Next: C-News Up: Netnews Previous: What is Usenet Anyway?

How Does Usenet Handle News?

Today, Usenet has grown to enormous proportions. Sites that carry the whole of netnews usually transfer something like a paltry sixty megabytes a day.gif Of course this requires much more than pushing around files. So let's take a look at the way most systems handle Usenet news.

News is distributed through the net by various transports. The historical medium used to be UUCP, but today the main traffic is carried by Internet sites. The routing algorithm used is called flooding: Each site maintains a number of links (news feeds) to other sites. Any article generated or received by the local news system is forwarded to them, unless it has already been seen at that site, in which case it is discarded. A site may find out about all other sites the article has already traversed by looking at the Path: header field. This header contains a list of all systems the article has been forwarded by in bang path notation.

To distinguish articles and recognize duplicates, Usenet articles have to carry a message id (specified in the Message-Id: header field), which combines the posting site's name and a serial number into ``<serial@site>''. For each article processed, the news system logs this id into a history file against which all newly arrived articles are checked.

The flow between any two sites may be limited by two criteria: for one, an article is assigned a distribution (in the Distribution: header field) which may be used to confine it to a certain group of sites. On the other hand, the newsgroups exchanged may be limited by both the sending or receiving system. The set of newsgroups and distributions allowed for transmission to a site are usually kept in the sys file.

The sheer number of articles usually requires that improvements be made to the above scheme. On UUCP networks, the natural thing to do is to collect articles over a period of time, and combine them into a single file, which is compressed and sent to the remote site. This is called batching.gif

An alternative technique is the ihave/sendme protocol that prevents duplicate articles from being transferred in the first place, thus saving net bandwidth. Instead of putting all articles in batch files and sending them along, only the message ids of articles are combined into a giant ``ihave'' message and sent to the remote site. It reads this message, compares it to its history file, and returns the list of articles it wants in a ``sendme'' message. Only these articles are then sent.

Of course, ihave/sendme only makes sense if it involves two big sites that receive news from several independent feeds each, and who poll each other often enough for an efficient flow of news.

Sites that are on the Internet generally rely on TCP/IP-based software that uses the Network News Transfer Protocol, NNTP.gif It transfers news between feeds and provides Usenet access to single users on remote hosts.

NNTP knows three different ways to transfer news. One is a real-time version of ihave/sendme, also referred to as pushing news. The second technique is called pulling news, in which the client requests a list of articles in a given newsgroup or hierarchy that have arrived at the server's site after a specified date, and chooses those it cannot find in its history file. The third mode is for interactive newsreading, and allows you or your newsreader to retrieve articles from specified newsgroups, as well as post articles with incomplete header information.

At each site, news are kept in a directory hierarchy below /var/spool/news, each article in a separate file, and each newsgroup in a separate directory. The directory name is made up of the newsgroup name, with the components being the path components. Thus, comp.os.linux.misc articles are kept in /var/spool/news/comp/os/linux/misc. The articles in a newsgroup are assigned numbers in the order they arrive. This number serves as the file's name. The range of numbers of articles currently online is kept in a file called active, which at the same time serves as a list of newsgroups known at your site.

Since disk space is a finite resource,gif one has to start throwing away articles after some time. This is called expiring. Usually, articles from certain groups and hierarchies are expired at a fixed number of days after they arrive. This may be overridden by the poster by specifying a date of expiration in the Expires: field of the article header.

Next: C-News Up: Netnews Previous: What is Usenet Anyway?

Andrew Anderson
Thu Mar 7 23:22:06 EST 1996