Statistical modeling and performance analysis of multi-scale traffic

N.X.Liu and J.S. Baras

22nd Annual Joint Conference of the IEEE Computer and Communications Societies, INFOCOM 2003, San Francisco, CA, March30 - April 3 2003.

Abstract

In this paper we propose a new statistical model for multiscale traffic, and present an exact queueing analysis for the model. The model is based on the central moments and the marginal distributions of the cumulative traffic loads in different time scales. Only the first two moments are needed to characterize the traffic process, which greatly simplifies the representation and estimation. The queueing analysis uses a very general approach and can evaluate not only the steady state performance but also the transient queueing behavior. The analysis reveals that there exist two classes of packet losses, the absolute loss and the opportunistic loss, both of which can be examined exactly with the method. Based on the statistical model and the queueing analysis method, a compound model is constructed for the practical multiscale traffic, and its performance is evaluated from various aspects. This work provides a good basis for practical application of the multiscale traffic characterizations in network dimensioning and resource management.