mark (11/22/82)
For those of you wondering about the lull in the domain discussions, here is a status report. Brad Templeton made some very good points about problems with the ARPANET notion of the top level domains being .ARPA, .UUCP, and so on. I forwarded these points to Jon Postel, who is involved in the ARPANET decision process. Jon told us something that surprised me, that having the top level domain .ARPA was NOT cast in concrete, and that they were open to suggestions. Brad and I argued our positions back and forth and Jon said he will take the ideas to the powers that be. We're waiting for results now. (Results are going to have to happen rapidly, since all this has to go into effect by Jan 1, 1983 on the ARPANET, and once a domain goes into use it will be hard to change.) Basically, Brad and I were both in favor of having geographic domains at the highest level or two. I wanted to see the network go into the heirarchy below the country (e.g. mark@cbosgd.btl.uucp.usa.na) and Brad wanted the network dispensed with entirely and the state or province used instead (mark@cbosgd.btl.oh.usa.na). (Brad - I hope I'm representing your position correctly - it isn't fresh in my mind.) These are, of course, not the only alternatives, but will be considered along with other ideas. If anyone else has strong opinions or ideas on the subject, send them to me and I'll forward them to Jon. Mark
bstempleton (11/28/82)
Mark, my actual opinions are tempered by realism to some degree. My essential philosophy is that in the long run we will want to leave network topology out of the routing. In my view, computer networks will become as common as the phone network in time, with every business and eventually every citizen having a computer mail address of some kind. (Even if this address simply points to a post office service that prints the message and delivers it for first class mail rates) Once things get this big we won't be able to deal with network topologies - there will be too many for the average user to cope with. In the meantime, of course, we have no central authority to help keep the structure understandable, and in this case money talks. Thus we will have to use domains that have money backing them, ie. already exist in some sense administratively. Fortunately, as you say, in certain cases existing domains can match "ideal" domains, such as the cases of nations or continents. Right now I think we can safely get away with a domain for this continent, and in the case of a few provinces and states, we can get away with domains for them because they are aleady organized in some ways. The essential point is to get the trend established so we don't end up with a spider's web again. The complexity of the uucp net is already too high to easily keep track of and we don't want the global network to get in this shape. It's easy to see how it could with hundreds of operating systems and machines, all with possibly different networking software. A final note: domains that don't match network topology don't mean that we get inefficient mail, it just means we need a bit more work on the name servers so they can optimize paths. If we can generate some automated system (ie. via news) for keeping track of useful paths, we're in business. domains need simply be a user interface, and don't have to have any relation to the actual transfer path or mechanism.
knutsen (03/09/83)
#R:cbosgd:-282800:sri-unix:11900001:000:717 sri-unix!knutsen Nov 30 00:22:00 1982 There is one way that (as far as I know) the domain-address has quite a bit to do with the transport mechanism: for each domain, there is supposed to be a name-to-route table somewhere. The existence of such a table requires some sort of organization within the domain. Thus I believe a strict geographical scheme wouldnt work very well. It also seems to me that the choice of a "top-level" domain is not really neccesary. If every address has ".universe" tacked on the end, its just a waste of bandwidth; just as long as there are a reasonable number of high-level domains, say 10 or so, so everybody can keep track of them. Besides, due to the "multi-homing" problem, there arent really any absolutes anyway...