All things related to multicast

[Bill Owens <>]

It turns out to have been a Cisco problem. They have a bug in their 
handling of multiprotocol BGP when peers are configured in 
peer-groups. The BGP header includes the correct next-hop, but the 
next hop field in the MP_REACH_NLRI incorrectly contains the address 
of another interface on the router. Cisco appears to assume that all 
eBGP peers in the peer-group will themselves be fully meshed.

If you have access to the Cisco bug database, you can look at:

CSCds54055 - BGP next hop changes with peer-groups configured with 
next-hop-self
CSCds06065 - BGP mBGP should use nexthop value present in MP_REACH

They may in fact be two reports of the same bug, the descriptions 
aren't terribly clear. In any case, my peering with the Abilene CLEV 
router was moved out of the peer-group yesterday, and the errors and 
loss of multicast routes immediately stopped. Once that is done for 
my other router I'm sure it will be fixed as well. I'd guess that 
this bug could also affect other vendors' routers when they peer with 
a Cisco, but I'm not sure if anyone besides Juniper supports MBGP. If 
you have a Juniper and are wondering if you're affected, I can send 
your more details of how to detect the problem. . .

Bill.