Aggregation of Long-Range Dependent traffic streams using Multi-Fractal Wavelet Models

The operation of Diffserv QoS service provisioning requires the aggregation of IP or MPLS traffic streams into a limited number of DijjServ classes and queue them accordingly. To be able to analytically estimate the QoS of each DiffServ class, the characteristic of the class aggregate traffic must be determined This paper presents an analytical technique to characterize aggregate traffic. It is widely known that IP traffic stream is highly bursty and correlated, and can be modeled as a Long Range Dependent traffic stream. By using Multi-fractal Wavelet Models (MWM) to represent each Long Range Dependent traffic stream, the proposed technique calculates MWM model parameters for the aggregate traffic and uses them to estimate the IDC of the aggregate traffic. The calculated MWM parameters are also used to estimate QoS received by the traffic aggregate. Both analysis and simulation are used to examine the performance of the proposed technique for various traffic conditions and scenarios. The MWM parameters of a number of real traffic traces are estimated and used to determine the characteristics of the aggregate traffic. The analytically derived parameters are compared to those directly measured from the simulated aggregate traffic. The results show that the calculation technique is accurate and can be effectively used in DiffServ network.