kre@ucbvax.ARPA (Robert Elz) (07/13/85)
In article <1261@peora.UUCP>, jer@peora.UUCP (J. Eric Roskos) writes: (in an article with a Distribution of "usa" which I have deleted for this followup - this is obviously appropriate for the whole world. I have shortened the original article somewhat) > As is evident if you read net.sources.mac or net.micro.mac, lately there are > a lot of very good programs being posted in pieces, pieces of which don't > make it to some sites. > > Suppose you are at a site that doesn't get an important article. Now, > that means many sites who you feed won't get it either. However, you can > usually get somebody else to mail it to you. > > Now, given the previously lost article, which contains the original message > ID, if your system administrator is one of the above knowledgeable net.news > gurus, he can probably get it back into the network downline from where it > was lost (I think), with the same message ID it had before. > > It would seem (if I understand the Usenet software) that it would get fed > along the path which has it missing, until it hit a site that already had > it, after which it would not be passed further; thus filling in part of the > gap that previously existed. This isn't always going to work (it won't help the "corrupted article" problem), but it should certainly help reduce net traffic on repostings of articles that simply get "lost" somewhere. No new news programs are needed, "rnews" does just the right thing. Take an article (complete with headers, exactly as it appears in some sites news directory) and "rnews <article" will post it locally, and redistribute it to the sites that you feed (and via them to the sites that they feed) until a site that already has the article is reached. There some things to watch for though - the "Path" line will be useless as a uucp mail path - but that's not what it should ever be used for anyway (it's never guaranteed to be correct). The article won't appear to be a local posting, so it won't get sent out over links that you have marked with the 'L' flag. You should obviously consult with news admins at neighbouring sites, nothing at all will be achieved if everyone does this. See if you can find a branch of the news tree that has missed the article, and try to arrange that it is "reposted" only once in that area. Robert Elz ucbvax!kre (seismo!munnari!kre) Why are the good ideas always someone elses?
jerry@oliveb.UUCP (Jerry Aguirre) (07/16/85)
It has been suggested that, when a site doesn't receive an article and
later requests it by mail, it then reinsert the article into the news
system so other sites will receive it.
Wouldn't it be a whole lot simpler to just add a few more news
connections to the net. The problem seems to be the number of sites
which are only connected by a single path. If any site on that path
garbles the article then it never makes it. If there were an alternate
path then the article would slip in the back door and propagate to
sites "downstream" from the site that garbled the article,
automatically and with a correct return path.
With a few exceptions the net has a star topology with all the delays
and unreliability associated with such a topology. Star network
proponents please take note that the news software does not take
advantage of any of the advantages available with that topology.
Instead we have the worst of both worlds, lost articles, two week
transmission delays, and twenty site reply paths. This was discussed
about two years ago in a series of articles titled "sub optimal
topology" but nothing came of it.
This lack of connectivity is affecting not only the reliability and
speed of the network but also its politics. Lets face it, the network
is really controlled by only a few sites. I do not imply that these
sites are abusing their position but they still control the net in the
sense that they could destroy it. In most cases removing just one site
from the net would isolate that section from the rest of the country.
A star topology implies a central administration, something the net
claims not to have!
Would all the discussion about some sites dropping net.flame have
occurred if sites could reasonably expect to receive that group by an
alternate path? Most sites are hanging out there as a "leaf" totally
dependent on the local "backbone" site.
There have been few efforts to fight this trend. The count
modification to the local flag made by bytebug@pertec.UUCP (roger long)
is a great step towards allowing more efficient cross feeding of
articles between sites. How many sites use it?
At least half the problems of the net seem solvable by greater
interconnection. Trying to patch them here and there is doing it the
hard way.
Jerry Aguirre @ Olivetti ATC
{hplabs|fortune|idi|ihnp4|tolerant|allegra|tymix}!oliveb!jerryhokey@plus5.UUCP (Hokey) (07/16/85)
The real answer is to come up with a news transfer mechanism that guarantees correct delivery. I still haven't had the time to look into xfernews. This issue is hampered by the manual/batch orientation of UUCP. If something messes up, uucp expects somebody to fix it and try again. If a news batch messes up, it is usually lost. -- Hokey ..ihnp4!plus5!hokey 314-725-9492