Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!decvax!decwrl!pyramid!hplabs!oliveb!jerry
From: jerry@oliveb.UUCP
Newsgroups: net.news.b
Subject: Re: Sorting batches helpful?
Message-ID: <926@oliveb.UUCP>
Date: Fri, 20-Jun-86 18:55:50 EDT
Article-I.D.: oliveb.926
Posted: Fri Jun 20 18:55:50 1986
Date-Received: Sat, 21-Jun-86 13:50:14 EDT
References: <510@mecc.UUCP>
Reply-To: jerry@oliveb.UUCP (Jerry Aguirre)
Distribution: na
Organization: Olivetti ATC; Cupertino, Ca
Lines: 41
Keywords: sort news batch
Summary: sort batches by posting date

In article <510@mecc.UUCP> sewilco@mecc.UUCP (Scot E. Wilcoxon) writes:
>	When batching news, have sendbatch sort the batch file.
>
>Right now, articles get passed on in the order in which they are received.

Unfortunately that is not true.  For sites using batching it is mostly
true.  For others it is not true at all.  There is no guarantee that
UUCP will send jobs in the order that they are queued.  Also "L" news
feeds can result in a locally posted article getting faster
distribution.

Other people have discussed the various problems with articles showing
up out of sequence.  It is VERY common for a followup to an article to
arrive before the original.

I would suggest sorting the batches by the article posting date.  In
addition to moving the followups after the parent, it would tend to
shorten the worst case transmission time as older articles would have
priority.

There are two possible problems with this.  The most obvious of these is
the extra overhead associated with sorting.  In particular, having to
read and parse the header of all the articles to get their posting dates
would slow down the batching process considerably.  An alternative would
be to have rnews set the modification time of created articles to the
posting date of the article.  This could be done with trivial overhead
and the sort would require doing only an fstat(2) of the files, much
faster than parsing headers.  Having the mtime of the articles set to
the posting time would probably be useful for other purposes.  (Having
the "time-warp" articles show up with a date that hasn't happened yet
should be fun.)

The second problem would be the recurring one of some site suddenly
flooding the net with very old articles.  As the new versions of news
will junk articles which are over the expiration date this should not be
too much of a problem.

Comments?

					Jerry Aguirre @ Olivetti ATC
{hplabs|fortune|idi|ihnp4|tolerant|allegra|glacier|olhqma}!oliveb!jerry