Fighting Spam in an ISP Environment - Sponsored Whitepaper

This white paper describes the Roaring Penguin Training Network (RPTN), a mechanism for sharing Bayes votes among multiple CanIt instal- lations.

A recent paper at the 2004 USENIX Large Installation System Administration con- ference by Jeremy Blosser and David Josephsen presented evidence that Bayesian ltering could be quite effective using a shared Bayes database, even among hun- dreds or thousands of different users. This lead to the Roaring Penguin Training Network (RPTN), a mechanism for sharing Bayes votes among different CanIt customers. The theory of RPTN is as follows:

1. Each CanIt installation whose administrator volunteers to submit data to RPTN keeps track of messages that are hand-trained.

2. Periodically, the CanIt installation contacts the central RPTN server and au- thenticates itself for the purpose of RPTN data upload.

3. The central RPTN server grants permission to the CanIt installation to upload a certain number of spam and non-spam signatures.