All things related to multicast

["Wallace, Steven S" <>]

> 
>   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.  The reason, of course, is that we want to
> buy on price point because we'll buy so many.  Given the 
> liklihood that
> all of the vendors will shave cost as much as possible, it 
> seems likely that
> switches relying on IGMP snooping are primed for byzantine failure at
> the worst possible moments, i.e. when they're most wanted and the cpu
> overloads.
> 

While it is the case that vendors could choose implementations of IGMP
snooping that are cpu intensive,  the switches we are currently purchasing
(HP4000s) identify the IGMP host reports via ASIC, then punt them to the
cpu.  There's very little CPU overhead.  I imagine most current generation
of IGMP snooping switching use a similar approach.

Steven