Offline and online network traffic characterization

Abstract

This paper investigates a new technique called Bayesian-Block-Analysis (BBA) for analyzing the time varying rate of events. The first goal is to evaluate the accuracy of BBA in identifying the rate changes in synthetic traces that have a given interevent times distribution, and known rate change points. We find that BBA is highly accurate on traces with exponential interevent times and known rate changes, and reasonably accurate with more heavier-tailed interevent times. The second goal is to apply BBA to actual network event traces. And for request arrivals or loss rate traces, BBA identifies significant stationary-rate periods which are qualitatively consistent with previous results obtained with less efficient or less accurate techniques. For packet arrivals to gateways, BBA identifies stationary rate periods that are corroborated by binning the data on a new timescale. Finally, we also show BBonline rate estimation is accurate for synthetic as well as actual system traces. © IFIP International Federation for Information Processing 2007.