Skip to content

Measuring response time

The response time is the time from setting a PLC output until the model’s answer arrives as a PLC input. It is the number that decides whether your PLC logic sees the model react in time. CONNECT can measure it without a PLC program: it plays the PLC’s part itself.

The measured loop CONNECT writes a PLC output signal and sends it to the browser. The browser applies it, a simulation step runs and a model component writes the answering PLC input signal, which goes back to CONNECT. CONNECT measures the time between writing the output and receiving the input. CONNECT writes the output Browser receives the value Simulation step model reacts Model writes the input signal answer returns - CONNECT stops the clock
CONNECT times the whole loop, from writing the output until the model's input arrives.

Each measurement repeats this many times and reports the statistics. It covers CONNECT, the connection to the browser, the browser’s signal handling and one simulation step. It does not include your PLC’s own scan time or the network between the PLC and CONNECT.

  • CONNECT is connected, and the model is loaded in exactly one tab that holds the document. If no tab has the model, or several tabs could answer, the measurement is refused.
  • A signal pair: one PLC output that CONNECT writes, and one PLC input that the model answers on. Both must be Bool signals with the right direction. Something in the model must switch the input when the output changes, for example a signal that a component or logic step writes in response.
  • Neither signal is forced, and no CONNECT signal mapping writes the input signal. The answer has to come from the model.
  1. Open the CONNECT page and find the Response time section.
  2. Pick the Output signal and the Input signal from the lists. They only show Bool signals of the matching direction.
  3. Optionally change Samples (default 200), Pause between samples in ms (default 100) and Timeout per sample in ms (default 2000).
  4. Click Measure. Progress shows as completed over requested samples.

The original value of the output signal is restored when the run ends. Only one measurement can run at a time.

For the most useful result, set the performance level first and keep the tab visible while you measure. To compare levels, run the same pair once at Low and once at High and note which level each result belongs to.

ValueMeaning
MeanAverage response time in milliseconds over all successful samples
Min / MaxFastest and slowest successful sample
P9595 % of the samples were at least this fast. Use it instead of the mean when you care about the worst typical case.
SamplesNumber of successful samples
TimeoutsSamples that got no answer within the timeout. A timeout is counted, never entered as a time.

The values stay empty until the first sample succeeds. A run that ends early keeps its partial result.

If no sample gets an answer, the page shows No answer on signal. Check that the model is loaded, that a component actually writes the selected input, and that the tab holds the document (see document ownership).

CONNECT refuses to start when the setup cannot give a valid result, and names the reason.

RefusalCause and fix
invalid-requestA value is out of range. Samples: 10 to 5000. Pause: 0 to 5000 ms. Timeout: 100 to 60000 ms. Both signals must be given and different.
busyAnother measurement is still running. Wait for it to finish.
unknown-signalA signal does not exist on the gateway. Check the name.
not-boolA signal is not a Bool. Pick Bool signals.
wrong-directionThe output must be a PLC output and the input a PLC input.
forcedA signal is forced. Release the force first.
mapping-writes-inputA CONNECT signal mapping writes the input. Disable the mapping or choose another input.
no-viewerNo browser with the model is connected to CONNECT. Open the model.
ambiguous-viewerSeveral tabs could answer. Close all tabs except the one that holds the document.

A running measurement stops as aborted with a reason, and you keep the partial result:

Abort reasonCause
viewer disconnectedThe answering tab closed or lost its connection.
interface disconnected: nameThe interface that owns one of the two signals left the Connected state.
signal removed: nameOne of the signals was removed or renamed during the run.
no answerA sample timed out, and the next one timed out again after CONNECT resynchronized.

While a measurement runs, the browser also records its own part of the loop. Read it with web_status and detail: "full", or with connect.loopProbe.status. The signalChain block contains:

FieldMeaning
active, probeId, outSignal, inSignal, reasonThe current or last announced run and why it ended
received, samples, supersededOutput edges received, samples completed, and edges replaced by a newer value before they were applied
receiveToCommitFrom receipt of the value to its application in the model, including the wait for the next simulation step
commitToSendFrom application until the answering input value is sent back
receiveToSendThe whole browser part of the loop

Each timing has meanMs and maxMs, empty until a sample is complete. The browser part is a subset of the gateway’s total. The difference is the transport time between CONNECT and the browser. Superseded and incomplete samples are left out.

Both commands work from the page and from an AI assistant: connect.loopProbe.run through web_execute, connect.loopProbe.status through web_query. A measurement is not recorded into macros.

  • connect.loopProbe.run - starts a measurement with outSignal, inSignal and optional samples, pauseMs, timeoutMs. It returns at once; the run continues in the background.
  • connect.loopProbe.status - read-only. Returns state (idle, running, done, aborted), reason, progress, the statistics and the signalChain block. The gateway offers the same through GET /diagnose/loop-probe; see REST and WebSocket API.
  • The loop shows what CONNECT and the browser add. It cannot measure your PLC’s acquisition or CONNECT’s write to the PLC.
  • Only the latest value of a signal is transmitted. A pulse shorter than a polling or sending step can be overwritten, so use signals that stay set long enough, as the measurement does.
  • A hidden tab, a busy computer or a long frame raises the maximum and the P95. Repeat the measurement on a quiet machine before you draw conclusions.