[NNTP] LIST EXTENSIONS (again)

Russ Allbery rra at stanford.edu
Mon Nov 8 15:34:06 PST 2004


Mark Crispin <MRC at CAC.Washington.EDU> writes:

> I didn't say that it was easy to get rid of MODE READER; just that it
> was the right thing to do.

What would you think of documenting port 433 for transit connections in
the standard and saying that transit feeds SHOULD use port 433 instead of
port 119?  That to me seems like the most viable way of eliminating MODE
READER.

> Everything that I have heard indicates that this is ultimately a matter
> of implementation elegance in inn rather than an insurmountable barrier.
> It isn't a "doing that will make my server three orders of magnitude
> slower" issue.

Using separate ports is something that would be fairly straightforward,
since that's mostly just a change in installation documentation without a
lot of code change.  Some people are already using port 433, and of course
people wouldn't have to change right away.  (We may want to add some
functionality to innd to allow it to listen on both ports, as a transition
aid.)

Redesigning INN so that it can handle reader and transit connections with
the same daemon on the same port is something that would essentially
require rewriting INN completely.  This isn't just an incidental design
decision -- this was one of the core portions of Rich Salz's original
design, and it would probably be easier to write a completely new server
from scratch than to eliminate INN's split innd/nnrpd structure.

Not that that would necessarily be a horrible idea, given that INN is a
pretty old code base, and I've considered it from time to time myself, but
as Andrew points out, splitting transit and reader functions is still a
very attractive design from a server architecture perspective.  I do think
there are arguments to be made in favor of allowing that design even in
the absence of any consideration of what INN does.

Basically, you're talking about a design change on the order of merging an
MTA and an IMAP server into a single daemon.  (In retrospect, having the
transit and reader components of NNTP be the same protocol was a huge
mistake.)

-- 
Russ Allbery (rra at stanford.edu)             <http://www.eyrie.org/~eagle/>



More information about the ietf-nntp mailing list