FBM Model Based Network-Wide Performance Analysis with Service Differentiation

In this paper, we demonstrate that traffic modeling with the fractional Brownian motion (FBM) process is an efficient tool for end-to-end performance analysis over a network provisioning differentiated services (DiffServ). The FBM process is a parsimonious model involving only three parameters to describe the Internet traffic showing the property of selfsimilarity or long-range dependence (LRD). As a foundation for network-wide performance analysis, the FBM modeling can significantly facilitate the single-hop performance analysis. While accurate FBM based queueing analysis for an infinite/ finite first-in-first-out (FIFO) buffer is available in the existing literature, we develop a generic FBM based analysis for multiclass single-hop analysis where both inter-buffer priority and intra-buffer priority are used for service differentiation. Moreover, we present both theoretical and simulation studies to reveal the preservation of the self-similarity, when the traffic process is multiplexed or randomly split, or goes through a queueing system. It is such self-similar preservation that enables the concatenation of FBM based single-hop analysis into a network-wide performance analysis.