chris@GYRE.UMD.EDU (Chris Torek) (10/12/88)
[I think this discussion needs some irreverent irrelevance, so...:] Curiously, `dynamic address assignment' is trivial in an XNS environment. Xerox Network addresses consist of unique 48 bit host numbers and 16 bit network numbers. Imagine that the machine has the 48 bit number wired into its backplane. If the machine is connected to an Ethernet, it uses this as an Ethernet address; its network number is then a logical `cable number'. If it is on dialup connections, it borrows the network number(s) from the router(s) it can reach by these dialups. Its host number is fixed (wired into the backplane, remember?). The only part of its address that is dynamic is the network number. Some other things fall out of this naturally: A machine can be a packet forwarder if and only if it has multiple addresses; it can forward between any two of its networks. For hosts, routing is easy. If the network number is known and is the same as (one of) your own, send it on that network. If it is unknown, or if it is not the same as yours, send it to a `gateway' (router). (For routers, routing is hard. No surprises here.) What this really does is point up a particular problem in IP: host addresses are too closely tied to host locations. My class B network umdnet machine `imladris.cs.umd.edu' can be put anywhere on umdnet, but nowhere on ucb-ether (at least not without Mills' `tunnels'). The XNS scheme has its advantages.... [Hm, not as irrelevant, nor as irreverent, as I had hoped. Ah well.] Chris