There is precious little information available for manufacturers to use when designing EtherNet/IP networks. To be honest, there is a little finger pointing around this topic. Users expect their EtherNet/IP vendors to provide information on how to best implement EtherNet/IP. Vendors take the position that network implementation is beyond their scope: all they do is to provide a device that conforms to the EtherNet/IP specification. Both are making good arguments.
Luckily, distributors are taking up the challenge and developing the expertise to design, select, and configure EtherNet/IP devices. And I think that is appropriate. They have a foot in both worlds. They are more intimate with the user community than the vendors are, and they understand more about the vendors’ products (including switches, routers, and technologies) than the users do.
In today’s article, I am going to look at two of the concepts that are important to network design and provide some recommendations.
One of the costs of using Ethernet on the factory floor is that every Ethernet node must connect to a switch, and switches cost money. Managed switches cost more money than unmanaged switches (I’ve touched on that argument in other articles, and I’ll probably have to do so again). At GM, they’ve stated that it costs $100 to connect an Ethernet device to a managed switch. Your cost is probably less but still significant.
One way to do that is with linear segmenting. Many, if not most, factory floor EtherNet/IP devices have 3-port switches built right into them. One port is the input port, one port is the output port and one port connects to the device. Traffic that is not targeted for that device passes directly from the input port to the output port. You can connect a string of devices together using these ports with the first port connected to a port on the switch. This topology introduces a bit of latency but not enough to worry about if you keep the number of devices under ten or so. The upside of the linear segmentation is that you reduce the cost per device by 90% for 10 devices.
One of the issues with linear segmenting is auto-negotiation of the speed and duplex. How does that work for a string of devices in a linear segment? Your goal should be to configure the entire linear segment for full-duplex at the maximum baud rate. It’s best if your EtherNet/IP devices are delivered set to fixed duplex and baud, but that isn’t usually the case. You will most likely have to configure each device to only use full-duplex and the maximum baud rate (normally in 2020 to 100Meg). When you use linear segments, on powerup, the switch does auto-negotiation with the first device. The first device auto-negotiates with the 2nd device. The 2nd with the 3rd, and so on through the entire linear segment.
If there is an auto-negotiation failure with the first device, you, hopefully, have an SNMP alarm configured to let your system know that you have a port problem on that switch. If any of the downstream devices fail auto-negotiation, you will simply have a failed stream of devices beginning with that device and all the devices that follow it.
Another issue that comes up in many manufacturing system designs is the routing of messages. To learn more, check out part 1 in a series about Ethernet networks.