On modeling concurrent heavy-tailed network traffic sources and its impact upon QoS

Studies of local area network and wide area network traffic have provided much evidence of what has been called “self-similar”, “long-range dependent”, “fractal”, “chaotic”, “heavy-tail”, or “power-tail” (PT) network traffic. These studies have indicated that conventional traffic models (e.g. Poisson) generate overly optimistic performance measurements. In this paper we present an interarrival process which extends previous work in this area regarding a heavy-tailed ON/OFF source model described in Fiorini and Lipsky (1998). Extending their work, we construct a heavy-tailed ON/OFF traffic model which attempts to characterize traffic generated from concurrent sources. The primary application of this process is to model, say, traffic generated by (World Wide Web) WWW servers and/or, in general, Ethernet traffic. Characterizing and modeling the behavior are important for accessing quality of service (QoS) given much empirical evidence of heavy-tailed phenomena in network teletraffic. As a result, we derive an approximation for computing the “critical points” (e.g. system utilization beyond which performance rapidly degrades), and a means to compute the cell loss probabilities for, say, WWW gateways and/or routers