Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!mailrus!purdue!tut.cis.ohio-state.edu!ucbvax!ucsfcgl!cca.ucsf.edu!wet!epsilon
From: epsilon@wet.UUCP (Eric P. Scott)
Newsgroups: news.newusers.questions
Subject: Re: Usenet is not a BBS (was Re: (VERY) Long dead subjects
Message-ID: <541@wet.UUCP>
Date: 13 Sep 89 19:27:49 GMT
References: <250D145B.4189@ateng.com> <1989Sep12.124629.27897@jarvis.csri.toronto.edu> <14661@bfmny0.UU.NET> <8839@batcomputer.tn.cornell.edu>
Reply-To: epsilon@wet.UUCP (Eric P. Scott)
Organization: Wetware Diversions, San Francisco
Lines: 47

In article <8839@batcomputer.tn.cornell.edu> eacj@tcgould.tn.cornell.edu
	(Julian Vrieslander) writes:
>What is the "flood algorithm?"

Considering when netnews was written, I'd guess that the term
comes from ARPAland.  Each IMP (Packet Switching Node) maintained
a routing table that indicated the shortest path to all the other
IMPs.  The problem was, how do you distribute routing information
if you don't already know how to route?  The solution was to
"flood"--each IMP would send updates to all of its directly-
connected neighbors, which would do the same.  It's called a
flood because it spills out in all directions.

Usenet distribution works pretty much the same way, although
there's some filtering involved.

There are two basic mechanisms for keeping articles from
circulating around forever: Path and Message-ID.

The Path keeps track of where an article has been.  netnews isn't
supposed to propagate an article to any machine that's already
seen it.  For this to work properly, each machine must have a
unique name.

Each article also has a unique Message-ID.  Netnews maintains a
"history database" of every Message-ID it's seen.  If an article
arrives that's already in the history, it stops there.  Each site
normally keeps articles for two weeks, and history for four.


>                             But I am really interested in what happens
>when the system saturates.

"Imminent death of USENET" predictions start to appear.

>                            Does this account for situations where I have
>missed seeing messages that other netters are commenting on?  I always
>assumed that missed messages were due to hardware outages.

Since each article finds its own way, you'll at times see
followups before the base articles.  If you never see an article,
it probably means that something got wiped out upstream.  Many
sites consider their spool partitions relatively expendable, so
hardware failures tend to affect news more than anything else.
One "solution" to this is redundant connectivity.

					-=EPS=-