[net.news.group] Overload

putnam@steinmetz.UUCP (jefu) (09/15/85)

A lot has been said in this group recently about net apocalypse,
death and destruction, and so forth.

It seems to me that there are three issues involved :
     1) Transmission overhead
     2) Storage overhead
     3) Human information overload
     
The first may be helped by faster modems, compression, and StarGate.  
Even with moderation, StarGate and the like would help to reduce overhead
in transmission.  

The second may be helped in several ways :
   Expiring groups on a per-group basis.
   Sites with many machines linked together by a fast LAN could build
     network news servers so that articles need only reside on a single 
     machine.  (Hasn't anyone done this?)
   Just not subscribing to 'irrelevant' groups.  This one is a problem,
     as the manager in charge of the machine may well decide that 
     (for instance) net.motss is not at all worthwhile, but that net.cycle
     is.  (parenthetically, i wonder if this could be construed as a 
     First Amendment problem).  
   Keeping articles in compressed format - spends cycles instead of sectors -
     not perhaps very viable, but at least worth considering.
   
The third issue concerns me much more than the others as i think that there
are technological solutions to the first two.  

I think that there may well be a solution to the third issue as well, but
it _might_ involve some major software hackery.   

First, the news software _must_ be fixed to screen out multiple readings
of the same message.  Either that, or multiple postings and duplicate
postings (via software error) should be forbidden.  The problem is that
multiple postings are often reasonable.  An example is a brief mention
i just posted on a book "Winning Ways For...", on a mathematical analysis
of game playing.  It was a response to an article in net.math on nim, but
i felt that if readers of net.books and net.games had not seen references
to the book that it was worth bringing it to their attention.  I really
dont think, though, that anyone should have to read it more than once.

Multiple postings to some groups are obnoxious, especially things 
multiply posted to net.flame, net.politics....  It might be worthwhile
to build a description into the newsgroup description and the 
'allow this posting?' code to allow or forbid multiple postings.
Something like "net.flame:no_multiple_postings", 
"net.women:multiple_posting_to net.abortion" and so forth.

But... 

I would like to propose a more fundamental change.  Now i cannot
propose a precise mechanism, especially in this distributed network,
but allow me to suggest a few examples...

A discussion starts off in (say) net.news.group about the coming 
net.apocalypse.  So, a new group is (automagically) created called
net.news.group.apocalypse, when, after a while, it becomes unused and
expire finally empties it, and perhaps waiting for a decent interval
of mourning, it is removed from the newsgroup list, its directory is 
deleted, and its gone.  This would make the news more like a directory
tree.  

If the reader could then at some given news level, browse the sublevels
and look at the interesting stuff there (ok, this does require a major
new news reading program, picky, picky, picky), the reader could then
choose to read only the subgroups of interest, and in each of these 
subgroups a single discussion would reside.  

I would propose that only one message on a topic would not cause the
sub-group to be generated, but that there be some lower bound (probably
small, maybe different for each group....)  For example in {net,mod}.sources,
posting all the pieces of and bug reports for a single program in a 
subgroup would facilitate finding the source when needed.

Hard to implement?  Yah, but not impossible.  sub-group names do not
have to be the same as the directory names where they are stored, but could
be generated from the Subject: line.

Dumb idea?  Breaks everything we have already?  Oh well, sorry i mentioned
it.  Back to the apocalypse which is already in progress....



-- 
               O                      -- jefu
       tell me all about              -- UUCP: edison!steinmetz!putnam
Anna Livia! I want to hear all....    -- ARPA: putnam@kbsvax.decnet@GE-CRD

riferguson@watmath.UUCP (Rob Ferguson [MFCF]) (09/16/85)

>A discussion starts off in (say) net.news.group about the coming 
>net.apocalypse.  So, a new group is (automagically) created called
>net.news.group.apocalypse, when, after a while, it becomes unused and
>expire finally empties it, and perhaps waiting for a decent interval
>of mourning, it is removed from the newsgroup list, its directory is 
>deleted, and its gone.  This would make the news more like a directory
>tree.  

The real problem is that news groups are insufficently specific,
and that since news groups tend to have high overhead and long life-times,
people are unwilling to create groups which serve more specific
(and smaller) interests. Given this, there is no practical way
for a news reading program to filter out the articles which a particular
persons wants or does not want to read, simply because there is not 
enough information encoded in the header of the message to determine
with any great precision exactly what the article is about.

[ A side note: rn provides a mechanism whereby you can 'kill' discussions
on the basis of the article's header. This is a valient attempt at a solution,
but since it attempts to work within the current framework it is doomed
to failure. Discussions wander under 'Subject:' lines continually, and since
there is a program which relies on the 'Subject:' line for article ordering,
people are even less inclined to alter the 'Subject:' line to reflect the 
content of the article than they were previously. I'm not trying to knock
Larry Wall - I use rn and think it is a wonderful piece of work and a great
improvement over readnews.  I just don't think that the design criteria 
were right...]

What you suggest has the advantage of maintaining some of the structure of
the existing news system, but also has the disadvantage of not truly
addressing the fundamental problem. Your proposal reminds me a great
deal of the 'notes' system, where discussions are organized under
'basenotes'; but since you want to maintain the existing top level
structure of groups, I don't think you will be able to construct truly
intellegent news reading software - at some level we must put more information
in the header, and I don't think that automatic creation and deletion
of newsgroups is going to do the job.

I think newsgroups are wrong. I think we should throw them away.

What I propose is that we should replace newsgroups is a keyword based
news system, in which every article would have a 'subject' and a
'keywords' header line, but would NOT have a newsgroups line. There
would be several thousand keywords, and some software to help the user
choose appropiate sets of words to describe an article. If appropriate
words don't exist in the data base, the user could add them himself,
since one of the design criteria would have to be that individual keywords
would have very low overhead.

This has the advantage that discussions could be as specific as one wants,
and they won't disturb anyone else. Article selection criteria would resemble
regular expressions containing sets of keywords rather than lists of newsgroups.

I can't claim credit for this idea; I originally heard it from Brad Templeton
(brad@looking). He has thought through the idea in some detail and even
wrote up an implementation strategy which was posted to the net some time
ago. If you are interested, I'll ask him to repost the article; it goes
into much more detail than I can here. I don't agree with all of the
reasoning, but at least it is a step in the right direction.

One day I intend to implement a news system which is keyword based.
If we can get a consensus in time, perhaps we can even use keywords to
help save the net.

.......................

Rob Ferguson 	{allegra,clyde,decvax,ihnp4,linus,utzoo}!watmath!riferguson