I'm configuring 7 interconnected 2960s (not a switch stack) with numerous multicast video streams running in a VLAN. Multicast is new to me, but I am coming to understand IGMP snooping and the need for one of the switches to be configured as a querier. (There is no multicast routing, strictly level 2). There is a single central switch connected by single trunk ports to each of the other switches. I find that when I configure a switch as a querier, all the multicast traffic is flooded to that switch over the trunks.
In reading the 2960 configuration guide:
http://www.cisco.com/en/US/docs/switches/lan/catalyst2960/software/release/12.2_58_se/configuration/guide/swigmp.html#wp1118428
I see that this must be because of the behavior that the port connected to the multicast router is added to the multicast forwarding table. Since there is no multicast router, one of the switches is acting as the querier, so presumably that is what gets added instead. But, why does the multicast router/querier get added to the table? The querier has not sent a "join" to the group, so why does it need to receive the multicast data?
The net result is highly congested trunk ports carrying multicast streams to a device that doesn't need them. Isn't this precisely what IGMP snooping is supposed to avoid? What am I missing? More importantly, is there a way to avoid this?
I can envisage workarounds:
- Try to consolidate as many of the multicast sources as possible into a single switch, and make it the querier
- Remove the multicast VLAN from the trunks and interconnect it separately to the switches (wasting ports)
But, wouldn't it just be better if the querier was NOT added to the multicast forwarding table? What would go wrong if it wasn't?
