bstempleton (10/01/82)
Ok, let's open the can of worms on this by going over some of the things that have been discussed. Ideal: We want mail to people to be based on who they are and where they are, not as much on what computer they use and certainly not on what network that computer is on. Thus a geographical hierarchy based on user@host.domain is something like: user@<minor_designation>.city.province.continent where a <minor_designation> might be a company, or a ward in the city or a street name or an institution or even, god forbid, a particular computer. By province, I mean province or state, shortened to the unique two letter code the post offal has. ie. "Brad Templeton@University of Waterloo.Waterloo.Ontario.NA" for a full designation. (somehow those dots will have to go by the time this comes, I think all @s is the best and can't figure why you need two symbols. Naturally, things can get shorter, and you don't need to use a domain if you are in it. To somebody in Toronto, I might be: Brad@UW.waterloo or to somebody in the states, I might be Brad@UW.waterloo.ON This assumes lots of databases and a central authority of some sort. After all, your machine has to know where to mail to get to the ON domain, which means there has to be a central machine for the ON domain. The "UW" machine has to know every person at UW, but that's not too much to ask. There are problems when net connectivity is involved, since it's hard to get a central machine for Massechusets when some machines are on the Arpanet and others are not. Still that is what it would be nice to strive for. It's not as hard as it seems, and we can even do this in most cases on the uucp net. I think we should look into it. We just need a volunteer machine for each state, which is not too hard since it is often set up that way right now. If we can't get this, we can kludge it by having a database that knows that there is a different way to mail "Columbus.OH" from "Cleveland.OH" because one is btl and the other is a university. I think it can be done. 2) For now At first, I think we just have to set up domains. This is of course up to the people who own the machines in the domains. I know here there should not be much problem getting an "ON" domain as well as a "wat" domain since that's the way it is mostly set up now. Also, while the "wat" domain might sit in the "ON" domain, it will also sit probably in the "uucp" domain for convenience, since this is one of the world's largest cs universities and there's a lot of traffic. The same will be true at most other places. Things like time zones don't make much sense since they're even harder to coordinate than states. After all, if a domain is to mean anything without databases getting extremely complex, a domain has to mean a program on a specific machine that knows how to decode the inner address below the domain. Essentially if a "btl" domain exists, there has to be a "btl" machine (perhaps not its name as far as users are concerned) that can decode foobar.btl for you. Thus domains turn the net into more of a tree. In certain cases, for efficiency, we will have to kludge it up a bit. For example, the "uucp" domain (which will mean nothing to our machines, they will strip it off) will have to represent calling a certain program on the arpanet. Chances are that this will be at Berkeley. Of course, there are several gateway type nodes around, so we'll have to have some smarts that say that certain sites ought to get their mail out of cca, sri-unix or nprdc. If you are the arpanet, they can probably arrange to have the database on every machine that knows this, or they can keep it all at berkeley. The questions remain. What domains are we going to set up? Who is willing to set one up and handle the database for that domain? If set up properly, each machine has to only know mappings for a few domains. These are the global ones such as "arpa" and "csnet", plus any sibling domains to its own domain[s]. In the case above, every machine need simply know a routing for the institutions in its city, the cities in its province and the 60 odd states, provinces and territories. For now, as the net is small, even less must be known. A lot of this information can be machine generated for one site in a domain by any other, so there is not a lot of work in maintaining it. This is what I would like to strive for. A final note: I suspect that we will have to tack a "root" on to the end of domain strings to keep things unique. I propose this root be the null string. This way there are no reserved names in the internet. With this system, I am bstempleton@watmath.uucp. (note the trailing dot) If this is installed, then there could even be a local domain called "uucp" anywhere, since it can be distinguished from the big one by not having the root character after it. Ie. user@blat.uucp might expand to user@blat.uucp.btl.uucp. in the final sense, while user@blat.uucp. is a final string. ---- This is the first set of comments. More to come. Discussion is invited.