Op 8 nov. 2012, om 03:03 heeft Ted Lemon het volgende geschreven: > On Nov 7, 2012, at 5:33 PM, "Ole Troan (otroan)" <otr...@cisco.com> wrote: >> Disagree. Hierarchical or flat PD (with relays) don't work for multihomed >> sites, have problems with arbitrary topplogies etc. > > You said this before, but you didn't describe any arbitrary topology in which > PD wouldn't work. Could you do that, please? > > To be clear, although a PD topology is always hierarchical with respect to > the relationship between any given requesting router and its delegating > router(s), this does not mean that the topology of the homenet is a > hierarchy. > > Let me give you an example. Suppose you have a CPE router, A, at the > customer edge. And you have two homenet routers, B and C, each with an > interface connected to A. And you have a fourth homenet router, D, with an > interface connected to B and one connected to C (I'm assuming direct > twisted-pair ethernet links for simplicity). > > I think you are supposing that because D has two paths to A, it will appear > twice in the PD topology, even though it exists only once. But that is not > so. It is only attempting to configure one interface—its downstream > interface. So the DUID and IAID will be the same in both copies of the DHCP > solicit that A gets. And so A will assign a single prefix, not two.
I would say that in case D doesn't have an additional interface, it doesn't request a prefix. It just autoconfigures one global address, although it may configure addresses on both links (to B and C). And I suggest that if D requests a prefix, for an additional interface, it uses unicast to A, with an autoconfigured address. So A receives only one request. Teco > > Suppose D is also a edge router. We are now multihomed; A will be a > requesting router with respect to D for the prefix that D got from its ISP; D > will still be a requesting router with respect to A for the prefix that A got > from its ISP. The relationships between B and C and A, and between B and C > and D, are likewise straightforward. > > I can diagram this out for you if you want, but hopefully that illustrates > the point I'm making. Are you thinking of an even more contorted topology > than this one? > _______________________________________________ homenet mailing list homenet@ietf.org https://www.ietf.org/mailman/listinfo/homenet
