Hi Erik: I think Zach and I came up with a common understanding. My concern had not to do with a backbone. It has to do with 2 routers in a same mesh under topology, and without the assumption that any node can reach any other node using the mesh under technology.
To illustrate this, say you use the ND base draft in a Frame Relay network. You build a hub and spoke with the router at the hub. The router is at a central office, and there are DTEs at the branch office. I agree, the draft works great! The router can resolve any incoming packet because it has a proactive registration for it. Then we want to place a second router. We place a switch between those and we split the DLCIs in half between the routers. So each of the routers get half of the registrations. There is simply no DLCI for the hosts to register to the other router. In a same fashion, you do not want a node to register to all routers and receive multicast from all routers in a LoWPAN that grows to the thousands. So with the draft alone, if router 1 gets a packet for a node attached to router B, the registration does not help. So the router will either multicast, which we wanted to avoid, or drop. That's my concern, that's why I've been telling you it was broken for mesh under. Zach and I agreed we could resolve this question by saying that 2 routers is already an extended LoWPAN, that will be addressed by the spec left to be done. Would you agree with that resolution? Continuing on the FR image, the deployment may choose either: - route over. define an OSPFv3 instance and connect those 2 routers as P2MP. Describing this is sort of operation out of scope for 6LoWPAN. - whiteboard. Both routers write onto the subnet white board which nodes they own, and let the mesh under figure out how they can reach one another, can be using PVC or calling out an SVC... This would be what extended does. Extending further the model, it can either: - larger route over. Extend your OSPFv3 instance in P2MP mode, or setup a number of them if you need to meet different needs, and we end up with something similar to what RPL does. - proxy ND. This can be done for specific topologies, like trees, or in the simplest expression, a backbone link, that would be point to point between 2 routers or a fully meshed core, wherever multicast can be emulated at a reasonable cost. Just like ND itself, ND proxy model seems more appropriate for transit networks than for NBMA. I hope it's clearer now, Pascal > -----Original Message----- > From: Erik Nordmark [mailto:[email protected]] > Sent: Monday, May 10, 2010 1:46 PM > To: Pascal Thubert (pthubert) > Cc: Zach Shelby; [email protected]; [email protected]; > [email protected] > Subject: Re: [6lowpan] [Roll] how does a node get an IP address > > On 05/10/10 03:38 AM, Pascal Thubert (pthubert) wrote: > > Zach, > > > > You're not addressing my question. My question is how do you do mesh > > under with 2 routers and the answer cannot be route over. So? > > If the draft does not answer that question and cannot point on a draft > > that does, then, no, it does not stand on its feet, sorry. > > Are you talking about having two 6LBRs? > Or is this a concern about the wired backbone? Please be specific. > > For the case of two 6LBRs it is done the same way as having two routers on > an Ethernet: both routers are configured with the same prefix(es) which will > be advertised in the Router Advertisements. > > The only added thing in 6lowpan-nd is the Context option for header > compression. That implies that the two routers need to be configured with > the same compression information. > > Note that the hosts will register with both of the routers, hence both have > the full information about all the hosts. > > What do you see as missing in 6lowpan-nd-09 for the case of 6LBRs? > > Erik _______________________________________________ 6lowpan mailing list [email protected] https://www.ietf.org/mailman/listinfo/6lowpan
