Analyzing packet interarrival times distribution to detect network bottlenecks

Abstract

This paper analyzes the properties of packet interarrival time (PIT) distribution functions of network segments including bottlenecks. In order to show the correlation between bottleneck behavior and packet interarrival time distribution, the alteration of probability distribution function (PDF) is observed through simulations including tighter and tighter bottleneck connections. The process of network bottleneck detection by passive monitoring requires effective metrics for distinguishing seriously congested links from normal or underutilized connections. The paper evaluates the third and fourth central moments (skewness and kurtosis, respectively) of PIT distribution as possible metrics for bottleneck detection. Simulation results as well as real measurement data analysis showed that PIT kurtosis can be a powerful measure of bottleneck behavior. © 2006 International Federation for Information Processing.