Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP Posting-Version: version B 2.10.2 GARFIELD 20/11/84; site garfield.UUCP Path: utzoo!utcsri!utai!garfield!dave From: dave@garfield.UUCP Newsgroups: net.news Subject: How to Defeat the Backbone Cabal in your Spare Time Message-ID: <2633@garfield.UUCP> Date: Mon, 6-Oct-86 09:30:10 EDT Article-I.D.: garfield.2633 Posted: Mon Oct 6 09:30:10 1986 Date-Received: Fri, 10-Oct-86 03:02:39 EDT Sender: news@garfield.UUCP Reply-To: dave@garfield.UUCP (David Janes) Distribution: world Organization: Semi-Optimal Software Solutions (Inc.) Lines: 75 Outlined below is a scheme where persons and sites who are upset with the current Usenet setup (i.e. Backbone and leaf nodes) may rearrange things to their liking. Usenet can be considered a directed graph, where the sites composing the network are the nodes, and the directed connections between them are the edges. Furthermore, a cost function may be assigned to any edge in this graph; for our purposes, let this cost be the average time in hours it takes an article to be delivered on this edge. We extend this cost function to any path in the graph to be the sum of all the costs of this individual edges in that path. Divide Usenet into a number of partitions, such that: (i) every site in Usenet is in at least one partition. (ii) each partition is connected (i.e. articles posted at any site in a partition will reach every other site in that partition.) (iii) let A and B be any two sites in a partition. The _minimum_ cost of delivering a message between these two sites must not exceed N hours (where N is arbitrary, say 3 days = 72 hours). (A better partitioning function may be found. Also note that the total number of partitions should be fairly small, say, not to exceed 15. Also, for covenience we will only consider the North American part of Usenet, since other considerations will come into play in other cases.) Here is the fun part. Any article received by a (participating) site is passed on as normal. Any article *posted* by a participating host is also passed on as normal. However, the posting site is also responsible for DELIVERING THE ARTICLE TO AT LEAST ONE SITE IN EVERY OTHER PARTITION. The beauty of the system is: (i) there is no `backbone' - every participating site is responsible for doing what the current backbone is presently doing. (ii) posting sites are responsible for a more realistic share of the cost of posting an article (most other partitions will require a long distance phone). (iii) No software changes are required. (use the L0 option in your `sys' file). (iv) Administrave power is decentralized. See (i). (v) Only sites who wish to participate have to. (Assuming that only a small number of sites refuse to deliver all the newsgroups, otherwise participating sites will have to make an effort to reconnect the graph a bit by calling `past' offending sites). (vi) quality of articles may go up (see (ii)). (vii) less articles may be posted (see (ii)). Of course, there are several drawbacks to this scheme: (i) no one forces any site to call _all_ the other partitions, so some partitions may not receive all messages. (ii) some sites cannot make long distance phone calls (some nearby sites will have to take the responsibility - and the cost - for them). (iii) Someone has to figure out what the partitions are. This is a one shot problem though. (iv) One site in a partition may become responsible for all or most of incoming articles. To avoid this, every participating site should be willing to receive articles from any other participating site. The site to send articles to in each partition may be picked randomly at run time (this requires a bit of cleverness), or is (hopefully) randomly picked when a site `joins' the network. Comments? dave -- .oOo. The UUCP: {utcsri,ihnp4,allegra,philabs}!garfield!dave o.o.o Mercenary CDNNET: David Janes .oOo. Programmer "There can only be one!"