CoV(t)-based Traffic and Queuing Modeling
Résumé
Accurate performance evaluation and network dimensioning are required for the design and the deployment of new telecommunication networks based on packet-switching technologies (e.g., Ethernet, IP, G/T-MPLS, etc.). In this context, network engineers need simple, efficient and realistic traffic models. In this paper we investigate traffic modeling based on the coefficient of variation CoV(t) of the data rate observed at different time scales t. We derive a simplified CoV(t)-based traffic model, the novelty being that it requires only four parameters easily tunable with the following characteristics: the mean-rate mu, the link capacity C constraining the traffic, the Hurst parameter H and the transition time scale t0 linked to the burst duration. Confronted to some real traffic traces, our model proves to be accurate. We illustrate how to use it in analytical performance calculation by applying it to the "ideal switch" case: thanks to our four-parameter simple model, one can derive formulas for node performance estimation or dimensioning. Finally, we propose a new fitting procedure that automatically translates the four-parameter traffic model of this paper into a superposition of Markov fluid ON-OFF sources, and that is simpler than already existing fitting procedures. This superposition can be used to generate traffic for packet switch simulations, simplifying the set-up of the test traffic to the specification of the four parameters we mentioned above.