IMDHMIControls LibraryShop main store

Explainer

Designing a 1734-AENT Point I/O fiber ring on EtherNet/IP

Industrial networks Published 8 min read

Short answer

Yes — enable IGMP snooping on all seventeen switches: the input data a 1734-AENT adapter produces is multicast, so without it every switch floods the ring and all sixteen adapters reject each other's packets. Full-duplex switching ends collisions but not congestion, each fiber span must meet its own optic's distance limit, and ring recovery must beat four times the RPI.

On this page
  1. Does this ring need IGMP snooping when there is only one controller?
  2. Will sixteen adapters collide when they all send to the PLC at once?
  3. How does a ring topology avoid an Ethernet loop?
  4. How fast must the ring recover when a fiber span is cut?
  5. How far can each multimode fiber span run?
  6. Is ControlNet the safer choice for this application?
  7. What should I prove at commissioning?
  8. Questions asked

Enable IGMP snooping on all seventeen switches — that one setting decides whether this network behaves. A 1734-AENT adapter sends its input data as multicast, so without snooping every switch floods it to every port and all sixteen adapters spend their cycles rejecting each other's packets. Switched full-duplex links end collisions but not congestion, the ring must block one link and recover from a cut faster than the controller's I/O timeout, and each multimode fiber span must satisfy its own optic's distance limit rather than the site's 8 km total. The numbers for a 1756-L61 with a 1756-ENBT and sixteen adapters follow, one decision at a time.

Does this ring need IGMP snooping when there is only one controller? #

Yes — enable it everywhere: the input data these adapters produce is multicast, so without IGMP snooping every switch floods it to every port and all sixteen adapters process and reject each other's traffic, about 750 unwanted packets per second each at a 20 ms RPI.

One receiver does not reduce multicast; it only means the flood lands on ports that never asked for it. A switch that is not snooping has no record of which ports joined which group, so it forwards group traffic out every port in the VLAN. Each adapter then receives fifteen other adapters' I/O streams, examines every frame, and discards it.

Worked example with illustrative values (RPI 20 ms, one input connection per adapter): each adapter produces 1 ÷ 0.020 s = 50 input packets per second, so the ring carries 16 × 50 = 800 multicast packets per second in total. Without snooping, every adapter receives the other fifteen streams: 15 × 50 = 750 packets per second of unwanted traffic. With snooping, the controller side joins each group and the switches forward a group only along the path to that port: each adapter receives zero foreign packets.

Snooping also has a dependency people miss: the forwarding tables age out unless something queries membership. Give the switches IP addresses on the I/O VLAN and enable the IGMP querier function on one of them, or the containment decays back to flooding. A drive fault traced to IGMP snooping configuration is a documented way this fails in the field, and it pays to read the factory IGMP snooping defaults on managed switches rather than assume what your units ship with.

Will sixteen adapters collide when they all send to the PLC at once? #

No — every link in a switched full-duplex network is collision-free by construction, so sixteen adapters transmitting in the same instant cannot collide; their frames converge in switch queues toward the controller, and the busiest span carries about 0.4 Mbps in this design.

Each fiber link gives the switch port a dedicated transmit and receive pair, so the collision backoff the question imagines never runs. What remains is congestion: when several adapters send at once, the switch serializes the frames onto the shared span and buffers what does not fit. Whether that matters is arithmetic on the offered load. The worst link is the span next to the controller switch, which carries all sixteen streams: 800 packets per second. At the 64-byte minimum Ethernet frame that is 800 × 64 × 8 = 409,600 bits per second, about 0.4% of a 100 Mbps link; at a generous 200-byte frame it is 800 × 200 × 8 = 1.28 Mbps, about 1.3%.

Queues build only when offered load approaches capacity, and the only mechanism in this network capable of producing that load is the multicast flooding above. So set speed and duplex explicitly on both ends of every fiber link, then watch queue, discard and error counters; the collision counters will read nothing forever. The RPI sets the packet rate in that arithmetic, and sizing a Generic Ethernet Module connection for WAGO 750 I/O shows the same calculation from the connection-configuration side.

How does a ring topology avoid an Ethernet loop? #

Managed switches running a ring protocol block exactly one of the seventeen fiber links during normal operation, so the live topology is a line even though the cable is a ring, and the blocked link enters service only after a break.

A ring with every port forwarding is a Layer 2 loop: the first multicast frame circles and multiplies until the network is unusable. The loop-prevention protocol — a vendor ring mechanism with declared ring ports, or RSTP — holds one link blocking, which turns the physical ring into an active line from the controller switch around to its neighbor. Three requirements follow: every switch runs the same protocol, because ring mechanisms generally do not interoperate across vendors; the ring ports are configured as ring members; and exactly one link stands blocked at any time. Commit to one switch family for all seventeen nodes — a mixed-vendor ring running two half-protocols is not a supported design.

How fast must the ring recover when a fiber span is cut? #

Faster than the I/O connection timeout: on these controllers that timeout defaults to four times the RPI, so 80 ms at a 20 ms RPI — classic spanning tree at roughly 50 s misses that bar by a factor of 600, RSTP at a second or two misses it by an order of magnitude, and a ring protocol rated under 50 ms passes.

The multiplier sits in the module's connection properties; four is the usual factory value, and confirming it there takes a minute. The tolerance arithmetic at a 20 ms RPI: 4 × 20 ms = 80 ms of link interruption before the controller declares the connection lost. Classic spanning tree spends 20 s detecting the failure plus 2 × 15 s in listening and learning, about 50 s in total. RSTP cuts that toward a second or two — still ten times the bar. Managed-switch ring protocols publish a recovery figure, typically tens of milliseconds; take the number from the datasheet of the exact model you are buying and hold it to half the tolerance, 40 ms, for margin.

Then prove it by pulling a span under load during commissioning. After a cut the ring runs as an unprotected line until the fiber is repaired, so route the ring-health status to the controller as an alarm rather than silent state.

How far can each multimode fiber span run? #

Plan per span, not per site: about 550 m per span for gigabit SX optics on 50 µm multimode, 2 km for 100 Mbit/s FX-class optics on multimode, and any vendor's higher multimode figure only with that datasheet's exact optic-and-fiber pairing in hand.

Optic class Fiber it is specified for Design distance per span
Gigabit SX, 1000 Mbit/s 50 µm multimode (OM2 or better) 550 m
FX-class, 100 Mbit/s multimode 2 km
Single-mode optic single-mode per the optic datasheet

The site's 8 km total is irrelevant to any of this: fiber distance is a per-hop limit. This ring has seventeen switches, therefore seventeen spans, and 8000 m ÷ 17 = 470 m per span if the route is even — inside the 550 m SX budget with about 80 m of margin. Averages are not design inputs, though: list the actual route length of each span and put it against its optic's row, and upgrade any hop that exceeds its row on that hop alone. Vendors quoting 4-5 km over multimode are quoting named transceivers on named fiber grades, so treat that as a claim about a specific optic rather than a planning number. The wrong optic-fiber pairing fails in its own way, as this analysis of QSFP28 CWDM4 optics failing on multimode fiber shows, and every span still deserves a measured insertion-loss check against its budget before the ring is trusted.

Is ControlNet the safer choice for this application? #

No — for reporting sixteen breaker states and issuing close commands on a dedicated network, this EtherNet/IP design runs at roughly 0.4% of one link's capacity with recovery inside the I/O timeout, and ControlNet's scheduled guarantee addresses a bound this duty cycle never approaches.

ControlNet's strength is a scheduled, time-slotted network in which every node's transmission time is guaranteed — the right tool when you must prove a hard worst-case bound. This application's bound is human-scale: a breaker state a tenth of a second old, and a close command that leaves within a second, are already beyond what an operator or a switching sequence can perceive, and the design above delivers far better than that on hardware the site needs anyway. The price of ControlNet here is a second network type, its taps and trunk cable, and node-schedule maintenance for a performance property nothing in this duty cycle uses. Decision: EtherNet/IP on the dedicated ring, with snooping, one blocked link, and per-span validation as above. One line that does not move: breaker opening and protection stay hard-wired to the protection relays — this network monitors and closes, and it must never become the path that opens a breaker.

What should I prove at commissioning? #

Pull one fiber span under load and demand three results — recovery inside the I/O connection timeout, multicast present only on the path to the controller, and zero discards or errors on every port after a soak — because each decision above fails silently when it is wrong.

  1. Before power-up, record every span's route length, fiber grade, optic type and measured insertion loss; upgrade any span outside its table row before testing starts.
  2. With the ring healthy, confirm exactly one blocked link in the ring status and full duplex negotiated or set on all seventeen switches.
  3. Inspect the switches' multicast forwarding tables: each adapter's group may appear only on ports along the path to the controller, and the querier must show alive.
  4. Force all sixteen breakers to change state in the same second and verify update latency at the 1756-L61 with zero queue discards on the busiest span.
  5. Pull the longest span under that same load, record recovery time against half the connection timeout (40 ms at a 20 ms RPI), restore the fiber, and confirm the ring returns to one blocked link.
  6. Bring ring health and connection faults to the operator as alarms before handover.

Questions asked

What RPI should I set for breaker status monitoring?

100 ms suits this duty: a state a tenth of a second old is fresher than any operator reaction, and close commands are operator-paced anyway. It also cuts the multicast load by five compared with 20 ms, so an adapter would face 150 flooded packets per second instead of 750 if snooping ever failed. Reserve 20 ms and faster for networks that will later carry interlocks.

Can I split the ring onto two subnets with A and B fiber ports instead?

You can, but it solves a different problem. Two subnets on A and B ports keep traffic classes apart on shared switches, which matters when SCADA or office traffic rides the same hardware. It does not by itself survive a fiber cut. A ring protocol with one blocked link gives cut protection on a single subnet, and IGMP snooping applies to either arrangement.

What happens if ring recovery misses the connection timeout?

Every adapter connection times out and the controller raises module faults until the connections rebuild. If you cannot buy recovery under four times the RPI, set the adapter modules' fault action so the controller keeps running, route the connection loss to an alarm, and let the operator retry the close command. Never leave a dropped span silent.

Can I mix single-mode fiber into the long spans?

Yes. Each span is an independent Layer 1 link, so fit single-mode optics and single-mode fiber on any hop that exceeds the multimode rows and leave multimode on the short hops. The rest of the ring does not change: verify the long hop against its own optic datasheet and measure its insertion loss like any other span.

Keep reading

  • Troubleshooting

    Sizing a Generic Ethernet Module connection for WAGO 750 I/O

    Enter the Generic Ethernet Module connection from the coupler's process image: input assembly instance 107 sized to input-image bytes plus one, output assembly instance 101 sized to the output-image bytes exactly, configuration assembly instance 1 at size 0, Comm Format SINT, and the remaining settings at their defaults. The 750-352 orders analog data before digital, so build both byte maps from the coupler's image view and round the totals once, never slot by slot.

    Controllers and I/O

  • Troubleshooting

    Why a Schneider M251 does not appear in RSLinx

    RSLinx and FactoryTalk Linx browse one transaction: the EtherNet/IP List Identity request, and a device appears only when the port you are cabled to answers it. Ping and Schneider software prove IP reachability, nothing more. Enable the M251's EtherNet/IP service on the connected port, browse through the right PC adapter, and register the M251 EDS when the entry shows a red X with a yellow question mark.

    Industrial networks

  • How-to

    Reach a CompactLogix through a ControlNet bridge in RSLinx

    When autobrowse through a 1784-U2CN ControlNet path lists the bridge's EtherNet/IP module but no CompactLogix or third-party EtherNet/IP devices, add the target IP address manually under that module's Ethernet port in RSWho or the FactoryTalk Linx browse tree, then select the resulting device as the online path in the Logix programming software. The devices are reachable; only discovery fails, and a manual entry needs no discovery.

    Industrial networks