by Jonathon W. Donaldson
Current network data rates have made it increasingly difficult for cyber security specialists to protect the information stored on private systems. Greater throughput not only allows for higher productivity, but also creates a "larger" security hole that may allow numerous malicious applications (e.g. bots) to enter a private network. Software based intrusion detection/prevention systems are not fast enough for the massive amounts of traffic found on 1 Gb/s and 10 Gb/s networks to be fully effective. Consequently, businesses accept more risk and are forced to make a conscious trade-off between threat and performance.
A solution that can handle a much broader view of large-scale, high-speed systems will allow us to increase maximum throughput and network productivity. This paper describes a novel method of solving this problem by joining a pre-existing signature-based intrusion prevention system with an anomaly-based botnet detection algorithm in a hybrid hardware/software implementation.
Our contributions include the addition of an anomaly detection engine to a pre-existing signature detection engine in hardware. This hybrid system is capable of processing full-duplex 10 Gb/s traffic in real-time with no packet loss. The behavior-based algorithm and user interface are customizable. This research has also led to improvements of the vendor supplied signal and programming interface specifications which we have made readily available.