Traac Estimation and Resource Allocation Based on Periodical Wavelet Analyses

In a high-speed network such as asynchronous transfer mode Syk91, Uni94] (ATM) a user can specify traac parameters that describe the intended usage pattern in more detail. These include the peak and sustainable (mean) cell rate as well as the maximum burst size. Based on these parameters the network must decide whether to accept or refuse the connection, and it must reserve resources to guarantee the requested quality of service constraints. Because of the multiplexing of many connections on a link, a certain smoothing is expected that allows the reservation of less than the peak cell rate. Recent publications Lel93, Lel94, Arl95] have revealed the self-similar (fractal) nature Man77, Man83, Pei92] of networking traac. Owing to positive correlation, the multiplexing of several such traac streams onto a link can have a negative impact on queuing performance Geo94, Geo95] because the smoothing eeect appears more slowly than expected. This paper describes a new algorithm called WAAN to derive an eeective capacity based on periodic wavelet analysis applied to a window of traac measurements (cell counts), that can capture and even exploit correlation structures. To cope economically with the numerical complexity of the wavelet transformation, a DSP is required to implement the method. The algorithm is highly adaptive over a wide spectrum of parameters, e.g. workload and intensity of the correlation, thus making it capable of coping with the high traac dynamics expected of future broadband networks. In addition, the algorithm can be applied to diierent conglomerations of connections, making it scalable to very high-speed networks where monitoring and controlling individual connections on every link along the connection's path becomes extremely crucial.