Internet of things DDoS mitigation. Preventing DDoS attacks using learning algorithms on limited hardware

Abstract

DDoS attacks are becoming more and more common, and threatens the current infrastructure of the internet. Cheap new IoT devices have led to a lot of new devices that are poorly secured and can easily be compromised and used for such nefarious purposes. While there are many attemps at solving this problem this thesis looks at a solution which could be applied to typical home router. This would stop malicious traffic before even hitting the internet, as a compliment to the greater effort.

IoT devices typically have fairly simple traffic patterns during normal operations. The system tries to learn these patterns in order to block traffic which would be outside of normal. A home router however is an extremely limited device from a hardware perspective, so a balance has to be struck between learning capability and resource consumption. This becomes especially apparent when considering that most of the chips in home routers doesn't even support floating point operations, which are commonly used for various learning methods.

The proposed system, with the accompanying implementation, shows promising results throughout the testing suite while remaining very low in resource consumption. However dealing with false negatives and implementing the result in a QoS algorithm are still difficult questions. Over all however the solution shows promise and by implementing something like this along with other existing DDoS mitigation efforts a substantial dent can be made in the viability of these attacks.