Fast Thresholds for Binary Clustering of Packet Traffic Properties

摘要

Many cases of processing traffic can be generalized to splitting values of properties into two clusters -small verses large values. For example, such clustering can be useful to detect gaps between consecutive TCP windows in a packet flow. Fast calculation of efficient thresholds is important for successful clustering. This paper compares performance of a number of threshold models applied to several properties of individual packet flows and as well as aggregate traffic.