24 Industrial Protocols: Families and Timing
More
Depth: Practitioner
Progress saves as you read. Mark sections complete as you go; your progress appears in the sidebar.
- Podcast audio coming soon
- Character narration audio coming soon
24.1 Start with the Conveyor
A protocol is a shared set of rules for exchanging data. Industrial communication adds a practical question: can those rules carry the right information before a machine must act? This chapter compares legacy and Ethernet families through timing, bandwidth, and cycle-time evidence. You will distinguish serial from industrial Ethernet protocols, compare latency and throughput, and use ISA-95 levels to place control and integration duties in brownfield and greenfield architectures.
Imagine a bottle conveyor that must stop before a jam damages the machine. Its stop message and a status report for next week’s maintenance plan share a factory, but have different deadlines and failure costs. A fast cycle alone cannot establish safe timing. The path must also handle stale, duplicated, missing, or wrongly accepted messages, and reach a defined safe state when communication fails.
Start with the job: who sends, who acts, how soon action must happen, and what happens if the link fails. Give the safe stop, motion loop, machine state, trend, and update separate records. Each needs its own time, loss, load, and recovery rule. Then check distance, topology, device support, and who can diagnose the network. These facts eliminate choices that cannot fit before a protocol name becomes a preference.
The plant’s history matters too. Equipment may run for 20 or 30 years, with spare parts, electrician skills, PLC programming tools, support contracts, cable trays, and downtime windows already fixed. A new packaging line can optimize around a motion network; a brownfield water plant may need to preserve Modbus RTU instruments and add gateways, diagnostics, and security segmentation. The problem is to satisfy each duty while keeping the installed plant operable. The approaches below separate those duties, compare the families, and test what their timing claims actually prove.
24.2 Prerequisites
For background, Industry 4.0 Fundamentals explains ISA-95 automation levels. Networking Basics introduces latency, bandwidth, and reliability; Transport Protocols covers TCP/IP and UDP.
KEY CONCEPTS
- Determinism
Cyclic traffic arrives within a bounded timing window, with a known jitter budget and timeout behavior.
- Fieldbus
Machine networks use device profiles, diagnostics, and installation rules suited to industrial equipment.
- Industrial Ethernet
PROFINET, EtherNet/IP, and EtherCAT share Ethernet foundations but have different real-time behavior.
- OPC UA
Vendor-neutral information models expose structured plant data to SCADA, MES, historians, and IT systems.
- Brownfield integration
Connect existing PLCs, drives, instruments, and serial networks without unnecessary device replacement.
- Industrial DMZ
A segmented boundary controls and inspects data exchanged between enterprise IT and OT systems.
24.3 Learning Objectives
- Compare industrial protocols (Modbus, PROFINET, EtherCAT, OPC UA) by latency, throughput, and use case
- Evaluate protocol requirements for different industrial applications based on ISA-95 levels
- Justify protocol selection based on latency, determinism, and security constraints
- Distinguish between legacy serial protocols and modern industrial Ethernet protocols
- Design industrial communication architectures for specific brownfield and greenfield use cases
24.4 Separate the Communication Duties
Monitoring reads values for dashboards, historians, alerts, or maintenance without directly controlling the machine. Automation exchanges I/O and commands between PLCs, drives, remote I/O, and controllers within a bounded cycle. Integration moves contextualized plant data to SCADA, MES, CMMS, historians, analytics, or cloud systems. Motion and safety need their own requirements within that picture: a conveyor status light, servo axis, safety interlock, and ERP report should not inherit one communication pattern.
Modbus TCP may fit simple register reads. A register is an addressed data value exposed by a device. PROFINET IO or EtherNet/IP may fit distributed I/O and drives. EtherCAT can serve tightly synchronized motion. OPC UA provides semantic IT/OT integration once the real-time boundary is protected. MQTT Sparkplug B can publish normalized telemetry upward; it does not supply a deterministic fieldbus for closed-loop control. Choose among these responsibilities before comparing headline speeds.
Figure 1 answers why industrial is not one requirement set, by splitting the industrial side in two before any protocol is named.
Start at the strip above the table in Figure 1. A tank-level trend sampled every hundred milliseconds and a drive loop that runs every millisecond can share a plant, a network and a vendor, yet only the first can be retried or buffered. The table then keeps them in separate columns through six rows, from timing and reliability to safety, security and lifecycle. Read across one row at a time and the same phrase means different work in each column. Best effort is fine for a consumer status update, tolerable on a dashboard that flags stale data, and unacceptable on an interlock. The numbered rule at the foot puts the order right. Classify the path first and choose a protocol last.
The consumer/industrial distinction is about consequence as well as speed. Consumer status updates can often be retried, and best-effort delivery may cause only inconvenience. Plant monitoring can also buffer or retry if it flags stale data, but drive, I/O, interlock, and closed-loop paths need explicit cycle-time and jitter budgets. Availability and timeout behavior must match the machine or process consequence.
Consumer security often centers on user data and convenience. Industrial design must protect process integrity and use safety-rated operation where the function requires it. Multi-year support, spare parts, vendor tools, and brownfield compatibility can dominate choices that would follow shorter refresh cycles in consumer systems.
Bandwidth is another constraint, not a substitute for timing evidence. Include protocol headers, retransmissions, and management traffic rather than counting only payload. The bandwidth workbench varies digital inputs and outputs, 16-bit analog inputs and outputs, and target cycle time, then compares 10 Mbps, 100 Mbps, and 1 Gbps links. Its overhead allowance is an estimate; measure overhead for the actual frame sizes and link. Targeting less than 50% utilization leaves headroom for bursts and diagnostics. Sustained 80% utilization is a red flag for link upgrades or traffic optimization, rather than proof that a control deadline has already failed.
24.5 Match Protocol to Control Loop
A defensible selection records the loop type, timing budget, safety role, ownership boundary, installed base, security zone, and lifecycle plan before naming a protocol.
Start by writing the required cycle time, jitter tolerance, distance, topology, safety role, and ownership boundary. A cycle time is the interval between repeated control exchanges; jitter is the variation in that timing. A 100 ms tank-level trend can tolerate retry and buffering. A 1 ms drive loop cannot wait for a broker, VPN, or cloud API. A brownfield Modbus RTU segment may be left in place and bridged, while a new motion cell may justify EtherCAT or PROFINET IRT. The same plant can therefore use several protocols at once, each with a different responsibility.
Protocol choice also follows vendor and tooling realities. Siemens-heavy plants often standardize on PROFINET and TIA Portal. Rockwell/Allen-Bradley plants often use EtherNet/IP with CIP objects and Studio 5000. Beckhoff and many motion-control systems use EtherCAT. Legacy panels may expose Modbus RTU over RS-485 or Modbus TCP. Integration layers may use OPC UA, MQTT Sparkplug B, or historian connectors. Maintenance teams must be able to diagnose the chosen stack at 2 a.m., not only during a vendor demo.
The selection tree in the linked figure in Part 2 turns these constraints into an inspection order: start with cycle time, then account for determinism, safety, installed assets, and the boundary where plant data moves into integration systems.
For a practical selection review, ask for an example message path. A temperature trend might move from a Modbus RTU transmitter to a gateway, then to OPC UA, then to a historian and dashboard. A robot-axis command might stay inside EtherCAT or PROFINET IRT and never cross the enterprise boundary. A plant KPI might be aggregated by an edge node and published through MQTT after quality checks. These paths reveal where latency, safety, cybersecurity, and ownership actually sit.
Keep real-time protocols inside the cell or line, then publish cleaned context upward through OPC UA, Sparkplug B, a historian, or MES interface.
For the conveyor, mark each control unit, switch, gateway, and service between source and final actor. Put an owner and clock at each step, and state what safe state follows if that step is late or unavailable. Measure at the end that acts, not only at the sender. Keep the worst valid cycle: a low mean does not show that every stop arrives on time.
Test with normal reads, alarms, start-up bursts, tool access, logs, and updates running alongside control traffic. A large file or scan must not starve the time-critical path. If jobs need separate paths or traffic classes, record where and how they are separated. Then break a cable, turn off a switch, restart a controller, fill a queue, and change a device. Record both the safe action and the time to return. A second path helps only if switching to it meets the allowed recovery time.
Keep installed speeds, wiring, connectors, vendor tools, and controller limits in the record. A new link can provide a clean data path while the old controller remains the bottleneck. Replacement cost and plant stop time still count. For data crossing into business systems, state which values may leave and who may send changes back. Test a lost outside link and a blocked user while local safety work continues.
24.6 Establish the Timing Mechanism
Industrial Ethernet does not automatically make a network deterministic. PROFINET RT prioritizes cyclic I/O on Ethernet, PROFINET IRT adds reserved time windows for tighter synchronization, EtherNet/IP uses CIP objects and can use CIP Sync with Institute of Electrical and Electronics Engineers (IEEE) 1588 PTP, and EtherCAT processes frames as they pass through each slave. These details explain why two protocols with similar cable connectors can behave very differently under load. Determinism comes from scheduling, clocking, frame handling, switch behavior, and device implementation, not from the word Ethernet alone.
Legacy protocols have different tradeoffs. Modbus RTU over RS-485 is simple and widely deployed, but it is request-response, usually single-master, and carries registers rather than rich semantics. PROFIBUS DP and DeviceNet solved many fieldbus needs before Ethernet became common. IO-Link often connects smart sensors and actuators to a master that then speaks PROFINET, EtherNet/IP, or another higher-level protocol. These protocols can still be correct when downtime, replacement cost, or certification risk makes wholesale migration unattractive.
Modern designs increasingly combine protocol layers. A drive network may stay on EtherCAT, an I/O island may use PROFINET, a gateway may expose an OPC UA server, and a DMZ broker may publish Sparkplug B topics to analytics systems. Time synchronization, VLANs, QoS, firewall rules, certificate management, firmware compatibility, and spare-part availability become part of the protocol decision. International Electrotechnical Commission (IEC) 62443 zones and conduits help decide which messages can cross from cell, area, site operations, and enterprise networks.
- Timing machinery: Look for cyclic scheduling, PTP/IEEE 1588 clocking, time-aware shaping such as TSN IEEE 802.1Qbv, and device-rated update times.
- Data machinery: Distinguish raw coils/registers, CIP objects, PROFINET GSDML descriptions, EtherCAT ESI files, IO-Link IODD files, and OPC UA information models.
- Operations machinery: Check diagnostics, packet capture support, replacement-device commissioning, firmware policy, and who can troubleshoot the network at 2 a.m.
Testing should match the consequence. A monitoring gateway may need stale-data flags, retry behavior, and historian gap handling. A motion cell needs cycle-time and jitter tests under worst-case load. A safety-related network needs certified devices, validated safety functions, and a change-control process. A cloud integration path needs proof that no inbound enterprise connection is exposing a control system. The selected protocol is only credible when those evidence paths are defined.
A time-aware shaper illustrates why scheduling needs more than a fast link. Each outgoing port has queues that hold waiting messages. Gates allow or block their transmission; a repeating gate control list assigns turns. Participating devices need aligned clocks and a shared schedule start. Before the control window, a guard band blocks lower-priority frames from starting too late and overlapping protected time. Admission checks that added traffic fits each relevant outgoing port’s reserved budget.
In the tour’s Motion Cell example, a 1,000 μs cycle assigns 8% (80 μs) to guard time, 30% (300 μs) to control, and 22% (220 μs) to video. Best effort receives the remaining 40% (400 μs), never below zero. Guard, control, video, and best effort fill the cycle in that order. With clock error at 1 μs, the schedule has different margin than the Clock Drift example’s 40 μs error, even though all four windows stay unchanged. The model reports high risk for that drift: clock disagreement can shift gate openings across devices. Local window arithmetic cannot establish timing across the full path, and these teaching values supply no measured proof for a real motion cell.
TSN time-aware shaper narrated tour
Narrated audio tour. The full transcript is below.
Play the tour to open the workbench. Pause, replay a beat, or read the full transcript below. Some situations cannot be shown by this teaching model; those beats keep the existing frame and provide captions only.
Full transcript — TSN time-aware shaper (8 beats)
B1 — Motion cell
You are the engineer responsible for a group of machines whose movements must stay coordinated. This group is called a motion cell. Its control commands share an industrial network with video and routine data.
Model caption: You manage a motion cell’s timing. Control, video, and routine data share scheduled transmission opportunities; late control messages put timing at risk.
B2 — Gate control list
A time-aware shaper gives each traffic queue scheduled turns to send. A queue holds messages waiting to leave through a network connection called an outgoing port. A gate allows or blocks transmission from a queue. A repeating gate control list sets when each gate opens and how long it stays open.
Model caption: A time-aware shaper schedules queue gates at an outgoing port. Deterministic timing bounds delivery time, including understood arrival variation and timeout behavior.
B3 — Guard band
Participating devices need aligned clocks and a shared schedule start time. Before control traffic gets its turn, a guard band blocks lower-priority transmissions. It prevents a frame, or unit of network data, from starting so late it overlaps the protected control window. The guard band protects time by leaving it unavailable for data.
Model caption: Aligned clocks share a schedule start. Guard time blocks lower-priority transmissions before control, preventing a late-starting frame from overlapping the protected window.
B4 — Control window
Use the model’s Motion Cell example. The cycle lasts one thousand microseconds. Control gets thirty percent, and the guard band gets eight percent. Clock error is one microsecond. Eight percent of one thousand gives eighty microseconds of guard time. Thirty percent gives three hundred microseconds for control frames to transmit.
Model caption: Motion Cell: 1000 us cycle; 1 us clock error. Guard = round(1000 × 8 / 100) = 80 us. Control = round(1000 × 30 / 100) = 300 us.
B5 — Remaining windows
The model assigns video twenty-two percent, or two hundred twenty microseconds. For best effort, start with one hundred percent. Reserve eight percent for guard time and thirty percent for control. Reserve another twenty-two percent for video. Forty percent remains: four hundred microseconds for non-critical data. The model never lets the remaining share fall below zero. The windows and guard time fill the thousand-microsecond cycle.
Model caption: Video = 22% = 220 us. Best effort = max(0, 100 − 8 − 30 − 22) = 40%; round(1000 × 40 / 100) = 400 us. Guard 80 + control 300 + video 220 + best effort 400 = 1000 us.
B6 — Service order
After the guard band, the control gate opens while the other data gates stay closed. Video follows, then best effort uses its turn. The schedule repeats when the cycle ends. Admission means checking whether added traffic fits the reserved time budget; control frames must fit their window at every relevant outgoing port.
Model caption: Control transmits first after guard time, then video, then best effort; the gate list repeats. Admission checks whether added traffic fits reserved time. Control frames must fit at every relevant outgoing port.
B7 — Clock drift risk
A common mistake is to assume unchanged window lengths mean unchanged timing risk, but control timing depends on devices keeping their schedules aligned. Keep the same windows but increase clock error to forty microseconds, as in the Clock Drift example. The model reports high risk: clock disagreement consumes timing margin and can shift gate openings across devices. Local window arithmetic cannot establish timing along the whole path.
Model caption: Same 80/300/220/400 us allocation, but clock error rises from 1 to 40 us. The model reports High risk because clock disagreement consumes margin and can shift openings across devices. A local gate list cannot establish whole-path timing.
B8 — Deadline takeaway
A dependable control deadline rests on shared time, traffic fitting its windows, and measured timing from tests.
Model caption: A dependable control deadline rests on shared time, traffic that fits its windows, and tested timing evidence. Teaching values illustrate the schedule; this asset supplies no measured proof for an actual motion cell.
24.7 Follow the Families from Serial to Ethernet
The following diagram shows how industrial communication protocols evolved from simple serial protocols toward fieldbus, Ethernet-based control, and IT/OT convergence standards:
Figure 2 answers why one modern plant still runs protocols that were designed forty years apart.
Read Figure 2 downward and watch the problem being solved change twice. The early entries, from Modbus in 1979 through the fieldbuses, answer a physical question: how to get rugged, predictable access to a shared bus. The middle block moves onto Ethernet without giving up determinism, which is why 2003 produces two different answers in one year, one dividing time into slots and one passing a single frame through every device. The last entry changes the question again, because OPC UA is not competing on speed. It adds security and a described data model so plant information can safely leave the plant floor.
Figure 3 sets out the data-motion patterns served by the industrial protocol families.
Read Figure 3 across four different jobs. Modbus has one master poll registers, which suits simple monitoring and brownfield links. PROFINET organizes controller-device exchange as cyclic I/O, with IRT tightening the timing for motion. EtherCAT passes one frame through each device and back, enabling sub-millisecond cycles across flexible topologies. OPC UA instead gives SCADA, MES, and IT clients structured plant information. The comparison is not a speed ranking: control timing, topology, and semantic integration determine where each pattern belongs.
24.8 Legacy Industrial Protocols
Existing networks are part of the design constraint, so compare their communication model and maintenance burden before planning replacement.
24.8.1 Modbus (1979)
Modbus is one of the oldest and most widespread industrial protocols:
Its simple polling model is easy to implement and has minimal overhead: one master asks devices for data in turn. Modbus RTU uses serial links, while Modbus TCP uses Ethernet and TCP/IP. RTU provides addresses for up to 247 servers; this is not an RS-485 electrical segment capacity or a Modbus TCP network limit. Traditional Modbus has no built-in security. Its large installed base in utilities, building systems, and brownfield industrial sites explains why gateways remain useful.
Figure 4 answers both halves of the Modbus reputation: why it is so easy to bridge, and what it cannot do.
One box in Figure 4 holds the initiative. The master sends a request, a single device answers, and the numbered sequence beneath shows the cycle repeating around the bus. Nothing on the lower row can speak on its own, so a sensor with urgent news waits its turn, and that is the cost of the design. The benefit is in the same picture. Every device is just an address holding registers, which is why the protocol maps onto serial wiring or Ethernet with equal ease and why bridges to newer systems are simple to build. The serial address limit printed at the foot is separate from the device capacity of an installed electrical segment. Together these explain the long installed base and the modern gateway sitting in front of it.
Typical applications: Building automation, energy management, simple machine control
24.8.2 PROFIBUS (1989)
Process Field Bus, dominant in European process automation:
Token passing provides deterministic bus access, and a multi-master arrangement lets multiple PLCs coexist. Copper links can run at up to 12 Mbps. A 100-device network should not be treated as a universal limit: validate segment and repeater requirements for the installation. PROFIBUS is established in European process automation, including chemical plants and refineries.
Bus access and link speed describe different parts of the timing problem. Check the required cycle time and worst delay on the installed path under busy traffic and device restarts. Existing controllers, wiring, spare parts, and maintenance tools also shape the choice. Keeping a working fieldbus may be appropriate when replacement cost and available downtime make migration unattractive.
24.8.3 DeviceNet (1994)
CAN-based protocol for discrete manufacturing:
DeviceNet uses the CAN physical layer associated with automotive-grade reliability and a producer-consumer model for efficient broadcasting. It connects low-level sensors, drives, and valves, and can supply device power over the network.
These characteristics place DeviceNet close to machine devices, where communication must fit the control job. Efficient broadcasting alone does not establish that the installed path meets its timing budget. Check cycle time, jitter, timeout behavior, and recovery under load. For an existing installation, include supported devices, maintenance tools, spare parts, and downtime windows when deciding whether to retain the network.
24.9 Compare Industrial Ethernet Timing
Ethernet changes the transport foundation, but each family still answers the conveyor’s timing problem differently. Compare the scheduling and device behavior before treating a shared connector as evidence of equivalent control performance.
24.9.1 PROFINET (2003)
Siemens’ industrial Ethernet successor to PROFIBUS:
Performance tiers:
- PROFINET IO: Standard I/O; the <100 ms example is a requirement to validate, not a fixed protocol limit
- PROFINET IRT (Isochronous Real-Time): <1 ms, deterministic, motion control
- PROFINET CBA: Component-based automation
Inspect Figure 5 to see how cyclic control and ordinary configuration traffic share Ethernet without receiving the same timing treatment.
Begin Figure 5 at the shared Ethernet hardware, then separate the traffic classes above it. Standard PROFINET RT gives cyclic I/O priority over ordinary traffic. IRT reserves time windows for tighter synchronization and motion workloads below one millisecond. Configuration, diagnostics, and IT-style exchanges remain in the non-real-time layer on the same backbone. The key distinction is scheduling, not the cable: industrial Ethernet does not become deterministic merely because control devices use it. A design must select the timing class and configure the network that enforces it.
Key features:
- Uses Ethernet hardware; IRT requires timing-capable devices and infrastructure
- Backward compatible with PROFIBUS via proxies
- Supports web services and IT integration
- Large installed base in Siemens-centered automation ecosystems
24.9.2 EtherNet/IP (2001)
Rockwell Automation’s industrial Ethernet protocol:
Characteristics:
- CIP protocol: Common Industrial Protocol (same as DeviceNet). This shared protocol connects the Ethernet family to an established fieldbus family.
- Standard TCP/IP: Uses unmodified Ethernet. Ethernet alone does not establish deterministic timing; scheduling, clocking, switches, and device behavior also matter.
- Producer-consumer: Efficient multicast messaging. Messaging efficiency still needs to be assessed against the control loop’s timing budget under load.
- Widely adopted: North American manufacturing. Installed controllers, vendor tools, and maintenance skills help explain its fit in Rockwell-centered plants.
24.9.3 EtherCAT (2003)
Ethernet for Control Automation Technology, ultra-low latency:
Architecture:
Figure 6 answers how EtherCAT reaches short cycles using hardware processing in its devices.
Follow the single frame across Figure 6. The master sends one Ethernet frame into a chain of devices, passing a drive, an input module, a sensor and a robot. Each marker along the way carries the key idea, because a node reads the controller’s output data and inserts its own input data while the frame is still moving past, in hardware, rather than receiving it and sending it on. The return path then brings the whole set of answers back in one piece. Compare that with a scheme where the controller addresses each device in turn. Here the cost of adding another node is a few nanoseconds of delay, not another round trip, which is how one pass can serve a thousand points.
Performance:
- Cycle time: Can be engineered into sub-millisecond ranges for suitable motion and I/O systems
- Jitter: Controlled jitter is critical for synchronized motion
- Topology: Line, tree, star, or any combination
- Data processing: Each slave processes data as frame passes through
24.10 Putting Numbers to It
EtherCAT can achieve short cycle times through its “processing on-the-fly” architecture. The example below is illustrative; always verify device-level timing with vendor data and load tests.
Adding Ethernet frame transmission time for 1,500 bytes at 100 Mbps:
Total cycle time in this simplified model is . Real results depend on frame size, node count, device implementation, topology, synchronization settings, and other traffic on the network.
Use cases: High-speed motion control, packaging machines, robotics.
A robot factory gives a second scale example: Temperature Terry sees 50 robots building cars. Modbus polling resembles passing notes, PROFINET cyclic exchange resembles taking turns, and EtherCAT resembles a letter that each participant reads or adds to while it passes. OPC UA plays the translator role through a described information model. These analogies identify different jobs; they do not establish deadlines. When one robot welds a car door, its controller needs fresh position and drive data within a tightly bounded cycle. EtherCAT or PROFINET IRT can be engineered for that loop; ordinary Wi-Fi is a poor fit for its real-time path.
24.11 EtherCAT Cycle Time
The separate cycle-time workbench varies node count and frame size to expose the two terms in this simplified model: node-processing delay and time to send the frame on the wire. Its illustrative targets are below 100 μs for high-speed motion and synchronized robotics, below 1 ms for general factory automation and synchronized drives, and below 10 ms for process control and monitoring. These are application targets to investigate, not guarantees from the sum or a protocol name. The 25 ns per-node assumption needs device-level verification. Validate the complete cycle, jitter, topology, synchronization, timeout behavior, and recovery under the intended load.
24.12 Test Timing Against the Process
The conveyor establishes the need for timely action; water treatment shows why timing evidence must also remain tied to trustworthy measurements and a safe fallback.
Figure 7 grounds the protocol discussion in a closed-loop industrial decision: measure residual chlorine downstream, compare it with the allowed band, and adjust the dosing pump while preserving mixing and contact time.
Follow Figure 7 from inlet flow and chlorine dose through mixing to the residual sensor. The PLC compares the measured 0.68 mg/L with the 0.5–1.0 mg/L target band; the decision remains hold unless a bounded adjustment is justified. The evidence ledger must preserve sensor calibration, sample time, command, alarm state, and operator ownership, because a fast protocol cannot repair an untrustworthy measurement or an unsafe fallback.
Water treatment automation demonstrates critical infrastructure IoT where precise control directly impacts public health. Chlorination systems maintain safe disinfection levels while optimizing chemical consumption through real-time feedback control.
For a bounded release record, name the job, path, traffic load, settings, measured worst case, safe failure, owner, and next test. Reopen the choice when a machine, switch, traffic class, safety requirement, or plant layout changes. Protocol selection remains credible only while that evidence matches the installed system.
24.13 Summary
The conveyor’s stop and maintenance report need different communication duties. Start with cycle time, jitter, stale-data behavior, safety role, ownership, and installed assets. Modbus polling suits simple register access and brownfield links; PROFIBUS and DeviceNet remain useful where supported equipment, tools, and downtime constraints favor retention. PROFINET, EtherNet/IP, and EtherCAT provide different Ethernet control mechanisms. OPC UA and Sparkplug B belong in the integration discussion once the real-time boundary is protected.
Bandwidth headroom, clock alignment, and a calculated transmission time each provide part of the evidence. None replaces worst-case testing of the complete path under traffic, failure, and recovery. Keep measurement quality and operator ownership tied to the action, as the chlorination example demonstrates.
Choose the protocol for the job’s deadline, failure consequence, installed equipment, and integration boundary. Prove the complete path with measured timing and a safe response to loss.
Carry that evidence into Industrial Protocols: Selection and Security, where the families become a defensible selection record.

