On Mon, 2 Dec 2002, David Howe wrote: > I think what I am trying to say is - given a "normal" internet user > using IPv4 software that wants to connect to someone "in the cloud", how > does he identify *to his software* the machine in the cloud if that > machine is not given a unique IP address? few if any IPv4 packages can
Of course it should be given an unique IP address. IPv6 is pretty popular with the ad hoc mesh crowd, btw. It's the only address space where you can still get large address slices for free or nearly so. (The space is probably large enough so that one could really map WGS 84 -> IPv6, and have very few direct collisions -- if it wasn't for small well-populated address slices and addresses and networks with magical meaning). But it should also get a geographic address, preferrably one refinable to ~~um scale, if needed. Bits are cheap, right? > address anything more complex than a IPv4 dotted quad (or if given a DNS > name, will resolve same to a dotted quad) > > odds are good that "cloud" nodes will be fully aware of geographic > routing (there are obviously issues there though; given a node that is Hopefully, > geographically "closer" to the required destination, but does not have a > valid path to it, purely geographic routing will fail and fail badly; it Geographic routing stands and falls with some (simple) connectivity assumptions. These are present in wireless dense node clouds in urban areas. > may also be that the optimum route is a longer but less congested (and > therefore higher bandwidth) path than the direct one. The connectivity in a line of sight network is not very high, and it is perfectly feasible to maintain a quality metric (latency, bandwidth) for each link. Given short range and high bandwidth within each cell that's not worth the trouble. > For a mental image, imagine a circular "cloud" with a H shaped hole in > it; think about routing between the "pockets" at top and bottom of the > H, now imagine a narrow (low bandwidth) bridge across the crossbar > (which is a "high cost" path for traffic). How do you handle these two > cases? High-dimensional networks don't block (map a high-dimensional network to Earth surface to see why). But that doesn't help much with current networks, where no satellite clouds are available. It hurts, but for nodes at and nearby the edge one would need to use special case treatment (implementing backpropagating pressure flow, so there would be less incentive to send packets to nodes at a wall).
