Hi all,

I recently encountered a strange problem when using a CRS328-24P-4S+RM. I'm not really sure if this is even a RouterOS problem, as it also happens on SwOS, but interestingly just on the CRS328 and not when using a CRS112-8P-4S-IN. But one by one...

We are doing an installation in a theatre with a so called "light sign" system. Stage Manager has a RGB keyboard which controls RGB light signs at various positions at the stage to give commands to the actors. Imagine it as a bunch of keyboard controllable traffic lights. These are custom made devices which use Microchip MCUs with TCP/IP-stack. The light signs establish a 10MBit/s half-duplex connection to the switch and are PoE powered. Communication is done by multicast, all devices send to 239.192.0.10.

Pressing a key on the keyboard sends the following 133 byte package to all devices:
The package everytime includes the color scheme for all connected light signs (max. 18 pcs). If a light sign changes its color after receiving this packet (as the stage manager selected a new color for any device) it sends an 11 byte "ACK" package back to the keyboard:
And now comes the part where I get stuck. It reproducibly happens that the light signs are not reachable from the RGB keyboard anymore. So the 133 byte package doesn't reach a specific light sign anymore, presumably after this light sign has sent the 11 byte ACK package. The link from the CRS328 just stops working in this direction. All other light signs (that didn't send the ACK package at that time as there was no color change for them) are still reachable at this moment. If you change their colors, the problem will also happen to them at some point. Only restarting the light sign by disconnecting the port or power cycling PoE helps.

But now comes another curiosity. The light signs have a button that can give a command back to the stage manager. Pressing it sends the following 8 byte packet to the RGB keyboard:
This packet reaches the RGB keyboard even if the other direction doesn't work anymore. So the link from the light sign back to the keyboard is still alive. But no traffic coming from the keyboard reaches the light sign. This is not a problem of the light sign's ethernet stack. I can even see in the interface overview that no TX traffic (0B) is going out through the switch port although there should be packets sent to the light sign as IGMP snooping is off, so multicast packets should be forwarded to all ports anytime. But this justs stops working randomly when a port received the "ACK" package (not necessarily the first time).

As already mentioned, this seems to happen only on the CRS328, even when using SwitchOS, which is quite concerning. Firmware is up to date. I fear some hardware-issue regarding 10MBit half duplex on the CRS328 and (maybe, but not necessarily) multicast...
What encourages me in this assumption is that putting a 5€ switch between the CRS and the light sign solves the problem.

Any ideas on the this? I'm pretty sure that I checked all software options that could have influence on this (IGMP, storm protection, PoE handling, ...). Having the problems even on SwitchOS where the CRS acts pretty much as a "dumb" unmanged switch leaves me clueless.

It is likely to be a low-level hardware or software issue. Have you sent the above to support@mikrotik.com? I'm afraid they're the only ones with a way to debug this kind of issues.

Just did. Will keep y'all updated...