[54086] in North American Network Operators' Group
Re: /8s and filtering
daemon@ATHENA.MIT.EDU (bdragon@gweep.net)
Tue Dec 10 13:27:23 2002
To: tme@multicasttech.com (Marshall Eubanks)
Date: Tue, 10 Dec 2002 13:26:52 -0500 (EST)
Cc: nanog@merit.edu
In-Reply-To: <29605DBA-0C4F-11D7-9B57-003065BA697E@multicasttech.com> from "Marshall Eubanks" at Dec 10, 2002 09:53:36 AM
From: <bdragon@gweep.net>
Errors-To: owner-nanog-outgoing@merit.edu
> Here is one reason for some to care :
>
> If you want to do interdomain multicast, and your
> address space is not announced globally (say because you have a /24 that
> is
> not from the swamp), you are likely to be black-holed
> due to RPF failures (as your unicast and multicast routing is likely to
> be different).
>
> This has caused problems from time to time...
>
> Regards
> Marshall Eubanks
Until such time as vendor C determines that RPF lookups should
use the MRIB first, then the URIB, rather than the final RIB,
the following will help:
router bgp
distance mbgp 19 19 19
ip mroute 0.0.0.0 0.0.0.0 bgp Null0 20
or if using 12.1:
router bgp
address-family ipv4 multicast
distance bgp 19 19 19
ip mroute 0.0.0.0 0.0.0.0 bgp Null0 20
not having looked at vendor J's multicast, I can't speak to it.
This does, however, make absolutely certain that only MBGP routes
determine RPF choices. If you expect or desire unicast routes
to drive multicast routing decisions, the above is not for you.
I had to do this to get around a bunch of RPF hedge cases.
