Host-based versus network-based: Could it be that this topic stems purely from "shim6-multihoming" versus "lack-of-network-based multihoming" ? Imho: Both hosts and network nodes should be able to trigger a change of path. I believe "host-versus-network based" - discussions should also deal with a topological network view where hosts are as well nodes as are routers. I wouldn't be afraid of it - not wrt scalability. Quote from ALayeredMappingSystem.pdf, taken from the LISP-list : "It is unimaginable that an EID is individually allocated to one user. As a result, mapping nodes can store the EID prefix-to-RLOC mapping once an item, which will greatly reduce the scale of mapping storage." Why is this unimaginable? Even the writer of this statement hereby demonstrates that he can imagine this. I guess he worries about the update churn if mappings between individual EIDs and RLOCs are to be disseminated. However, by utilizing an existing routable RLOC-location information, no dissemination is required at all. DNS and others may map the user name to IPv4 + geogr.coordinates (RFC1712) as of the last point of routing. The geographical coordinates are a given "third instrument". If both the host as well as the router complies with it, then routing to the egress-(DFZ?)-node can be done without churnfull mappings and without looking looking at the dest.IP-address. In some way really all solutions are hierarchical. Yes, including the present pure-prefix-building mechanism. I think a more profound analysis of hierarchical routing wouldn't be inappropriate. E.g. the mentioned .pdf is pretty fond of 2 hierarchical levels. Why just 2 ? Why not more than 2 ? As you know, myself, I am in favor of 1 (flat) level, i.e. flat topology, properly sparsed due to some hierarchical process. Heiner
_______________________________________________ rrg mailing list rrg@irtf.org http://www.irtf.org/mailman/listinfo/rrg
