Latency

What is Latency?

Latency (delay) is the time a message, data packet or signal takes to travel from source to destination, including any processing along the way, expressed in milliseconds (ms) or microseconds (µs). When both the outbound and return trip are measured, this is called the round-trip time (RTT), which is what the familiar ping command shows. In IT, latency mainly determines how responsive an application feels; in OT, it determines whether a control loop stays stable and whether a machine stops in time.

🧠 What are the components of latency?

Network latency is the sum of four delays that every packet accumulates on its way:

On the plant floor, the network is only part of the story. The end-to-end latency of a control loop is the chain sensor → Remote IO → network → PLC program → network → output module → actuator. You can therefore distinguish three types:

• Network latency - delay in transmitting data across Ethernet, fieldbuses or Wifi
• Processing latency - the time a controller needs to process data, often set by the PLC’s cycle time
• Device latency - response time and filtering in sensors, I/O modules and drives

On a short 100-metre plant connection, propagation is negligible (about 0.5 µs). There, transmission, queuing and above all device cycle times dominate.

⏱️ What latency is typical in IT and OT networks?

The table shows why real-time control stays local: a round trip to a cloud region easily takes longer than a complete PLC cycle.

🔄 What is the difference between latency, jitter and determinism?

• Latency - how long a message is in transit
• Jitter - how much that delay varies from packet to packet
• Determinism - the guarantee that a message always arrives within a fixed, predictable time
• Bandwidth - how much data can be sent per second; a wide connection is not automatically a fast one

For real-time control, a constant latency of 2 ms is often better than an average of 0.5 ms with spikes up to 20 ms. A controller can compensate for a fixed delay, not for an unpredictable one. That is why industrial networks focus on the worst-case latency rather than the average.

🏭 Why is latency critical for control loops and motion control?

Every delay in a control loop acts as dead time: the controller reacts to a measurement that is already out of date. In a PID controller this reduces the stability margin, so you either have to tune the loop more conservatively or risk overshoot and oscillation. The faster the process, the more this matters:

• Motion control - synchronised servo axes, robots and electronic cams need cycle times below 1 ms and jitter in the microsecond range, via EtherCAT or ProfiNET IRT
• Discrete manufacturing - communication between PLCs, HMIs and Remote IO usually needs a few to several tens of milliseconds
• Process industries - temperature and level control in energy, chemicals or water treatment is slower, but alarms and monitoring of critical processes must remain timely
• Safety functions - for safeguards such as a light curtain, the response time of sensor, safety controller and communication counts towards the safety distance, because ISO 13855 uses the overall stopping time of the system; an emergency stop and an SIS must also act within a defined response time

⏱️ Excessive or unpredictable latency can lead to unsafe situations, production problems, rejects or delays in emergency shutdowns.

📏 How do you measure latency in an OT network?

1. Decide what you are measuring - one-way latency, round trip or the full chain from sensor to actuator.
2. Start with a round trip - ping quickly gives an RTT, but ICMP gets low priority on many devices and says little about cyclic I/O traffic.
3. Analyse real traffic - with a network TAP or mirror port and Wireshark, you see the timestamps of request and response, and the spread (jitter) between cyclic frames.
4. Synchronise clocks for one-way measurements - with Precision Time Protocol (IEEE 1588), two measurement points can be compared to within microseconds.
5. Benchmark equipment - RFC 2544 describes how to test the latency, throughput and frame loss of switches and firewalls.
6. Use the controller’s diagnostics - PLCs and engineering software show scan times, update times and watchdog overruns.
7. Measure under load - the worst case appears during network congestion, backups or large file transfers, not on a quiet network.

🔧 How do you reduce latency in an industrial network?

Other factors that influence latency are distance (for example between field and cloud), the quality and configuration of switches and routers, the number of hops and the system load on servers or controllers.

🔐 What do security controls do to latency?

Security and latency pull in opposite directions. Every control that inspects or encrypts traffic adds processing time:

• Stateful firewalls - track connections and, depending on the model, add tens of microseconds or more
• Deep Packet Inspection - an industrial firewall that checks Modbus or S7 function codes has to unpack every packet; under load this can reach milliseconds
• VPN and TLS - encryption and extra headers cost processing time, especially on lightweight embedded hardware
• Passive IDS - listens via a mirror port or TAP and adds no latency to the traffic

IEC 62443 takes as a principle that security measures must not adversely affect essential functions. Place inspecting firewalls at zone boundaries in the Purdue Model, between cell and supervisory level or between OT and IT, and not in the middle of a deterministic I/O ring. Test the added latency under full load before you put a control into production.

❓ Frequently asked questions

What is a good latency for an industrial network?

A good latency depends on the application: motion control needs cycle times below 1 ms with microsecond jitter, standard PLC communication works with a few to several tens of milliseconds, and SCADA visualisation with hundreds of milliseconds. More important than a low average is that latency stays predictable within the cycle time.

What is the difference between latency and bandwidth?

Latency is the time a single message spends in transit, while bandwidth is the amount of data a connection can carry per second. A high-bandwidth connection can still have high latency, a satellite link being a classic example. For OT control, low and stable latency usually matters more than plenty of bandwidth.

What is the difference between latency and jitter?

Latency is the delay of a message, while jitter is the variation in that delay between successive messages. In industrial networks, jitter is often more harmful than slightly higher latency, because a controller can compensate for a fixed delay but not for a fluctuating one.

Can a PLC be controlled via the cloud?

Real-time control via the cloud is unsuitable for most control loops, because a round trip to a cloud region easily takes 10 to 100 ms and is not guaranteed. The cloud is therefore used for analytics, reporting and optimisation, while the PLC or an edge system runs the control locally with low latency.

Is 5G fast enough for industrial control?

5G URLLC is designed for a latency of 1 ms at 99.999% reliability, but that target requires a specific network setup, often a private 5G network. For mobile robots and AGVs, 5G is a serious option, while the fastest motion control usually stays wired with EtherCAT or ProfiNET IRT.

Does a firewall increase latency in an OT network?

A firewall always adds some latency, because every packet is checked before it is forwarded. With simple filtering this is usually negligible, but deep packet inspection and encryption can cause noticeable delay under load. Place firewalls at zone boundaries and test the latency before you put them into production.

📌 In summary

Latency is the delay between action and response within a network or system, and in OT what matters most is a low and predictable worst-case latency. Propagation, transmission, processing and queuing make up network latency, while the cycle times of PLCs and I/O fill the rest of the chain. With QoS, TSN, edge computing and careful placement of security controls, you keep OT processes under safe and accurate control.