ietf-nntp New Chairs for NNTPEXT

[Nick Christenson]

> Mario Valente wrote:
> >   Since I only recently joined the mailing list I would like to pose a
> > question,
> >  related to an idea of mine on the redesign of NNTP and Usenet: currently
> >  the RFC states that for a SENDME or ARTICLE request, the *entire* article
> >  should be sent. I have this idea whereby only the headers would be sent
> >  and stored and the body of the articles requested on demand. This would
> >  lead to changes in both NNTP and probably in the format of the messages
> >  themselves.
> > 
> >   Has this been discussed already ? If yes under what subject, since I was
> >  unable to find any info on the mailing list archive ?
> > 
> >   If needed I can expose further on my idea. Its quite simple in RFC terms,
> >  quite simple in implementation terms (at least INNwise) and would save
> >  tons of bandwidth and disk space.

I think this idea has a great deal of merit.  We mention this possibility
in our paper,  "A Scalable News Architecture on a Single Spool", in the
June '97 ;login: and at: http://www.earthlink.net/company/scale_news_art.html.

It would be my recommendation that we finish work on the 977 update
first (finally!), get some consensus on what the current necessary 
extensions are (like auth), and then start thinking about next generation
architectures.  At this point, I think talking about how to go about
retiring store-and-forward is probably a pretty good idea.

> Somone once mentioned to me that they consider one of the strengths of
> NNTP/Usenet is that there is never just one copy of something. It makes
> it alot harder for the Scientologists when they need to sue 10,000
> admins to remove content.

Well, I don't see this as a real problem if you overcome some of the
other things that are necessary to make this work.

For example, if someone posts a very popular porn image from their small
ISP, podunk.net, this machine could get slammed after distributing the
header for article requests.  Certainly a caching model would be necessary
and a hierarchical cache along the channels of news feeds would make most 
sense, it seems to me.  In this case, the scenario you fear is not 
significantly more likely than it is now.  Of course, this suggests a 
lot of other problems, but I'll leave those as an exercise. ;-)

I probably shouldn't have gone this far, as it's not really germaine to 
what needs to be accomplished now, but perhaps this will stimulate
some discussion.  Hopefully, it provides impetus to finish the boring
stuff so we can get to work on the fun stuff.

-- 
Nick Christenson
npc at earthlink.net