cunningh@noscvax.UUCP (03/03/84)
<> There are complex reasons why many sites do not extensively interconnect across-country. Part of the explanation lies in that there are are de-facto 'domains' within Usenet. Nonetheless, it would be useful if more sites realized that long-distance phone rates are not as distance-dependent as they used to be, and consider calling a selection of sites in very different localities. This would increase the interconnectivity of the net considerably, though, and could lead to some unanticipated changes in traffic (and some unanticipated phone bills). For sites outside of North America, the difference between calling one coast in NA vs. the other coast is almost insignificant compared to the trans-oceanic phone rates. It makes little difference, from their point of view, where in NA they call. -- Bob Cunningham (mail: ...vortex!islenet!bob)
cunningh@noscvax.UUCP (03/12/84)
[] In response to the query "what is the Usenet topology anyways?" The Usenet links are roughly a restricted subset of the sum total of existing uucp links. A link only exists because one site has agreed to connect up (usually by dialup modems, usually using the uucp protocol) to another site on a regular basis, say every few hours. The 'net' effect exists only because there are many sites willing to forward mail and news traffic. Not all existing connections are published. Those that are, for Usenet news transfers, can be found in the newsgroup 'net.news.map'. The extensive lists, broken down geographically are re-published every month by Mark Horton (cbosgd!mark) and friends. In those 'maps' you'll find listings of the various sites, their organizations and connections ... along with some indication of how to contact the administrator of each site. Only the information that the site administrators have chosen to publish in 'net.newsite' tends to get into the maps. It's not only possible, but probably happens more often than most people think, that a site hooks into a Usenet neighbor and doesn't post, or doesn't notify cbosgd!map ... or changes the sites it swaps newsgroups with and doesn't bother to publish the changes. There's over a dozen well-known "backbone" sites that make a consistent, and sometimes un-appreciated effort to provide as reliable as possible news transfer service by interconnecting with each other, and a generous selection of both local neighboring sites and sites geographically much further away. The article which originally sparked this discussion noted that much of the long-distance traffic goes through a relatively small number of sites, and often their's no relatively-direct connections apparent for mail or news. Thus, it's not uncommon to see a news article on Usenet that has bounced back and forth across the continent more than once. Pathological cases have shown up that have also bounced back and forth across the Atlantic ocean more than once. This means that probably more money (untold thousands of dollars?) is being spent on long-distance phone calls than is really necessary, and that the current topology can be called 'sub-optimal'. The suggested solution is to encourage as many sites as possible to set up connections with an orderly series of sites in various parts of the country. It was suggested that this would lead to more cost-wise optimal routine of messages. Obviously, though, an individual site can only minimize its own phone bills by minimizing its long-distance connections. A decision perhaps in that site's best interest, but not necessarily in the interest of the network of all sites. Sigh. -- Bob Cunningham (mail: ...vortex!islenet!bob)
mart@utcsrgv.UUCP (Mart Molle) (03/16/84)
The observation that "if you need to go from A to B, you should do it in one hop" because of the telephone rate structure is obvious to anyone who has ever looked at the initial pages of a telephone book. Sure, to send a message to your friend on the opposite coast it is [globally] cheaper to the Usenet community if you send it yourself in one hop. The problem is that news is BY DEFINITION BROADCAST communications, rather than point to point. Now the issue is how best to distribute each news item to many [5000?] sites. No-one can afford the dialer time to call all those sites directly, so the only sensible thing is to treat news distribution as a hierarchical tree with limited "fan-out" at each node. Thus, if a site is willing to exchange news with, say, 5 sites, the average path length from the originator to everyone else is AT LEAST 6 hops by analogy with the way "chain letters" grow, i.e., 1 + 5 + 5**2 + 5**3 + 5**4 + 5**5 = 3906 [too small] 1 + 5 + 5**2 + 5**3 + 5**4 + 5**5 + 5**6 = 19531 [ok] Now, we can't expect all those thousands of sites to be unique, and the originator probably isn't sitting at the ideal root node of the tree, even if it IS unique, so it stands to reason that the path length of news items that you see should be more in the range of a dozen hops or so. If you REALLY want to make news more efficient, then talk some satellite utility out of a communications channel at certain off-peak hours and have all the "major sites" buy themselves earth stations. That way you can use a broadcast channel to deliver broadcast messages efficiently. Since dishes for pirating TV signals from satellites are cheap enough for Joe Public to buy, the cost of the equipment to bypass Ma Bell shouldn't be that bad.