Path: utzoo!attcan!uunet!clyde.concordia.ca!news-server.csri.toronto.edu!cs.utexas.edu!usc!snorkelwacker!ira.uka.de!smurf!urlichs From: urlichs@smurf.sub.org (Matthias Urlichs) Newsgroups: comp.mail.uucp Subject: Re: Paths and Precedence (Re: Question about From: lines) Message-ID: Date: 18 Jul 90 11:07:44 GMT References: Organization: University of Karlsruhe, FRG Lines: 122 Warning: Some or all of the following may be bleedin' obvious to some people, blatantly wrong to others, and just disconnected ramblings to still others. You have been warned. In comp.mail.uucp, article , peter@ficc.ferranti.com (Peter da Silva) writes: < In article urlichs@smurf.sub.org (Matthias Urlichs) writes: < > Maybe I phrased this wrong. I _know_ that the people hanging off me are such < > hosts. < < Like I said, maybe that's how things are in Germany, but here in the states < Usenet is not a tree. Look at the postscript maps some time: you have this < nice tree structure in Europe, and an incredible mesh of redundant links in < the U.S.. Even clumps of sites hanging off the internet are going to look the < same way, as individuals with PCs hook on and drop off. < Umm, like I said before, the German UUCP network is divided into two parts: One consists of EuNET members and subscribes to that tree structure (more or less -- see below), the other part does as it damn well pleases, much like you do in the US. I am part of the latter. Anyway, none of my configuration depends on this. I am just using knowledge about my neighbors, based both on what they tell me about themselves and what they do about my fancy RFC822 addresses I send them when testing the link. What higher-level structure (if any) my neighbors and I are meshed into does not enter the picture. <>The only problem was writing a Perl script which distills domainized host <>names from the maps, and then aliases all UUCP names to domain names before <>trying to get any intelligent routing done. After all, if anyone wants to send <>his/her mail through a ten-hop path instead of directly beaming it across the <>Internet, then that's their decision, but definitely not mine. < < What if that message contains material that violates the Internet's charter. < Say, digitized playboy pics? Besides, I don't consider the growing dependence < of Usenet on the Internet a healthy trend. If it wasn't for the independent < nature of Usenet, your little subnet wouldn't have stood a chance. I'd like to see "The Internet's Charter", please. Sending the EMail equivalent of alt.sex.pictures across the Internet may violate some charter or other, but so does mailing said picture across multiple UUCP hops violate the purpose these links have been set up for. Besides, what do you want me to do? Make my mailer analyze your (and anyone else's) traffic, using some sort of AI, when deciding how to route it? I don't think I have to mention that this won't work. If two sites want to exchange traffic which violates anything, either in content or in volume, they'll have to set up a private link. No problem. Or to be more exact, their problem, not mine. Reserving some links for some special traffic is easy if you use MMDF. Don't say "your little subnet". There are 570 sites in our map, which isn't exactly little. ;-) < And there is still the problem of getting to sites that you know about but < aren't in the maps? Not everyone publishes all their connections: look at < the map entry for Intercon. And that's a site that fully supports domains! If I know (a path to) someone who isn't in the maps, I have to add a route to that site into my database. No problem; the algorithm only breaks down when I locally attribute a link with higher cost than with which it is published. Besides, if Intercon has a private link which isn't published, then either the linked-to site is in the maps or it isn't. If it is not, then I'll have to use source routing, and my MTA knows about that. If it is, then presumably Intercon doesn't want that link used for general traffic anyway and Pathalias will find a better route. < < Finally, routing everything through the internet is expensive. That 10-hop < path might be cheaper than sending the mail up the hierarchy and a few < thousand miles to the friendly internet site you leach off, only to have < it beamed back to a location 30 miles away. Better to send it through your < buddy's system because a friend of his is a student at the local college. < Hod do you find that out? The maps. I didn't say that I want to route everything through the Internet. There also isn't a hierarchy. The Usenet map (which is not relevant to this discussion since it's a News map, not a UUCP map) still looks like a tree over here, and so do the UUCP maps for that matter, but today the "backbone" links are Internet lines. Why generate a path along these links instead of sending the mail directly to the destination? I _am_ using pathalias to find out how the mail would be routed. Who said I don't? Assume the pathalias-generated path sitea!siteb!sitec!sited!user, which (according to the maps) is equivalent to sitea.dom.one!siteb!sitec.dom.x! !sited!user. These equates are in the maps, but so far no one seems to actively use them. Now I can do what everyone else seems to be doing which is to take the first bang path and dump it on sitea, which I'm trying to avoid. Alternately, if I know sitea to be intelligent enough, I say "user%sited@sitec.dom.x" (or whatever style of source routing both I and sitea like). Or, if sited is registered in a part of the map which both I and sitea know about, I can say user@sited. If siteb isn't intelligent enough to understand that sort of address, then sitea has to generate a path through it. I'll do the same with any of my neighbors who requires it (and if I publish links to those, they'll be marked DAILY*2 or higher in order to avoid them, if possible), but if siteb needs bang paths it's not my problem, and not my responsibility either, but sitea's. Or, I can take the short cut across the network, and say "user%sited@sitec.dom.x" (or "@sitec.dom.x:user@sited". Or you might use "sited!user@sitec.dom.x" if you want to. I'm not that optimistic) to sitec's SMTP ports. Or to the SMTP port sitec.dom.x is MXed to, which amounts to the same thing -- I'm willing to assume that links to MXes are reasonably fast. I'll admit that how to determine which of these options to use may not always be obvious, but they all should work. The reverse problem is far more common, i.e. when someone generates a ten-hop-path (hosta!....hostj!user), and no one realizes that hostb could send the stuff directly to hosti through the Internet instead of letting the addresses get mangled beyond recovery. I've seen lots of that happen with my bang paths. That's why I decided to cut through all this, and hand off the addresses to my neighbors according to their intelligence. See above. -- Matthias Urlichs -- urlichs@smurf.sub.org -- urlichs@smurf.ira.uka.de Humboldtstrasse 7 - 7500 Karlsruhe 1 - FRG -- +49+721+621127(Voice)/621227(PEP)