Modbus RTU Polling: Budget for a Silent Device Before Setting the Scan Rate
Updated: Nov 5
A Modbus RTU network can look fast during commissioning and become slow as soon as one instrument stops replying. The working devices may still communicate correctly, yet their values arrive late because the polling master spends time waiting for the silent address. Increasing the baud rate alone does not remove that waiting time.
Before promising a scan rate, budget both the healthy polling cycle and the cycle with a failed device. Include the number of requests, response timeouts, retries and delays. Then verify the result on the actual system. This guide concerns sequential unicast polling on one serial RTU bus; it is not a timing guarantee for every TCP gateway, wireless link or PLC driver.
Start with requests, not just instrument count
The Modbus serial specification describes a master that starts one transaction at a time. It sends a unicast request and waits for a reply or an error outcome before proceeding. A silent device can therefore occupy time that would otherwise be available for another transaction.
Count the requests needed to obtain the required data. A meter read through four separate register groups creates four transactions, even though it is one physical device. Conversely, a supported contiguous block can return several values in one read. Confirm permitted register blocks, function codes and lengths against the exact manual; an unsupported block is not a useful optimisation.
Use a real device list
For example, the ACREL DTSD1352-C family provides RS485 Modbus RTU communication, as documented by Acrel. A project worksheet should identify the exact meter variant and register-map revision, rather than simply listing “Acrel meter”. The photographed catalogue instrument below illustrates a device that could be included in such a worksheet. No response-time or inverter-compatibility claim follows from the photograph.
Record the address, serial settings, requested register groups, expected update requirement and the responsible master for each device. A second independent master must not be added to the same RTU bus as a shortcut to faster reads. Where an inverter already controls meter communications, have the integrator assess an approved architecture before adding monitoring equipment.
Keep four timing terms separate
Transaction duration is the time a successful exchange actually occupies. Response timeout is the configured waiting limit when an acceptable reply does not arrive. Retry count determines whether another attempt follows a failure. The interval between frames or requests adds further waiting, including after errors where the implementation specifies it.
Schneider Electric’s serial-master documentation distinguishes response timeout from time between frames and permits node-specific timeout configuration. Other drivers expose different names and scopes. Check whether a setting counts retries after the initial attempt or total attempts, and whether it applies per request or per device. Identical-looking values can produce different cycle times.
Worked example: one silent device
Consider a hypothetical sequential master with three devices and one read per device. Assume each healthy transaction, including its normal gap, takes 100 ms. The healthy cycle is approximately 3 × 0.1 = 0.3 seconds. These are assumed planning figures, not measured performance of the pictured meter.
Now assume one device is silent. The master allows a one-second response window and two retries after the initial attempt. That creates three waiting windows: 3 × 1 = 3 seconds. Add the two healthy exchanges, 0.2 seconds, and the cycle is at least about 3.2 seconds before including the failed requests’ transmission time and any additional retry gaps, processing or cooldown delays.
A one-second collection target that seemed comfortable under healthy conditions is no longer achievable with that retry policy. If the silent device has four separate reads and the driver retries every read independently, those same assumed windows could consume 4 × 3 = 12 seconds. A driver that suspends further reads after a device failure may behave differently. Verify which case applies.
Do not shorten the timeout blindly
A timeout must accommodate legitimate device processing and communications delays under the expected load. Setting it below valid response time can turn a slow response into an avoidable failure. Continental Control Systems’ timing guidance shows that acceptable reply timing can depend on the meter’s firmware; its WattNode examples are not default settings for other manufacturers.
Espressif also documents the problem of late responses arriving after a master has moved to another transaction, and provides implementation-specific serial cooldown behaviour. The lesson is to account for late-response handling in the selected driver. Do not copy its configuration symbols into unrelated gateways or assume every driver clears pending traffic the same way.
Check freshness separately from refresh speed
An HMI can redraw an old value rapidly. Specify how the application identifies a valid new sample, retains the last successful-read time and exposes a communication fault. Decide how stale values should be displayed or excluded from calculations. These are application requirements to implement and test, not automatic benefits of selecting Modbus.
For a commissioning record, capture healthy cycle duration, transaction timings, retries and the age of values during an authorised simulated outage. Record behaviour when the device returns, including whether queued work causes a burst of requests. Perform interruption tests only under an approved plan that protects the process; do not disconnect an operating control or protection instrument casually.
Turn the timing budget into a design decision
If the fault case exceeds the required data age, review the request grouping, priority schedule and supported failed-device retry policy. Separate slow energy totals from faster operational measurements where the driver supports it. Additional independent serial segments may help distribute work, but require compatible hardware and an engineered design.
Send the device list, register groups, driver details, measured timings and required maximum data age for review. A realistic polling specification states what happens during a missing-device fault as well as during normal operation. Mains installation, process controls and safety functions require competent people and applicable requirements; a fast polling cycle does not establish a certified safety function.
Sources and pictured equipment
Need help with your application?
For instrument enquiries, visit ProSense Instruments, email sales@prosenseinstruments.com.au or call 1800 560 854. Include the device list, register groups and required maximum data age.
For monitoring and automation projects, visit ProSense Solutions or email sales@prosense.com.au.
For software enquiries, visit ProSight Software or email info@prosightsoftware.com. Device and protocol integration must be confirmed for your application.




Comments