ietf-nntp "Common NNTP Extensions" document updated

[Vincent Archer]

On Mon, Dec 01, 1997 at 09:47:29AM +0000, Clive D.W. Feather wrote:
> Vincent Archer said:
> > infrequently connected to the network. I've discussed firsthand with the
> > admin at the other end on when NEWNEWS was to be used, and the answer was,
> > after a few tries, "never".
> 
> You clearly didn't talk to us.

You're too far from my site, net-wise :)

[lots of NEWNEWS arguments removed, refer to original message]

> Eh ? If you are going to fill in gaps with NEWNEWS, you need to store one
> date per feed. That's all. To do it by articles, you need to store a last

The message I was replying to used NEWNEWS as an example of how to go back
to a currently unrequested group and getting previous articles. A
functionality which is closer to client reading than regular feeding.

> a lot of overlap between feeds, NEWNEWS allows you to optimize the traffic
> because you only fetch article IDs initially, then only fetch *new*
> articles. Using other techniques sends you a load of superfluous data.

XHDR reading does this too, except at a newsgroup granularity rather than
server granularity. If you are fetching a lot of widely-spaced newsgroups,
using X times NEWNEWS or X+x times GROUP+XHDR isn't much of a difference.

> None of this is rocket science - on the contrary, we have tens of thousands
> of customers reading news like this every day.

Are all of your customers using NEWNEWS? How big is your server physical
memory? Because, unless you have a server specifically designed to use
NEWNEWS feeding (i.e. indexing on an article's date in addition to the
message-id), any call to NEWNEWS will end up pulling the full history base
in memory. If your server hasn't enough physical memory, the 2nd requester
will reload the history from disk again.

The only conclusion I can derive from this is "Your mileage may vary".
However, I strongly believe that any RFC which insists on the presence of
the NEWNEWS command on servers will simply end up ignored, because for a
very large number of implementations of the NNTP protocol, NEWNEWS is
considered harmful. We may disagree on the usefulness of NEWNEWS, but I
sure don't want a RFC that people will ignore because it goes against
established usage.
-- 
Vincent ARCHER  -=-=-  Herve Schauer Consultants -=-=-   archer at hsc.fr
Tel: +33 1 41 40 97 00                          Fax: +33 1 41 40 97 09
        01 41 40 97 00                                  01 41 40 97 09