Path: utzoo!mnetor!uunet!husc6!bloom-beacon!tut.cis.ohio-state.edu!osu-cis!att-cb!att-ih!ihnp4!ihlpg!ejbjr
From: ejbjr@ihlpg.ATT.COM (Branagan)
Newsgroups: news.admin
Subject: Re: No longer about April Fools (the Path: header)
Message-ID: <5137@ihlpg.ATT.COM>
Date: 5 Apr 88 15:19:59 GMT
References: <1590@sigma.UUCP> <4750002@hpscdc.HP.COM> <47838@sun.uucp> <2047@epimass.EPI.COM>
Organization: AT&T Bell Laboratories - Naperville, Illinois
Lines: 95
Summary: Message IDs and history are sufficient, Path: is not!

In article <2047@epimass.EPI.COM>, jbuck@epimass.EPI.COM (Joe Buck) writes:
> In article <47838@sun.uucp> chuq@sun.UUCP (Chuq Von Rospach) writes:
> >>This "bug" prevents your transmitting the news items you receive back
> >>to your newsfeed. 
> >>And maybe them sending it back to you, and you to them ......
> >
> >It prevents transmiting it back, but the loop prevention is the reason
> >why we have the history file. It comes back, it's recognized as a duplicate,
> >it dies.
> 
> Chuq is right on this one.  Message-IDs and the history file are
> sufficient to prevent loops.

So far, so good.  Message IDs and history indeed are sufficient to
prevent loops.  They are also NECESSARY to prevent duplication of
articles recieved from multiple independent paths through the
network.

>				They aren't sufficient to prevent
> wasted transmissions of articles;

This however is not true.  Rather than checking the Path:, it would
be possible to ask the remote machine if it already has a given
message ID in its history file BEFORE transmission, rather than
waiting to check locally on the remote machine after transmission.

This would be perfectly sufficient to prevent retransmission of
articles back and forth between machines.

It would also prevent the tremendous amount of wasted transmission
of articles by multiple paths through the network for which the Path:
header is not sufficient.

>				    getting rid of Path: will
> significantly increase backbone news traffic (adding roughly one
> full feed's worth of traffic), since every backbone site talks to at
> least two others and passes full feeds back and forth.

Only if article transmission continues to ignore the existence of
history (i.e. eliminating Path: WITHOUT adding a history check
would significantly increase network traffic.  Eliminating Path:
and using history instead would DECREASE network traffic).

So why don't we use history instead?  Anything that would decrease
the cost of the net would seem obviously worthwhile.

The answer seems equally obvious.  HISTORY.

When the origonal netnews software was written, and the network was
much smaller (<20 machines) and more structured Path: was sufficient
to prevent retransmission - there weren't multiple paths through the
network to cause unnecessary retransmission.  As the network grew
this "problem" was ignored - it wasn't a big enough problem to be
worth solving until the network grew to be so large as to make a
solution a major, expensive hassle.

Past history tells us it is impossible to make coordinated changes
in netnews software on such a large anarchistic network.

Hence any change made would have to be made in an upward compatible
manner and phased in slowly over the next umpteen years.

Someone would have to invest significant time and effort to implement
a history/message ID based handshaking in the news transmission
software - including a  new field in the appropriate system files to
tell which transmission scheem to use for each machine to which a given
upgraded machine talks.

Anybody want to invest the time to design, implement, test and
introduce such a feature?  I didn't think so.

Perhaps cost of unnecessary transmission will eventually force the
backbone to propose and implement a solution.  Does anybody have or
care to generate statistics on the cost of unnecessary transmission
by multiple paths?  Without some estimate of the potential savings
it will be impossible to justify such an effort and the almost-gaurenteed
chaos which would temporarily result from a major change in transmission
protocall.

Before I stick my foot too far down my throat, does such a protocall
already exist?  Based on my limited knowledge of netnews internals
and the discussion I've seen here, I don't think so.  If it does,
could someone more knowledgable tell us about it?  If it does exist,
is it limited to major backbone sights? - if so should it be more
widely distributed?

Note that the Path: header would still be needed - as an audit trail to
track down problems like sites which mangle headers.  It would be nice
if the Path: header were more hidden though, to prevent attempted 
misuse.
-- 
-----------------
Jim Branagan
ihnp4!ihlpg!ejbjr
(312) 416-7408 (work)