All things related to multicast

["Gordon Rogier" <>]

> If you do have non-Cisco switches with IGMP Snooping within
> the same LAN
> with CGMP switches, you may have problem, because for CGMP to
> work correctly,
> the CGMP-router must receive all IGMP-reports sent by hosts downstream
> to a CGMP switch. An IGMP-Snooping switch may actually suppress IGMP
> reports for reasons of efficiency (only send those reports up
> to the router
> that are necessary). Our IGMP Snooping switches actually do
> this optimization,
> but stop to do it if they see CGMP in the network.

In my previous life at KU and with GPN (I just recently took a job with
Cisco as an SE, but still provide some transitional support for GPN), I also
saw this optimization problem when using a multi-vendor implementation
approach.

Also, you may want to consider how you will deal with layer-2 traffic when
SSM roles out.... I wonder how many of the IGMP snoopers types will be
upgradable to IGMPv3 (???).  I suspect that CGMP will be able to adapt to
SSM much easier and quickly than IGMP snooping... why??.. because most of
the successful IGMP snoopers have to prefilter the IGMP joins/leaves/etc out
with ASIC assists... are these ASICs going to work with IGMPv3?  I really do
not know.

Please do not hit me up with a "you're with Cisco now" comeback... I really
do not have a clue on how Cisco even looks at this issue, but I do know
(esp. based on Eckhart's insight above) that there are some out there whom
look at multicast as not just a layer-3 challenge, but also layer-2.

I only added my 1/2 pennies worth here because about a year ago at KU I ran
into this same set of issues and had to work with some of the vendors (I
will not name them here... only to say it was not Cisco) to correct some
other multicast issues, then found in their improvements this "optimization"
issue existed.  I do not know if this vendor later realized this issue
existed... understand that this "optimization" issue effects more than
Cisco's CGMP method... it also effected normal IGMP interactions of group
members.  My analysis was that doing these "optimzations" blindly was a bad
thing (and I did not even know until now that Cisco's version of IGMP
snooping took this into account).

--*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*--
Gordon Rogier                        

Great Plains Network