All things related to multicast

[Michael H Lambert <>]

If you happened to be on the Internet2 connector call on backbone migration a 
couple of weeks ago, you've probably already heard much of this.  If so, my 
apologies for the duplication.  At any rate, please bear in mind that I'm not 
an MPLS expert so some of the technical details presented might not be 
exactly right.  I'll also throw James Deaton and Bill Owens under the bus as 
co-conspirators.

There is a significant, but not insurmountable, issue with IPv4 multicast 
when moving from the current backbone to the MPLS-based backbone.  Because of 
the way Juniper implements a corner case (meaning it's unlikely to receive 
attention), each MSDP SA received generates a Type-5 L3VPN BGP announcement.  
Under normal circumstances, this generates an acceptable number of 
announcements.  However, with a full MSDP mesh, an MSDP storm can generate 
hundreds of thousands of BGP announcements.  Believe it or not, we are still 
seeing MSDP storms.

One workaround is to filter the BGP announcements.  This does mitigate the 
effects of the storm, but is not RFC compliant.  Other options that have been 
discussed include not having the full mesh or limiting the number of 
Internet2's MSDP peers, effectively requiring connectors to do pairwise MSDP 
peering.

In order to help Internet2 with resource planning, we'd like to get a better 
idea of the current multicast landscape.  If you would, please take a moment 
to respond to the poll at https://doodle.com/poll/zkuvs2vc6dh5dmvr.  Please 
note that we are concerned only with IPv4 multicast at this time.

The poll will close at 2100 UTC on Friday, 30 June.

Thanks,

Michael