Le 18 nov. 09 à 23:50, William Herrin a écrit :
On Wed, Nov 18, 2009 at 4:31 AM, Scott Brim <Scott.Brim at gmail.com> wrote:Bill, I don't understand exactly what you're saying but it seems to be something like: - An endpoint may have multiple addresses assigned to it on a single interface. It could even have addresses from different PA prefixes from upstream providers. - If source address and upstream provider path don't match, packets could be dropped. - Address selection itself can't help here, so you suggest that routing take source address into account until packets get beyond where uRPF is enforced.Hi Scott, That too, but I'm trying to convey what I think is a more important notion: Address selection is classically a simple function on the routing table: pick the primary address of the interface the destination routes out of. With multiple administrative domains, address selection may *still* be a simple function on the routing table... the more significant difference may be in the routing table itself. That notion is suggested by the following example: (A,B,C)-D-(E,F) Where A, B, C are hosts, D is a router and E, F are ISPs supplying two different administrative domains with unique blocks of PA addresses. The two domains propogate from E and F down through D to hosts A, B and C. A, B, C and D *all* need to do the right thing with respect to source address selection and route selection based on source address.
Yes, that's the key.
*AND* if D is a complex network instead of a simple router, that complex network needs a *dynamic routing protocol* that helps the routers in D do the right thing, not just static policy routing.
In a complex D, an alternative to a dynamic routing protocol consists in a stateless address mapping, in border hosts and routers of D, with encapsulation of global packets to traverse D.
This is feasible with the SAM approach described in draft-despres- softwire-mesh-sam-01. A customer-side border node, when it has to forward a global packet toward the global Internet, selects the destination of the encapsulating header based on which PA prefix it finds in the source global address. This destination is the address of the provider-side border router to which this PA prefix has been delegated (known by some parameters received in DHCP ).
If you have a look at this draft, maybe you can tell whether you agree or not on the relationship with the point you make.
Regards, RD
I'm not sure if MIF is correctly scoped for that, but offer the thought that maybe it should be. Regards, Bill Herrin -- William D. Herrin ................ herrin at dirtside.com bill at herrin.us 3005 Crane Dr. ...................... Web: <http://bill.herrin.us/> Falls Church, VA 22042-3004 _______________________________________________ mif mailing list mif at ietf.org https://www.ietf.org/mailman/listinfo/mif
Note Well: Messages sent to this mailing list are the opinions of the senders and do not imply endorsement by the IETF.