| Summary: | This thesis explores issues regarding the implementation of a real-time traffic characterization algorithm (the Hurst-parameter algorithm) for Asynchronous Transfer Mode (ATM) data networks. Because of the multiplexing of data cells from a variety of sources at ATM network nodes, traffic over ATM is highly statistical in nature, and also exhibits the qualities of long-range dependence and self-similarity. Conventional traffic measures (e.g., peak cell rate, sustained cell rate) do not measure these long-term properties, and thus fail to properly characterize ATM traffic, which can lead to inefficiencies in bandwidth allocation and traffic control strategies. The Hurst-parameter algorithm methodology, developed at TRLabs Winnipeg, calculates the Hurst-parameter, which is a measure of self-similarity for the traffic stream. This thesis attempts several implementations on a variety of platforms in an attempt to determine how best to implement the algorithm in real-time. Implementations include a parallel DSP implementation, and a single-processor implementation that can run in real-time (i.e., with real-time ATM traffic). As an aside, this thesis examines implementation issues through a DSP system design tool called Cossap. The results of the implementations reveal that the Hurst-parameter algorithm can run at the OC-3 rate of 2.83$\mu$s per cell using the parallel DSP implementation. The real-time implementation, although it cannot operate at full OC-3, demonstrates the operation and usefulness of the Hurst-parameter as a traffic measure. Finally, the Cossap simulations provide a non-real-time method to explore how changes in the algorithm (e.g., varying the number of data points used in determining the Hurst-parameter) can affect the resulting Hurst-parameter values.
|
|---|
