On Thu, Jul 28, 2022 at 08:02:18AM -0700, Tony Li wrote: > Thank you for re-opening this. > > I will point out that what you’re really hitting on is the known > architectural deficiency of IP: how does a multi-homed network deal > with multiple locators? We’ve discussed this to death previously and > effectively came to the conclusion that we didn’t want to change the > architecture.
The architecture is indeed deficient in dealing with multiple locators for one identifier; I would argue that this draft is the result of accepting that premise, accepting the fact that multiple *identifiers* will be used in parallel, and having the network deal with that. > What’s changed? Are we now open to changing the architecture? Depends on what exactly you're referring to with "architecture". This isn't an attempt to make multiple locators work for a single identifier. If that worked (widely), we wouldn't need dst-src-routing. > Routing based on source address is a band-aid fix for the specific > symptoms. I think that if we are open to making changes, we should > not assume that routing based on source address is the solution, It might be a band-aid fix for general symptoms - what it is intended to be the solution for is correctly routing "multi-prefixed" (a specific kind of multihomed) networks. > and that this draft we be better served by focusing on highlighting > the architectural issue and should avoid talking about solutions. I think that happened in rtgwg-enterprise-pa-multihoming (though it does verge into solutions too), which is now RFC 8678. Cheers, -David P.S.: as far as implementations are concerned, the first time dst-src-routing popped up in the wild was January of 1998, when RTA_SRC got introduced in Linux 2.1.79: (it's "Pedro's subtree work") https://git.kernel.org/pub/scm/linux/kernel/git/history/history.git/tree/net/ipv6/route.c?h=2.1.79#n832 _______________________________________________ rtgwg mailing list [email protected] https://www.ietf.org/mailman/listinfo/rtgwg
