Cloudmark has made a reputation for themselves with their peer-to-peer technology. Yet peer-to-peer hasn’t always been embraced as the next best thing and is often confused with illegal activity such as sharable/downloadable music files. When asked why he feels the peer-to-peer architecture works, Jacob answers - "In my opinion, the reason peer-to-peer works is the same reason that ebay works. When you empower individuals, even in a tiny way, and take that and lash it together in a community so that one click turns into ten clicks, turns into millions of clicks. We’re still seeing what the power to peer-to-peer is actually. It leverages all the power of all the machines in the network and also, interestingly enough, leverages the people, and it turns what would normally be an insignificant deleting of a message into a powerful piece of feedback for the network, which helps that person and helps everyone connected to the network."
The key difference between Cloudmark’s peer-to-peer technology and say, a file-sharing peer-to-peer technology, is that Cloudmark shares fingerprints, the secure one-way hashes that can’t be turned into the content they represent. They do not ship content around the network, which adds up to a big difference. Cloudmark is essentially shipping around people’s opinions about spam, which people are willing to share freely. One of the challenges that Cloudmark learned early on is that as powerful as the peer-to-peer network seemed to be, the overall opinion of running a peer-to-peer network inside their company didn’t really inspire many folks in the enterprise world. These doubts were channeled that into what Cloudmark calls their spam DNA technology, which has been distilled down to something that can be run at the gateway. At the enterprise level, the same level of spam protection is provided without having to run SpamNet on all the desktops.
"Part of our core advantage is what we call the evolution engine. We have a genetic classifier that is about a generation beyond what other people call Bayesian classifiers. Other companies kind of train the classifier and then send it out to their customers. The problem with that is that with problems and feedback you go through this long iteration that makes improvement a very slow process. What we’ve done is simulate that process, using evolutionary techniques. We literally take the spam DNA cartridge, allow the system to make changes to it and then repeat the process. This is how we get the highest accuracy and lowest false positives," explains Jacob.
By subscribing to our early morning news update, you will receive a daily digest of the latest security news published on Help Net Security.
With over 500 issues so far, reading our newsletter every Monday morning will keep you up-to-date with security risks out there.