Traffic analysis, source modeling, and performance evaluation of multi-media ATM networks

Asynchronous transfer mode (ATM) has been identified as the underlying transmission mechanism for the future broadband networks, which are expected to support a wide variety of services that exhibit diverse traffic characteristics, while guaranteeing the quality of service (QoS) of individual users. However, some fundamental issues about the networks must be resolved before we can provide end-to-end QoS guarantees while still taking advantage of the multiplexing gain. One of the most important issues is that we must be able to estimate the end-to-end performance measures for each type of traffic along each path through the network. How to characterize the statistical behavior of a traffic source is also fundamental to the analysis and design of broadband networks.Statistical analyses of large sets of traffic measurements have shown convincing evidence for the presence of long-range dependence (LRD) in various kinds of actual network traffic. We use traces of actual traffic measurement data and a set of data processing methods in time and frequency domains to access the statistical features of the traffic and study the effects of traffic higher-order statistics and traffic time scales. We find that higher-order statistics only have minor effects on the network performance compared with that of second-order statistics and marginal distribution. The network performance improvements achieved by using dynamic link bandwidth allocation schemes instead of fixed ones will also be examined.We then propose a family of traffic source models and develop algorithms to estimate the parameters of the proposed traffic source models. Heterogeneous on-off sources with both exponential and Pareto distributed sojourn times are used to synthesize simulated traffic data which will match both marginal distribution and autocovariance function of each trace of actual traffic data. The mapping from on-off source models to the Markovian modulated Poisson process (MMPP) will also be discussed.We then consider a packet-switched network with virtual paths and MMPP traffic sources. We propose an algorithm for simplifying source models for a single-queue problem which greatly reduces the dimensionality of the problem while yields approximate solutions close to the exact ones. We then use this technique in an iterative algorithm for solving problems of multi-link networks with multi-media traffic. End-to-end QoS, as well as performance measures between any two intermediate nodes along a virtual path can thus be calculated.