Utilizing higher order statistics of packet interarrival times for bottleneck detection

This paper introduces a new approach for determining bottleneck link locations in a network. The considered measurement model is passive monitoring of a backbone link. We analyze the properties of packet interarrival time (PIT) distribution functions of network segments and make decisions whether the extracted properties of a link suggest bottleneck behavior or not. The correlation between bottleneck behavior and packet interarrival time distribution is demonstrated through simulations featuring tighter and tighter bottleneck connections. Locating shared bottlenecks with passive monitoring requires effective metrics for distinguishing seriously congested links from normal or underutilized connections. The current paper presents the third and fourth central moments (skewness and kurtosis, respectively) of PIT distribution as possible and promising metrics for bottleneck detection. According to the simulation results, kurtosis of PITs is found to be a powerful measure of bottleneck behavior. This is further validated by investigation of real measurement data.