On Thu, Jan 3, 2013 at 5:05 PM, Jared Mauch <[email protected]> wrote: > Oh, I understand all these use-cases, but there is a case for a well-designed > network not always sharing/mixing the NLRI. eg: We don't transport v4 NLRI > in v6 transport, nor v6 NLRI in v4 transport. If you have a single session > with massive shared fate, perhaps it's not a protocol design error but a > network design error.
it is not just a shared-fate problem. Even if you have distinct BGP sessions per AFI, the RainbowPoop scenario still breaks the datacenter network if they actually depend on NewVpnThing. Most networks will still have a very hard time troubleshooting it, and they'll be bleeding customers until they do. I noticed you've omitted MPLS VPNs. We also do not use the same BGP sessions for IPv4 and IPv6, but we do use IPv4 sessions for MPLS VPNs. It is avoidable but I would bet that most networks do not avoid it. Do yours? On Thu, Jan 3, 2013 at 5:05 PM, Tony Li <[email protected]> wrote: > On Jan 3, 2013, at 1:48 PM, "Chris Hall" <[email protected]> wrote: >> However bad or difficult to clean up, what if that's not as bad as the >> alternative ? > > Clearly, the treat-as-withdraw alternative is better. You're not leaving > bogus state in the rest of the network. Does this mean you support treat-as-withdraw, even though you dislike its high complexity? My read of your postings today is that you dislike the whole error-handling idea. -- Jeff S Wheeler <[email protected]> Sr Network Operator / Innovative Network Concepts _______________________________________________ GROW mailing list [email protected] https://www.ietf.org/mailman/listinfo/grow
