John H. Robinson, IV wrote: > Todd Walton wrote: >> Also, PO Boxes often have a unique ZIP+4, so that a ZIP+4 will get the >> mail to exactly the right spot. > > I did not think about PO Boxes. You are right. Take La Mesa, CA. There > are four zip codes: 91941, 91942, 91943, and 91944. The last two are for > PO Boxes. That zip+4 would indiate a specific PO box, and you could map > to three-dimensional space coordinates. Very good point. > >> I once lived in a town where mail to me only needed "Todd Walton" and >> "Spickard, MO". The postmaster knew where the Waltons lived. I think >> we should build a network on that. > > There are very rural areas where addresses are actually directions on > how to get there. Not in this country, though. It was one of them more > third-worldy ones (you know - I heard the report on NPR one > afternoon...)
This thread has touched exactly on some the things I got out of VJ's talk. Traditional telco addresses, and zipcodes, and old-fashioned addresses-by-giving-directions, are all examples of addesses being, in reality, paths. They don't really know anything specifically about the endpoints. They break when a path segment fails, they don't scale well. Association of locality with endpoint is not even part of the spec, as I see it .. although it may be part of the implementation (eg, postal delivery routes). Of course misaddressed or misprocessed mail _can_ get delivered anyway, because of human intervention, but these path-based systems have reliability problems because total reliability is the product of the component reliabilities. (Ignoring that there might be alternative paths and recovery-strategies nowadays.) For IP, there is no path. Each network node tries to get to the desired endpoint, or to somewhere else thought likely to know how to get there (if any), or it may just throw the packet away. I'm a bit fuzzy about the details, but reliability now benefits from having parallel paths, and the endpoints (eg TCP) fixup for any packets that go astray. I wish I did have a transcript, because I'd like to recall VJ's words describing this. Regards, ..jim -- [email protected] http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list
