What is deterministic behaviour?
Deterministic behaviour is the property of a system that always produces the same output from the same initial state and the same input, and, in time-critical applications, always responds within a known, bounded time. In industrial automation it therefore covers two things at once: the right answer and the right moment. A PLC that makes the correct decision but sometimes switches 40 ms late is just as unpredictable to a machine as a PLC that makes a calculation error.
🧠 What two kinds of determinism exist?
In practice two meanings are used interchangeably. It helps to separate them:
| Kind | Meaning | Example of a violation |
|---|---|---|
| Functional (logical) determinism | Same state + same input always gives the same output | Two tasks write the same variable at once (race condition), so the result depends on execution order |
| Temporal determinism | The answer always arrives within a known upper bound (worst case) | A network frame sometimes waits behind heavy traffic and arrives milliseconds later |
Importantly, a deterministic system is not independent of its state. A counter, a timer or a PID controller remembers what happened before. Determinism means that this state plus the input fully fixes the outcome, without randomness or unknown outside influences.
Determinism is also different from speed. A system with a constant response time of 10 ms is more deterministic than a system that needs 1 ms on average but sometimes 50 ms. A control loop can compensate for a fixed delay, not for an unpredictable one. That is why OT engineers look at worst-case latency and at jitter, its variation, rather than the average.
🔄 How does determinism relate to real-time and reliability?
| Concept | Question it answers | Relation to determinism |
|---|---|---|
| Deterministic | Is the behaviour predictable and repeatable? | The foundation |
| Real-time | Is the deadline guaranteed to be met? | Hard real-time requires temporal determinism |
| Fast | How short is the average response time? | Independent; fast is not necessarily predictable |
| Reliable | How often does the system fail? | Independent; determinism does not prevent hardware failures, but it does make software faults reproducible |
| Available | What percentage of the time does it work? | Redundancy raises availability, but only deterministic failover avoids disturbing the process |
Real-time does not mean “fast” but “guaranteed on time”. In hard real-time a missed deadline is a system failure, in firm real-time a late result is worthless, and in soft real-time its value gradually degrades.
🏭 Why does deterministic behaviour matter in OT?
In office IT a buffer absorbs variation: nobody notices a web page loading 200 ms later. In OT, software drives physical processes, and timing has direct consequences:
| Application | Typical cycle time | Consequence of non-deterministic behaviour |
|---|---|---|
| Motion Control and motion controllers | 31.25 µs – 1 ms | Axes fall out of step, vibration, position errors |
| PLC control | 1 – 50 ms | Delayed or asynchronous actuation, wrong synchronisation between machines |
| Process control in a DCS | 100 ms – 1 s | Poorer control quality, overshoot, oscillation |
| SCADA visualisation | 250 – 1,000 ms | Inconsistent or delayed operator information |
| Systems with redundancy | Depends on the process | Failover is not synchronous and the process sees a bump |
| SIS and safety PLCs | Typically 10 – 100 ms | A hazardous situation is caught too late |
For safety functions, response time can literally be expressed in millimetres. Under ISO 13855, the total stopping time determines how far a light curtain must be placed from the hazard. With an approach speed of 2,000 mm/s, every additional 10 ms of response time costs 20 mm of extra distance. That is why functional safety under IEC 61508 requires the response time of a safety function to be specified in the safety requirements and demonstrably met.
Non-deterministic behaviour also leads to:
- Watchdog trips - a PLC goes to stop when the cycle time exceeds the configured watchdog time.
- False alarms or timeouts - communication that is occasionally late is treated as a fault.
- Difficult troubleshooting - a fault that only occurs under peak load is hard to reproduce during a root-cause analysis.
🔧 How is deterministic behaviour achieved technically?
Determinism arises in every layer: computation, operating system and network.
- Cyclic execution - a PLC following IEC 61131-3 reads all inputs into a process image at the start of each cycle, executes the program and then writes all outputs. Within one cycle the program therefore works with one consistent snapshot. A fixed cycle time instead of a free-running task keeps timing constant.
- Real-time operating system - an RTOS guarantees that a high-priority task pre-empts a running task within bounded time. Since Linux kernel 6.12 (November 2024) the PREEMPT_RT real-time extension is part of the mainline kernel, which makes soft PLCs on Linux easier to build.
- Worst-case execution time (WCET) - for critical tasks the longest possible execution time is determined, by measurement under load or by static analysis. Caches, pipelines and shared memory in multicore processors make that upper bound harder to establish.
- Deterministic networks - standard Ethernet is best effort. A single maximum-length frame already occupies a 100 Mbit/s link for about 123 µs. A critical frame queued in a switch behind a series of such frames is delayed, and without prioritisation or scheduling there is no fixed upper bound on that delay.
Industrial protocols each solve this in their own way:
| Technology | Mechanism | Typical performance |
|---|---|---|
| ProfiNET RT | Priority via VLAN tag, no TCP/IP stack for cyclic data | Cycles around 1–10 ms |
| ProfiNET IRT | Reserved time window per cycle, synchronised clocks | Cycle from 31.25 µs, jitter below 1 µs |
| EtherCAT | One frame passes all slaves on the fly, distributed clocks | Jitter well below 1 µs, cycles of 100 µs or less |
| TSN | IEEE 802.1AS clocks, time-aware queues (802.1Qbv), frame preemption | Bounded latency on standard Ethernet, provided it is configured end to end |
| PTP (IEEE 1588) | Common time base for all devices | Synchronisation in the sub-microsecond range |
🔐 Which IT/OT developments threaten determinism?
IT/OT convergence puts determinism under pressure, because IT technology is designed for average performance:
- Virtualization and containers - a hypervisor or container host shares CPU time among several workloads. A virtual PLC needs pinned, isolated CPU cores and a real-time kernel.
- Shared networks - office traffic, backups or video on the same network enlarge queues. Network segmentation, VLANs and QoS limit that influence.
- Security measures - an industrial firewall with deep packet inspection adds latency, and active vulnerability scans can load a PLC so heavily that its cycle is disturbed. That is why ISASecure EDSA certification, subsumed into the CSA programme in 2019, used communication robustness tests to check that a device maintains its essential functions under malformed and flooding network traffic.
- Wireless - Wifi has greater delay variation than cabling; 5G URLLC promises 1 ms at 99.999% reliability as a target value, not as a guarantee for every network.
- AI models - an AI model that gives different answers to the same question is functionally non-deterministic. Let such applications advise, and keep the deterministic control itself in the PLC.
🧭 How do you verify deterministic behaviour in practice?
- Specify the requirement - for each function, the maximum response time and allowed jitter, derived from the process and the safety analysis.
- Choose the right layer - motion and synchronisation in EtherCAT, ProfiNET IRT or TSN; control in cyclic PLC tasks; supervision in SCADA.
- Design with margin - keep the measured execution time well below the configured cycle time, so extensions and peak load do not immediately cause watchdog trips.
- Measure the worst case - record cycle time, maximum cycle time and jitter under full load during the FAT, not only at rest.
- Test disturbances - simulate network peaks, cable breaks and failover of redundant components.
- Monitor in operation - track cycle-time statistics and network diagnostics so gradual degradation, for example after a software change, becomes visible early.
❓ Frequently asked questions
What does deterministic mean in automation?
Deterministic in automation means that a control system always performs the same action given the same state and input, and does so within a guaranteed maximum time. It is about a predictable outcome and predictable timing. Without both, a machine cannot run safely and reproducibly.
Is standard Ethernet deterministic?
Standard Ethernet is not deterministic, because frames can wait in switches behind other traffic with no fixed upper bound on that delay. More bandwidth reduces the waiting time but gives no guarantee. Industrial protocols such as ProfiNET IRT, EtherCAT and TSN therefore add time synchronisation and scheduled time windows.
What is the difference between deterministic and real-time?
Deterministic behaviour means that outcome and timing are predictable, while real-time means that a deadline is guaranteed to be met. Hard real-time is only possible with temporal determinism, but a deterministic system can also have a long, predictable response time. Real-time is about the deadline, determinism about predictability.
Why does jitter matter for deterministic behaviour?
Jitter is the variation in delay and therefore a direct measure of how deterministically a system behaves in time. A controller can compensate for a constant delay, but not for a varying one. For motion control, jitter below 1 µs is common, while for process control a few milliseconds are often acceptable.
Can a virtual PLC be deterministic?
A virtual or soft PLC can behave deterministically if the underlying computer is set up for it, with a real-time kernel, pinned and isolated CPU cores and a deterministic fieldbus. Without those measures the PLC shares compute time with other workloads and its cycle time becomes unpredictable. For fast motion control, dedicated hardware therefore remains common.
How do you measure whether a system is deterministic?
You measure deterministic behaviour by recording the maximum cycle time, the response time and the jitter under realistic peak load and during faults, not just the average at rest. PLCs provide cycle-time statistics for this and networks offer diagnostic functions. The measured worst case must stay well within the requirement of the process.
📌 In summary
Deterministic behaviour means a system always does the same thing given the same state and input, and always within a known maximum time. For OT it is a fundamental requirement, not a luxury. Cyclic PLC tasks, an RTOS, deterministic networks such as EtherCAT, ProfiNET IRT and TSN, shared time via PTP and separation from IT traffic together deliver predictable, safe industrial automation, even as IT technology enters the factory.
