InterNetNews: Usenet

transport for Internet sites

Rich Salz, Open Software Foundation

NNTP, the Network News Transfer Protocol, has been labelled the most widely
implemented elective protocol in the Internet. The growth of the Internet has meant more
sites exchanging NNTP data. While the explosive growth in Usenet traffic places demands
on all sites, the goal of fast network access puts particular demands on NNTP hosts.

InterNetNews is an implementation of the Usenet transport layer designed to address
this situation. It replaces the standard UNIX server architecture with a single long -running
server that handles all incoming connections. It has proven to be quite successful, providing
quick and efficient news transfer.


Usenet is a distributed bulletin board system,
built as a logical network on top of other networks
and connections. By design, messages resemble
standard Internet electronic mail messages as defined
in RFC822 [Crocker82]. The Usenet message for-
mat is described in RFC1036 [Adams87]. This
defines some additional headers. It also limits the
values of some of the standard headers as well as
giving some of them special semantics.

Newsgroups are the classification system of
Usenet. The required Newsgroups header specifies
where a message, or article, should be filed upon
reception. Sites are free to carry whatever groups
they want. Most sites carry the core set of so -called
``mainstream'' groups. There are currently about
730 of these groups, and one or two new ones is
created every week.

Messages generated at a site are sent to the
site's ``neighbors'' who process them and relay them
to their neighbors, and so on. Sites can be intercon-
nected -- indeed, on the Internet, this is quite com-
mon. See Figure 1.

Figure 1 : Small Usenet topology (all links are two -way).

The Path header is used to prevent message
loops. For example, an article written at A could get
sent to B , D , C , and then back to A. Before pro-
pagating an article, a site prepends its own name to
the Path header. Before propagating an article to a
site, the receiving host checks to make sure that the
site that would receive the article does not appear in
the Path line. For example, when the article arrived
at site C , the Path would contain A!B!D , so site C
would know not to send the article to A.

Sites also keep a record of the Message -ID's of
all articles they currently have. If D receives an
article from B , it will reject the article if C offers it
later. For self -protection, most sites keep a record
of recent articles that they no longer have. This is
very useful when another site dumps a (usually quite
large) batch of old news back out to Usenet.

For the past few years, the amount of data gen-
erated by Usenet sites has been doubling every year.
A site that receives all the mainstream groups is
receiving over 17 megabytes a day spread out over
11,000 articles [Adams92]. About 20% of the data
is article headers, and while all of them must be
scanned only half of it is must be processed by the
Usenet software. 1

The number of sites participating in Usenet has
been growing almost as quickly. Based on articles
his site receives and survey data sent in by partici-
pating sites, Brian Reid estimates that there are
36,000 sites with 1.4 million participants [Reid91].
A ``sendsys'' message to the ``inet'' distribution in
June of 1989 received about 200 replies in the first
twenty -four hours. A year later, nearly 700 replies
were received. (Sendsys is a special article that asks
all receiving sites to send back an email message,
usually without human intervention; by convention,
inet is primarily the set of sites on the Internet.)

1 Yes, this means that, as far as the software is
concerned, Usenet is over 90% noise.
Summer '92 USENIX -- June 8 -June 12, 1992 -- San Antonio, TX