Detection of network infrastructure problems using artificial traffic

This paper introduces the approach of monitoring the characteristics of artificial network traffic for detecting routing infrastructure pathologies. The experimental set-up used to develop some aspects of this approach is presented. It consists of several hosts connected to a medium size IP network. These hosts exchange special packets and monitor a selected set of characteristics of this traffic. Based on measurements of characteristics such as packet return-trip time and loss rate, the system extrapolates acceptable traffic behaviour and establishes thresholds that indicate potentially suspect traffic behaviour. Traffic measurements are verified against these thresholds, and the system generates an alarm if a significant discrepancy is observed. The thresholds are continuously adjusted in order to accommodate changing congestion conditions in the network. Other special packets may be sent in order to identify more precisely the nature and location of the problem.