The report looks at performance data from a number of leading commercial fingerprinting and Bayesian content filters. The analysis shows that that fingerprinting systems have become less accurate as spammers have learned to engage in hash busting, and slower, because their developers have responded to the spammers by adding new layers of secondary filters. The bundling of multiple filters has potentially improved detection capacity, but has lowered the speed by which fingerprinting/
The report documents that many commercial antispam offerings are simply front ends that rely on the open source Bayesian core developed by the SpamAssassin project. While more accurate than fingerprinting systems, these first generation Bayesian filters are slower than fingerprinting programs. In 2007, a second generation of content filters have been released that deal with the throughput problem in older Bayesian filters. COMDOM Antispam running on Linux servers for instance combines high accuracy of Bayesian filtering, with a throughput rate that is at least 5 times faster than the speed of leading fingerprint systems. This high performance in second generation Bayesian content filters means significant reductions in the number of servers required to process the same volume of traffic, and lower network TCO.
The analysis is designed for administrators and managers at large enterprises and ISPs who face the rising costs of spam on network infrastructure. COMDOM Software is a member of ICSA Labs Antispam Consortium, an association of ten leading global companies focused on fighting spam. The full text of the technological impact assessment is available at http://www.comdomsoft.com/
