All things related to multicast

[Bill Fenner <>]

>  IGMP snooping seems to me to be almost insidious.  It requires non-
>trivial cpu activity from the very device we'd like not to have do
>significant computation.

I think that depends on the implementation; I understand there are
switches that have hardware helpers for IGMP snooping so there's very
little overhead.  In addition, I've seen at least one switch that does
per-port layer-3 filtering for IP multicast based upon its snooped
IGMP info - i.e. none of the IP->ethernet multicast address mapping
problems arise.

>  GMRP appears to me as though it may be workable.  Either that or I
>haven't gotten the word yet.  Among the hooks are that it needs an
>802.1p aware NIC.

This is actually a requirement for the driver, not the card itself.
It's theoretically possible for an operating system to support GMRP at its
generic ethernet layer; since Windows does not do so 3Com implemented GMRP
in their own driver.  My impression is that 3Com is the most interested
in GMRP since their NICs historically do no multicast filtering so if
they can offload the filtering to the switch it's a big win.

>Does igmp go away?

No, IGMP happens at layer 3 (for the multicast routers) and GMRP happens
at layer 2.

  Bill