Huh, it's a unique ethertype. Wireshark 2.4.1 knows what it is but 1.8.3 doesn't. Oddly, even though the newer Wireshark knows what it is I can't find where it was assigned. IEEE page shows 8700-8710 were assigned to Motorola, but 8783 falls outside that range.
Regardless, I stand corrected; ignore everything I said about the L2 querier. As rrcpgm said, IGMP snooping should only affect IP multicast since it is based on the multicast IP address. Even trying to statically assign multicast to an interface in IGMP snooping, or forcing flooding inside the multicast-snooping-options for the vlan, requires the multicast IP address. I spend most of my time dealing with IP multicast so I glossed over the "pure L2 multicast" bit.
Unfortunately I don't have any 2200s or 3300s in my test environment to test against but I did check against a couple of standalone QFX5100s on 16.1R5. I duplicated the frame on my Ixia. It isn't exact since IxNetwork is padding the frame quite a bit but the source/destination addresses are as shown in the pcap. Using that I can confirm that the frame is making it from a port on sw1 to a port on sw2, and vice versa. I can also confirm that it is flowing properly within a single VC. Due to other problems I can't tell if it will properly go from one VC across a LAG to another and still work properly.
So yeah, I'm with you in that it would appear to be a bug. I've skimmed through prsearch and see some stuff that might be related, but nothing that specifically matches. The closest one seems to be PR1115300, but that one seems to be describing the exact opposite problem.
Title | IGMP-snooping not preventing multicast flooding | |
Release Note |
|
Problem |
| |
Triggers |
|
| Resolved In | 13.2X51-D39 14.1X53-D35 15.1R2 15.1R3 |
Hmm, that gives me an idea though. Even though snooping shouldn't be touching this traffic, since it appears to be a possible workaround would be to set the interface for the AP as multicast router interface according to snooping. Not perfect since that means it will flood all multicast out the port, but most APs I've dealt with implement their own IGMP snooping to prevent multicast from flooding the airwaves.
I just tried this on an EX8200 with 12.3R11 as well. There I was able to recreate the behavior and confirm that the multicast-router-interface workaround does work.
Sorry for going down the wrong path, but hopefully that helps.
-Chad