Hey guys,
Case packer line, 1756-L73 with three 1756-EN2T modules. Each EN2T has a remote rack, two 1734-AENTR and one 1794-AENT. All of it on a 16-port unmanaged switch, with the FactoryTalk View SE client PC on the same switch.
Two weeks ago we added a fourth rack on the first EN2T, the only thing that's changed. Since then the HMI PC freezes for about 10 s at a time, and racks 2 and 3 show 16#0203 connection timed out roughly once an hour. RPI on everything is 20 ms.
I tried a new NIC in the HMI PC, then a different PC entirely, and it freezes just the same. Is the HMI PC too slow for this many tags, or is 20 ms too fast with four racks? Or does a fourth rack put something on that unmanaged switch that every port on it then has to listen to?
Unmanaged switch and four racks of I removed link Before you blame the PC, open Task Manager on it, Performance, Ethernet, and read the receive rate with the SE client closed.
Receive sits around 3 Mbps with nothing running, and the packets counter on the adapter is climbing at a couple of thousand a second. I never looked at it before, but I'm pretty sure it can't have been like that. Rack 4 is set up like the others, RPI 20 ms, Connection tab left at default, which says Multicast. Drives are on their own EN2T so they shouldn't be part of it.
Sorry, why does the HMI PC see the rack traffic at all? I thought a switch only sends a packet to the port it's addressed to, that's the difference from a hub.
It does that for unicast. Your I/O connections are multicast, and an unmanaged switch has no IGMP snooping, so it floods every multicast frame out of every port. Four racks at 20 ms is a couple of thousand frames a second landing on a NIC that has to look at every one and throw it away. The driver falls behind, the PC stalls, and whatever else got delayed in that switch misses its timeout at the same moment, 0203. The fourth rack tipped it over.
Two fixes. A managed switch with IGMP snooping and a querier. Or each adapter's Module Properties, Connection tab, Use Unicast Connection over EtherNet removed link Unicast you can do this week. Someone did a decent article on it: https://plctr.com/utilizing-ethernet-ip-in-plc-communication/
Set the three 1734 racks to Unicast at the shift change, about 40 s of I/O drop per rack. The box on the old 1794-AENT was greyed out so that one stays multicast for now. HMI receive is under 200 kbps idle now. No freeze since 06:00, no 0203 in the module fault logs.
Multicast flood on the unmanaged switch, not the PC and not the RPI. Managed switch goes on the capex list anyway. What I don't get is how it ran a year with three racks and nobody noticed, it must have been sitting just under the edge, I guess. Thanks petek, and billr for the receive rate, that was the number that made it obvious.
Right, so the switch was copying every rack packet to the PC. Got it.