Re: Changing NNTP servers on the fly.

Eliot Lear <>
Date: Tue, 31 Mar 92 9:42:42 PST
From: Eliot Lear <>
To: (Tim Berners-Lee)
Cc: Edward Vielmetti <>,,
Subject: Re: Changing NNTP servers on the fly.
In-Reply-To: Your message of Tue, 31 Mar 92 09:17:36 GMT+0200
Message-id: <>

Hello all,

It is true that the nntp working group has been pushing against all
sorts of retrieval issues.  How any of the following would be
implemented is completely an open question, right now.  I should say
that much of what follows was the result of informal brainstorming,
and a lot of discussion at various USENIXes.  I think everyone agrees
that the NNTP people do not yet have enough information to make a
decision, and there is a growing concern about scope of whatever
project we would choose to take on, as one could quickly envision a
very broad all-encompassing project that would serve everyone's needs
but never be implemented.  As we begin to discuss best ways to present
news to the user, we immediately come up against five questions.
Briefly described, they are the following:

[1]	How shall the user select and receive new information?
	Are we talking SQL or Z.39 or what?

[2]	Should the mechanism be a pull-update/lockstep mechanism, as
	it is now, or does the server need to have enough smarts about
	things like priorities such that the mechanism should be
	async/interrupt driven?

[3]	Should we be writing the protocol with some sort of RPC
	mechanism in mind, such that the application doesn't even know
	if the service is local?

[4]	How do we handle archives?  Should a saved article be treated
	just as any other article, or do we need stronger archive
	search mechanisms in NNTP?  OR, should archive support be
	placed in the netnews model, itself (e.g., sendme style
	OR, should netnews reading become a distributed model, as
	access to the Internet approaches ubiquity?  Here is where
	we begin to delve into resource and information location

[5]	Should whatever mechanism we design be limited to netnews, or
	should we also leave enough rope for someone to use it for

So what we have right now is a growing list of questions, and not very
many answers - YET.

I must clarify one point Tim made.  News is currently stored and read
locally mostly for historical reasons.  The plain fact of the matter
is that netnews has been and continues to be more popular than the
Internet, simply because it costs less.  Thus in past people have not
considered reading over the Internet as ``the mechanism'' because it
could not be used as such by a large portion of the participants.
There is also an issue of how to find new and interesting articles
under a distributed model.  That's an area I haven't given much
thought at all to.

The statement that the current NNTP is nothing more than a file
transfer protocol is largely correct.  It's a specialized version that
takes advantage of the netnews architecture.  In fact, it would have
been quite possible to implement NNTP *in* FTP as an extension.

Eliot Lear