F
Member
【Editorial Discussion】
A convenient Ethernet switch in the middle of a motion or I/O line can pass a ping and still spoil a cyclic bus. The first failure is often jitter or lost I/O connections under load, not a lack of average bandwidth. “It is only a few devices” is not a sufficient argument if the switch does not meet the profile the protocol expects.
1) Check whether the device is a managed switch with the required features (store-and-forward behaviour, QoS for the protocol’s traffic class, no energy-efficient idle surprises, and a documented forwarding latency) or an office unit that happens to link at 100 Mb/s. The label “Gigabit” does not answer the jitter question.
2) Measure cycle-time jitter and lost frames when the line is busy, including a burst of non-real-time traffic if that can appear in the cabinet. An unloaded lab line will not show a queueing problem that only appears when a camera or a laptop is also plugged in.
3) Topology matters. A star through one oversubscribed box, or a long daisy-chain of store-and-forward hops, adds delay that the controller’s watchdog may not tolerate. Count hops and the published latency, not only cable length.
4) If you must use a general-purpose switch for a non-critical segment, keep it off the isochronous or motion path and document the split. Mixing those roles on the same unconfigured box is how intermittent drops get blamed on the PLC.
Open question: have you had a line that pinged clean and still dropped I/O, later traced to an office switch or an unconfigured QoS path? What test do you now require before a switch is allowed on a cyclic bus?
A convenient Ethernet switch in the middle of a motion or I/O line can pass a ping and still spoil a cyclic bus. The first failure is often jitter or lost I/O connections under load, not a lack of average bandwidth. “It is only a few devices” is not a sufficient argument if the switch does not meet the profile the protocol expects.
1) Check whether the device is a managed switch with the required features (store-and-forward behaviour, QoS for the protocol’s traffic class, no energy-efficient idle surprises, and a documented forwarding latency) or an office unit that happens to link at 100 Mb/s. The label “Gigabit” does not answer the jitter question.
2) Measure cycle-time jitter and lost frames when the line is busy, including a burst of non-real-time traffic if that can appear in the cabinet. An unloaded lab line will not show a queueing problem that only appears when a camera or a laptop is also plugged in.
3) Topology matters. A star through one oversubscribed box, or a long daisy-chain of store-and-forward hops, adds delay that the controller’s watchdog may not tolerate. Count hops and the published latency, not only cable length.
4) If you must use a general-purpose switch for a non-critical segment, keep it off the isochronous or motion path and document the split. Mixing those roles on the same unconfigured box is how intermittent drops get blamed on the PLC.
Open question: have you had a line that pinged clean and still dropped I/O, later traced to an office switch or an unconfigured QoS path? What test do you now require before a switch is allowed on a cyclic bus?