IMDHMIControls LibraryShop main store

Explainer

Why Modbus polling is not the same as remote I/O

Controllers and I/O Published 8 min read

Short answer

Polled Modbus moves each remote point one request-response transaction at a time, so register data refreshes on the poll cycle — commonly hundreds of milliseconds — not in step with the controller's I/O scan. Native remote I/O refreshes the whole remote image in a time close to one scan and faults deterministically on node loss. Use polled Modbus for supervisory points that tolerate its measured cycle, and native remote I/O when interlocks or short pulses need every-scan behavior.

On this page
  1. What is the actual difference between Modbus polling and remote I/O?
  2. How long does one Modbus poll cycle really take?
  3. Can I make Modbus polling fast enough by raising the baud rate?
  4. What breaks when remote points are polled instead of scanned?
  5. What does scan-synchronous remote I/O give you that polling does not?
  6. Which architecture does my machine need?
  7. How do I test polling timing and failure behavior before commissioning?
  8. Questions asked

A PLC that acts as a Modbus master can read and write every register in a remote Modbus I/O rack, and the rack will work. What you do not get is remote I/O behavior: each point moves when one request-response transaction completes, so the data in your program refreshes on the poll cycle, not in step with the controller's I/O scan. On one serial port shared by 16 slaves and a 50 ms scan, that cycle is 800 ms. Pulses shorter than the cycle can disappear, and a dead link leaves the last good values in the registers. Supervisory points tolerate that; every-scan interlocks do not. Here is the timing arithmetic, what breaks, what native remote I/O does differently, and the decision for each case.

What is the actual difference between Modbus polling and remote I/O? #

Modbus polling is data exchange: the PLC issues a request, the remote device answers, and the value lands in a register once the transaction completes; remote I/O is an architecture: the controller's I/O scan services the remote image every scan, in a time close to one scan.

The phrase "Modbus master" describes a communication role, not an I/O service model. That is how one controller can carry Modbus master support in its communication documentation while a separate FAQ for the same product says it supports no remote I/O: both statements are true, and neither implies the other. One answers a question about transactions, the other about I/O architecture.

Master capability also certifies nothing about a particular slave. Register layouts, coil versus register types, data widths, byte order and update behavior are properties of the slave, confirmed against its documentation and the controller's master setup before anything is wired. Call the arrangement what it is — a polled data exchange — because the timing and stale-data risks you are about to design around are exactly what the remote I/O label hides.

How long does one Modbus poll cycle really take? #

With one request issued per controller scan, any one slave's data can be about 850 ms old when the logic uses it.

Worked example — 16 slaves on one serial port at 9600 baud:

  • On most PLCs the master gets one transaction per scan, so the scan time is the minimum interval between requests.
  • A read of 10 holding registers puts 8 characters on the wire out and 25 back, 33 total. Modbus RTU serial framing carries 11 bits per character — start, eight data, one parity, one stop — so the wire time is 33 × 11 = 363 bits; at 9600 bit/s that is 363 ÷ 9600 = 0.038 s, about 38 ms before turnaround gaps.
  • The per-slave slot is the slower of the two limits: scan-time pacing versus 38 ms of wire. Around the ring, 16 slots give the cycle stated above.
  • A value captured at the start of its slot then sits through 15 more slots and up to one scan before use: the full cycle plus one scan of worst-case age, the 850 ms figure above.

The structural limit is the single serial port shared by every slave; why Modbus RTU drive polling on one serial port is slow works the same arithmetic from the drive side. These numbers belong to this example — your scan time, baud rate and register counts move them — but the shape never changes: data age is set by the poll cycle, and the poll cycle is slave count times slot time.

Can I make Modbus polling fast enough by raising the baud rate? #

Raising the baud rate cuts the wire time — the same 33-character transaction takes about 3 ms at 115.2 kbaud instead of 38 ms at 9600 — but a master that issues one request per scan still caps the cycle at 16 slot times.

Baud rate only touches one of the two numbers in the slot. Once wire time drops below the request interval, more baud buys nothing, and 56 kbaud to 115.2 kbaud is about where RS-485 equipment tops out in practice. The standing guidance holds either way: do not reach for Modbus RTU when the application needs fast data response.

Squeezing the interval is its own trap. Devices exist that stop answering when polled every 30 ms — a reported field symptom, not a specification — so the shortest sustainable interval is a property of the slave, not of your schedule. Modbus TCP removes the shared serial wire, and 100 ms polling of a modest node count sits well inside its documented capability, but the transaction model is unchanged and everything in the next section still applies.

What breaks when remote points are polled instead of scanned? #

Three things: values in the program can be one or more poll cycles old when used, input pulses shorter than the poll cycle can be missed entirely, and a dead link leaves the last successfully read values sitting in the registers.

Age is the quiet failure. Logic that interlocks on a remote input acts on the last completed read, which in this example is up to 850 ms behind the physical contact. Pulses are the loud one: the read samples the register at one instant, so a 200 ms part-present pulse inside an 800 ms cycle is caught only when it overlaps that instant — 200 ÷ 800, one chance in four in this example. Polled inputs are level data, never event data; counting pulses takes hardware at the node, which is why PC-polled remote I/O cannot count encoder pulses.

Writes take the same path in reverse, so command-to-physical-output delay runs up to a poll cycle. When the link dies, the master reports transaction timeouts while the registers keep their last values and the outputs hold their last commanded states until your program acts on the timeout status. That failure handling is your design job — the same discipline, seen from the writing side, is covered in safe Modbus writes to PLC output registers from SCADA. One safety line belongs here: no personnel-protection function rides on polled remote points — safety functions use safety-rated I/O, and polled points carry sequencing and supervision, with outputs forced to a defined state on communication failure.

What does scan-synchronous remote I/O give you that polling does not? #

A native remote I/O system moves the entire remote image — every input and output — every network cycle, completing all of it in a time close to one controller scan, and node loss trips the network's own watchdog and faults the controller instead of leaving stale values in place.

The scanner works through a configured node list each cycle and moves the whole image whether or not anything changed. The classic PLC-5 remote I/O scanner is built exactly this way, and PLC-5/80E scan-list loss on remote I/O after a power cycle is what it looks like when that list is the part that breaks. Node loss is not something you program: the network watchdog or connection timeout marks the node failed within a defined time and the controller enters its I/O-fault behavior with outputs going to their configured state.

The trade is ecosystem. Native remote I/O adapters and networks belong to the controller's family, so you buy determinism and designed failure behavior with vendor lock-in. Modbus connects you to any slave that speaks it, and hands you the timing and failure design as your own work.

Which architecture does my machine need? #

If the remote points only supervise — temperature, level, run status, setpoints — and 100 ms-class updates are acceptable, polled Modbus meets the requirement; if logic interlocks on remote points every scan, or must catch pulses at or below the poll cycle, use native remote I/O or move those points local.

The decisions, case by case:

  • Supervision and slow process variables with updates of 100 ms or slower: polled Modbus, TCP where the nodes sit on Ethernet. Compute the poll cycle with the arithmetic above, and design the stale-data flag and output safe-stating before commissioning.
  • Every-scan interlocks, or pulses at or below the computed poll cycle: native remote I/O, or local I/O in the controller base. If the poll cycle is not comfortably inside your required response time, the decision is already made.
  • Personnel protection: safety-rated I/O either way.
  • Mixed machines, the common case: put the handful of points that matter into native remote I/O or the local base, and poll the rest.

For failure detection, count consecutive failed transactions and declare the node lost when that count times the poll cycle exceeds the stale time you can tolerate; at that point, force the node's outputs to their defined safe state rather than holding them.

How do I test polling timing and failure behavior before commissioning? #

Measure two numbers on the installed system — input-change-to-register-update time and command-to-physical-output time — then break the link on purpose and record what the program sees and what the outputs do.

Wire one input and one output as a test pair, toggle the input, and timestamp physical change to register change; repeat under representative program and communication load, because a single successful read proves connectivity, not timing. Then disconnect the link: confirm how the master reports the failure, what value the program sees, that the outputs reach the state you defined for a lost node, and that data resumes correctly on reconnect. File the measured times and the tested failure behavior with the rest of the remote I/O commissioning checklist for a modular station.

Questions asked

Does issuing a poll every scan make Modbus behave like remote I/O?

No. With a single slave polled once per scan, data age can approach one scan — but one transaction still completes after another, a multi-slave round-robin staggers ages across the cycle, and nothing in Modbus provides node-loss detection. That watchdog behavior is what a native remote I/O scanner brings.

My scan is 10 ms, so is the poll cycle 80 ms with 8 slaves?

No — the per-slave slot is the larger of the request interval and the wire time. At 9600 baud the 33-character, 10-register transaction still occupies the wire for about 38 ms, so 8 slaves need roughly 8 × 38 ms, about 300 ms. A faster scan only helps once it is the slower of the two numbers.

Is Modbus TCP fast enough for sub-100 ms updates?

Polling a modest node count every 100 ms sits well inside documented Modbus TCP capability on an uncongested switch. At faster update rates you are past what that capability statement covers, so measure on your own hardware: the floor is set by transaction processing in the master and each slave, not by the wire.

How do I size the timeout for a polled node?

Count consecutive failed transactions and declare the node lost when the count times the poll cycle exceeds the stale time you can tolerate. On the 2.4 s example — three misses on an 800 ms cycle — that is the point to force that node's outputs to their defined safe state rather than hold them.

Keep reading

  • How-to

    How to run analog I/O over a Modbus TCP bus coupler

    Yes: remote analog I/O behind a Modbus TCP bus coupler can close a control loop from the PLC. Prove the electrical path with the network out of the loop, bring up the connection with a real timeout, map and scale the channels against known test points, then close the loop on measured timing — one documented integration refreshes coupler data every 10 ms, but the valve and hydraulics set the real response, and a failed exchange is a fault state, never fresh data.

    Controllers and I/O

  • Troubleshooting

    Why PC-polled remote I/O cannot count encoder pulses for SCADA

    A 100 PPR quadrature encoder on a line at 200 m/min produces 333 Hz on channel A, about 1.3 kHz once the counter decodes quadrature at 4x. No PC poll, and no 4 ms discrete input, can follow that rate. Count the pulses in a PLC high-speed counter input and let the SCADA sample the accumulated register.

    Controllers and I/O

  • How-to

    Remote I/O commissioning checklist for a modular station

    Commission a modular remote I/O station in this order: match the coupler's protocol and addressing to the controller interface, budget and verify station power, seat modules and connectors against the drawing, then prove cyclic data exchange and every field signal before enabling outputs. A loading web page and lit status LEDs prove the coupler is alive; only a value that changes at the controller proves valid I/O data.

    Controllers and I/O