From: Brad Templeton (brad@main.templetons.com)
Date: Mon Sep 20 1999 - 17:21:09 CDT
On Mon, Sep 20, 1999 at 05:44:44PM -0400, Henry Spencer wrote:
> On Mon, 20 Sep 1999, Brad Templeton wrote:
> I would phrase this a little differently. We didn't ruin Distribution;
> what we did was to accelerate a trend that was already present. The basic
> problems with Distribution were vague specification and a lack of prompt
> user feedback on errors. This was already in the process of ruining the
> header's utility. What we did was to recognize that, and make it possible
> for admins to do what they already wanted to do, i.e. ignore the header
> except in special cases. This removed whatever slight feedback remained,
> in the form of limited propagation... which was not enough to stop the
> growing misuse of Distribution, but may have been slowing it slightly.
>
> The organization and discipline needed to make Distribution really work
> on a sustained basis simply was never present on Usenet at large.
Well, this is historical argument, but in fact both are true.
The ability to say "all,!foo" was convenient for the admin, but it had
the consequence that any leak anywhere else on the net would flow through
to the whole net because the above means "let through any distribution
you don't know about."
It was too hard for most admins to bother listing their own distributions,
and effectively administratively impossible to control the list on every site
that fed you. In most cases changing your distribution profile would mean
sending E-mail to several other admins and hoping they did what you asked.
The only way admins would ever manage their distribution would be if they
just had one file on their site where they listed what they wanted, and they
could change it at will.
The other thing which ruined the distribution line, though it came second,
was that big ISPs with national dialup decided they wanted to deliberately
get distributions that were not geographically right for them. Some big
sites did the same. Soon it was impossible to stop the articles from
propagating and today this is also true, so the only answer is to move
control in the software to the reader or receiving site. It's less
efficient, of course, but the other method doesn't work at all.
(I suppose one could have some sort of NNTP negotiation to control feeds
but it would be even more complex.)
Anyway, this is all history, and I'm not blaming you for any great error
by any means, it was an unintended consequence, and later when the big
ISPs came, an intended one.
Being able to subset the net by geography, policy etc. is useful, and
the distribution concept is a good way to do it. New hierarchies don't
really subset the net they divide the net, and only subset if you assume
people will then take the hassle of reading multiple hierarchies that differ
only in policy or other form. Sometimes it is right to divide the net,
sometimes it is right to subset. Distribution can almost do it, and
telling software authors to implement it in this fashion is simple and
lets it do the job.