Hi Carsten, Ralph, I think indeed it is important to consider mesh under and route over separately, as this has strong impact on the L2 behavior: in mesh under, - links are transitive, multicast can be costly In route over links are not transitive and multicast is not that costly. More comments inline.
> -----Original Message----- > From: [email protected] > [mailto:[email protected]] On Behalf Of Carsten Bormann > Sent: lundi 16 novembre 2009 07:06 > To: Ralph Droms (rdroms) > Cc: 6lowpan > Subject: Re: [6lowpan] Thoughts on draft-ietf-6lowpan-nd-07 > > Quick answers: > > On Nov 16, 2009, at 06:43, Ralph Droms wrote: > > > 1. In a mesh-under network, the L2 characteristics of the > lowpan are > > close to those usually assumed for implementation of IPv6; in > > particular, there are no "lowpan routers" at L3 and L3 > messages *can* > > be delivered (perhaps with lower probability of success) directly > > between lowpan nodes. Why is ND not sufficient in this model? > > RFC 4861 ND works in a pure RFC 4944 mesh-under network if: > 1) the mesh-under (L2) routing protocol provides subnet-wide > multicast, > 2) that is efficient enough to be used for routine ND messages, > 3) nodes are awake often enough to detect and reply to NS messages. > > Such networks do exist, but these assumptions are not > necessarily compatible with many application scenarios we > have in mind; this is the reason we started with ND optimizations. > > > Would > > proxy ND be sufficient? > > I haven't seen such a design yet, so I don't know. > > > 2. In a route-over network, all nodes are routers > > That is not the assumption: The assumption is that there are > hosts and routers, and that the forwarding function is > performed by the routers. There needs to be a protocol that > enables hosts to register themselves to routers, independent > of the routing protocol. I may be missing a shortcut. Where does the need for registration come from? In e.g. an ethernet IPv6 network hosts do not register (RS/RA). > Since the host-router relationship is somewhat ephemeral due > to the nature of the wireless links, address assignment needs > to be lowpan-wide instead of per host-router relationship. If you have one prefix per router you do not need this. What are your major concerns about this model? Julien > As ND-07 is not by itself supporting host-host communication, > address resolution only happens between hosts and routers. > > Re one other term: "relaying" is the term ND-07 is using for > what routers do between themselves to process host NR > messages; it is not IP forwarding. > It is probably more appropriate to give the inter-router > message a different name (e.g., relayed NR, RNR), so the > confusion between this relaying process and IP forwarding is > reduced; this is the part that will move to the second > document after the split. > > Gruesse, Carsten > > _______________________________________________ > 6lowpan mailing list > [email protected] > https://www.ietf.org/mailman/listinfo/6lowpan > _______________________________________________ 6lowpan mailing list [email protected] https://www.ietf.org/mailman/listinfo/6lowpan
