Design of a Portscan Detection Scheme with Random Packet Sampling

Abstract

Abstract. We consider the parameter design of a TCP portscan detection scheme with packet sampling. We assume that anomalous hosts infected by worms or bots perform portscan and generate a large number of scanning TCP flows as well as its usual traffic. We regard a flow consisting of SYN packets only (SYN-only flow) as a portscan flow and attempt to detect hosts who generate θ SYN-only flows or more, where θ denotes the threshold. In this framework, we present a method for determining the threshold of sampled SYN-only flows and the minimum sampling rate under the constraints of the predefined false positive and false negative ratios. We conduct sampling experiments with public trace data and demonstrate the effectiveness of our design.