afterward.

PathWhat publishes to the brokerFits whenWatch for
Edge adapters directHiveMQ Edge polls or subscribes to the PLCFew protocols, one team, no existing gatewayPolling load on the PLC; adapters mostly read-only
OPC UA server, then Edge or a gatewayAn OPC UA server (Kepware or a native one) feeds the brokerMany vendors already behind OPC UAPer-adapter subscription settings; subscription limits unconfirmed
Ignition with Cirrus LinkTransmission publishes Sparkplug BIgnition already runs the plantSTATE and ACL setup; buffering lives in Transmission
HighByte or another DataOps layerThe DataOps tool models and publishesA modelled namespace matters more than raw tagsTwo components to patch; unclear ownership of the model
Native MQTT on the deviceThe machine itselfNew equipment that speaks MQTTUneven payload formats between OEMs

In our view, choose by who will maintain the publisher in year five. A broker is the easy part to standardise. The publisher is where plant-specific work accumulates. For the wider design see MQTT in manufacturing.

How HiveMQ manufacturing deployments fail

Most of these come from HiveMQ’s own documentation, read as a plant engineer would read it. They are our reading, not HiveMQ guidance, and items we could not confirm are marked.

  • Polling is a load on the PLC. Edge’s polled adapters default to 1,000 ms and accept as low as 1 ms. The Siemens S7 adapter needs PUT/GET access enabled on the PLC, which is a security decision to make on purpose.
  • Changed-only publishing is not uniform. Modbus, S7, Beckhoff and BACnet default to publishing changed data only. The MTConnect and database adapters publish every interval. For the OPC UA adapter the documentation says unchanged values are still published, while a HiveMQ forum reply suggests otherwise, so bench-test it.
  • OPC UA settings are per adapter, not per tag. The default serverQueueSize of 1 means a fast signal inside a 1,000 ms publishing interval keeps only its latest sample unless the queue is raised. Mixing fast and slow signals needs separate adapter instances. This is our inference from the setting definitions.
  • Writes are not free. Writing to an OPC UA server through Edge needs a southbound mapping, a commercial feature. The EtherNet/IP Logical Addressing adapter also documents writes, and we did not confirm its licence tier.
  • Sparkplug permissions. In a January 2025 HiveMQ community thread, a user could publish Sparkplug data only as admin until the legacy STATE/# topic was allowed alongside spBv1.0/STATE/#.
  • Cluster replica count. HiveMQ’s documentation says a replica count below 2 can lose data on a node restart, and permanent loss of more nodes than the replica count makes persistent data unavailable until recovery.
  • Version drift. Pin versions in every plant and test upgrades on a bench.
  • WAN dependence. A plant that publishes straight to a central broker stops its data when the link drops. See multi-site architecture.

What we do here

We design MQTT and Unified Namespace architectures and integrate brokers with PLCs, SCADA, historians and enterprise systems as part of IT/OT projects. We are independent of HiveMQ: we have no commercial relationship with the company and do not resell its products.

  • Topic and namespace design that survives the second plant, drawn from general MQTT and UNS practice.
  • Choosing the publisher per machine family: Edge adapter, OPC UA, SCADA module or native MQTT.
  • Broker placement against your zones and conduits, with an OT network assessment where the segmentation is unclear.
  • Bench testing adapter behaviour (changed-only, queue sizes, reconnect) before anything is rolled to the line.
  • Independent review of a broker choice, including whether Ignition and SCADA already cover the need.

Limits and open questions

  • Customer stories are undated, so none can be tied to a HiveMQ version.
  • Adapter parameters came from a summarised read of HiveMQ’s documentation and should be re-checked against the live page before design use.
  • Whether Edge can publish Sparkplug B is not confirmed in what we read.
  • Licence tiers for the MTConnect and database adapters are not stated in the open-source list.
  • Kepware’s “embedding” has no documented mechanism.
  • WAN clustering is undocumented either way. See the multi-site page.

Deeper

Across to a sibling: HiveMQ Edge and its adapters. Back to the HiveMQ hub.

Frequently asked questions

Where does a HiveMQ broker sit in a manufacturing plant?

Usually at Level 3, as the plant broker that machines, HiveMQ Edge gateways, SCADA and historians connect to, with a bridge upward to a central cluster or cloud. HiveMQ's Siemens reference page describes one HiveMQ instance per factory. HiveMQ publishes no Purdue placement guide, so placement beyond that is a design decision.

Which manufacturers does HiveMQ say use it?

HiveMQ's published case studies include Mercedes-Benz (the page address still says daimler), Ford, Eli Lilly and Currenta. Mercedes-Benz gives the most figures: 24 factories, 10,000 test devices and nearly 470 million messages a month. Currenta gives more than 10,000 concurrent connections. All of these are HiveMQ's own claims. The pages are undated and not independently verified.

Does HiveMQ integrate with Ignition, Kepware and HighByte?

HiveMQ joined the Ignition Technology Ecosystem Program on 23 July 2025 and publishes a guide for Cirrus Link modules. It lists HighByte and Kepware partner pages, with little technical detail for Kepware. We found no HiveMQ integration with Litmus or Rockwell Automation. Check each vendor's current documentation.

Can HiveMQ Edge read a PLC directly?

Yes for several families. HiveMQ's documentation lists adapters for OPC UA, Siemens S7, Modbus TCP, Beckhoff ADS, EtherNet/IP by CIP address, BACnet/IP and others. Most are read-only northbound. Writing to an OPC UA server needs a commercial southbound mapping, and the EtherNet/IP Logical Addressing adapter also documents writes, with its licence tier not confirmed.

How does HiveMQ fail in a plant?

Mostly at the edges: polling that loads a PLC, changed-only publishing that differs by adapter, Sparkplug permission gaps, a cluster replica count below two, and WAN dependence. Few of these are broker faults. We mark which items HiveMQ documents and which are our reading.

References and further reading

  1. HiveMQ Edge documentation · HiveMQOpen-source and commercial features, platforms and licence keys.
  2. HiveMQ Edge protocol adapters · HiveMQDirections, polling defaults and limits per adapter.
  3. Mercedes-Benz Vehicle Diagnostics System case study · HiveMQUndated vendor case study with the scale figures quoted here.
  4. Eli Lilly equipment connectivity case study · HiveMQ
  5. Currenta case study · HiveMQ
  6. HiveMQ architecture page · Siemens Architecture HubA partner-solution reference design: local broker per factory, bridged to a central cluster.
  7. HiveMQ joins the Inductive Automation ecosystem · HiveMQDated 23 July 2025.
  8. HiveMQ cluster documentation · HiveMQReplica count, discovery, transport and recovery.

← HiveMQ

Have a project that needs technical ownership?

Send us the situation. If it is not something we should take on, we will say so.

The Plant-Side Brief

The weekly brief for the people who own a plant’s control systems.

One email every Wednesday for engineering and operations leaders responsible for controls, SCADA, MES and OT networks. Skim the top in two minutes; read the Deep Dive when you have fifteen.

  • One modernization problem, worked through to cost, risk and a schedule you can defend. The Deep Dive.
  • End-of-support dates before they reach your plant Windows, servers, controllers and HMI software. Lifecycle Watch.
  • A checklist your engineers can run that week not a framework to admire. Field Kit.
  • Every number sourced, no vendor agenda we are paid by plants, never by vendors.