Multicast Traffic and IGMP Snooping Configuration in EtherNet/IP Networks

How is IGMP snooping configured for EtherNet/IP? Implicit/explicit CIP, RPI and pps load, querier placement, choosing unicast, an example bandwidth calculation and a switch checklist.
Technical summary
Technology
EtherNet/IP, CIP, IGMP, IGMP snooping, Cisco IOS
Topic
Managing EtherNet/IP multicast I/O traffic
Level
Intermediate
Prerequisites
  • Basics of VLAN and switch configuration
  • The EtherNet/IP I/O connection concept
In this article8 sections

Behind the complaint “the network has slowed down and I/O connections drop now and then” on an EtherNet/IP line, there is often a single mechanism: because the switch does not recognise multicast I/O traffic, it floods it to every port. People look at the cable, the device, the firmware; the switch configuration should be checked too. Below, starting from the CIP messaging types, we describe the effect of RPI on network load, why multicast connections cause problems, the correct placement of IGMP snooping and the querier, and the choice between unicast and multicast. At the end there is an example bandwidth calculation and a switch checklist.

CIP messaging types: implicit and explicit

EtherNet/IP carries the CIP (Common Industrial Protocol) application layer over standard Ethernet and TCP/IP. CIP uses two types of messaging, and their behaviour on the network is very different:

Feature Explicit messaging Implicit (I/O) messaging
Purpose Parameter read/write, diagnostics, configuration Cyclic I/O data
Transport TCP 44818 (some services UDP 44818) UDP 2222
Scheduling Request–response, irregular Periodic, by RPI
Addressing Unicast Unicast or multicast

Implicit traffic drives the design of industrial network infrastructure. An I/O connection has two directions: O→T (originator to target), from the scanner (usually a PLC) to the adapter, and T→O (target to originator), returning from the adapter to the scanner. The O→T direction is unicast; the T→O direction can be unicast or multicast depending on the connection setting. Multicast T→O is designed so that several consumers (e.g. two PLCs or a “listen only” connection) can receive the same adapter data from a single stream.

The effect of RPI and packet rate on network load

RPI (Requested Packet Interval) determines the interval at which data is sent on a connection. The number of packets per second in each direction is roughly 1000 / RPI(ms). A connection with a 10 ms RPI generates 100 packets per second in each direction, 200 packets in total.

In EtherNet/IP devices, the real limit is usually not bandwidth but the number of packets the device can handle per second (pps). A communication module or adapter has a certain pps capacity; this capacity covers both the packets of its own connections and the unwanted traffic reaching it. An unnecessarily short RPI wastes this capacity. Choose the RPI according to the real response needs of the control application: a fast cutting position and a tank level do not need the same RPI. A common setting is to pull all connections down to the shortest RPI “so that it’s fast anyway”. Manufacturers such as Rockwell Automation provide planning tools that calculate pps and connection capacity per module; on large projects this calculation is done at the design stage, not during commissioning.

Multicast I/O connections and the flooding problem

A switch forwards unicast frames only to the destination port, according to its MAC address table. The default behaviour for multicast frames is different: if the switch does not know which port is interested in the group, it sends the frame to all ports in the same VLAN. On an unmanaged switch, you cannot change this behaviour.

Consider an example cell: the multicast T→O streams of 30 adapters reach every port in the VLAN. HMIs, drives, cameras and other adapters that have nothing to do with this traffic must also receive, process and discard every packet. On field devices with limited processing capacity, this load causes them to miss packets of their own connections and leads to connection timeouts. The symptom is typical: random dropouts that start as the number of devices on the line grows and whose root cannot be traced to a single device. The fault is not in one device but in the network as a whole.

EtherNet/IP multicast addresses are assigned from the administratively scoped 239.192.x.x range, and the TTL of multicast packets is 1 by default; this traffic does not pass through routers and stays within the VLAN. In other words, the problem stops at the VLAN boundary. Splitting I/O traffic into separate VLANs per machine or cell is the first step in limiting the multicast effect. For the general principles of VLAN planning, see the article VLAN design in enterprise networks.

IGMP snooping and the querier: correct configuration

IGMP snooping is the switch listening to IGMP messages between devices and the multicast router to learn which port is a member of which multicast group. When a scanner opens a multicast connection, it sends an IGMP membership report for the relevant group; the switch sees this report and forwards the group’s traffic only to that port. The details of snooping behaviour are defined in RFC 4541.

Snooping alone is not enough. For membership information to stay up to date, there must be a querier on the network that sends periodic IGMP queries. Without a querier, devices do not send regular reports; the membership entries in the switch time out (about 260 seconds with IGMPv2 defaults), and depending on the switch model, traffic either floods all ports again or is not forwarded at all. The second case looks like this in the field: the network is built, everything works, and a few minutes later the I/O connections drop. It is a hard fault to diagnose, because everything is normal in the first tests.

Rules to follow for querier placement:

  • One querier per VLAN is enough; if there are several candidates, the one with the lowest IP address is elected. Decide deliberately which one will be elected.
  • Snooping switches treat the port where the querier is located as a “multicast router port” and forward all multicast traffic to this port as well. If the querier ends up on a switch at the edge of the network, all multicast traffic is carried over the backbone links to that point. Put the querier on the central switch where I/O traffic is already aggregated.
  • In a ring or STP topology, some switches temporarily flood multicast traffic during a topology change. If ring recovery and this flooding happen at the same time, connection drops can occur despite the ring; check the manufacturer’s flooding behaviour after a topology change.

Example configuration and verification commands on Cisco IOS-based industrial switches:

! Snooping must be enabled globally (default on the Catalyst and IE families)
ip igmp snooping

! Enable this switch as the snooping querier
ip igmp snooping querier
ip igmp snooping querier address 10.20.30.2

! SVI address in the I/O VLAN (querier source)
interface Vlan30
 ip address 10.20.30.2 255.255.255.0

! Verification
show ip igmp snooping querier
show ip igmp snooping groups vlan 30
show ip igmp snooping mrouter vlan 30

The first place to look in the console is the output of show ip igmp snooping groups: each multicast group should map only to the relevant scanner ports. If the group list is empty, either the querier is not working or the devices are not sending IGMP reports. Unmanaged switches and devices’ embedded switches may not be able to do this filtering; on long daisy-chained lines, check in the device documentation how multicast traffic is forwarded through the embedded switches.

Unicast or multicast: choosing the connection type

Most current EtherNet/IP scanners support unicast connections for the T→O direction; the connection type is selected separately for each connection in the engineering tool. Selection criteria:

  • If the data has a single consumer, choose unicast. The risk of flooding disappears and dependence on IGMP configuration decreases.
  • If more than one controller consumes the same data (e.g. listen-only connections or several PLCs monitoring the same input module), multicast prevents the adapter from sending the same data more than once; in this case, IGMP snooping and a querier are mandatory.
  • If the connection will pass through a router or firewall, multicast is unsuitable because of TTL 1; unicast is used.
  • Older devices may only support multicast. Plan a separate VLAN and snooping for these devices.

Example bandwidth calculation

The calculation below is an example based on assumptions, intended to show the order of magnitude; in a real project, calculate the values from device data.

Assumptions:
  Number of adapters    : 30
  RPI                   : 10 ms  → 100 packets/s per direction
  Frame size (on wire)  : ~150 bytes (assumption incl. headers, preamble and inter-frame gap)

Port connected to the scanner:
  Packet rate = 30 × 100 × 2 directions = 6,000 packets/s
  Bandwidth   = 6,000 × 150 × 8         ≈ 7.2 Mbit/s

Multicast T→O, NO snooping (floods every port):
  Unwanted traffic reaching each port = 30 × 100 = 3,000 packets/s ≈ 3.6 Mbit/s

3.6 Mbit/s looks insignificant for a 100 Mbit/s port. The problem is not bandwidth. The problem is packet count. While each adapter processes a few hundred packets per second for its own connection, it also has to receive 3,000 packets per second that have nothing to do with it. As the number of adapters on the line increases or the RPI gets shorter, this load grows linearly. Assess the pps capacity in the device’s datasheet together with this unwanted traffic.

Switch configuration checklist

  • I/O traffic is in separate VLANs per machine/cell; IT traffic and engineering PCs are not in these VLANs.
  • IGMP snooping is enabled in every VLAN where multicast is used, with a single, deliberately chosen querier.
  • The querier is on the central switch where I/O traffic is aggregated; the mrouter ports are the expected uplinks.
  • The output of show ip igmp snooping groups matches the scanner ports; there are no empty group lists.
  • The handling of unknown multicast traffic (flooding or blocking) has been verified from the switch documentation.
  • T→O unicast is selected for single-consumer connections.
  • RPI values are set according to application needs; module pps capacity has been calculated.
  • There are no unmanaged switches in VLANs where multicast is used, or their effects have been taken into account.
  • Multicast flooding behaviour after a topology change has been observed in the ring test.

These checks are not independent of switch selection; IGMP snooping and querier support are among the lines to look for in the datasheet during industrial switch selection. For diagnosis with packet capture and counter analysis, the method in the article PROFINET fault diagnosis steps can largely be adapted to EtherNet/IP networks too. For EtherNet/IP network design and planning connections between PLCs, see our industrial Ethernet and factory networks and PLC-to-PLC and PLC-to-server integration pages. The protocol’s specifications and installation guidelines are published by ODVA.

Frequently asked questions

Why is IGMP snooping needed in an EtherNet/IP network?

If multicast I/O connections are used, without snooping the switch floods this traffic to all ports in the VLAN. Field devices have to process unwanted packets and may experience timeouts on their own connections.

Does snooping work without an IGMP querier?

It appears to work for a short time. Without a querier, devices do not send periodic membership reports, the entries in the switch time out, and depending on the switch model the traffic either floods all ports or is not forwarded at all.

When should a unicast T→O connection be preferred?

Choose unicast if the adapter’s data has a single consumer or if the connection will pass through a router or firewall. Multicast only offers an advantage if more than one controller consumes the same data.

Does shortening the RPI improve performance?

Only if the application really needs faster data. An unnecessarily short RPI consumes the packet capacity devices can handle per second and can reduce the stability of all connections on the network.

Support

Contact us for support or a discovery session on this topic.

Contact form
[email protected]

Eretek Industrial

PLC, PROFINET/PROFIBUS and SCADA communication, IT/OT network segmentation and industrial field connectivity. We install and monitor without stopping the production line.

Knowledge Center

Other articles