Many performance analysis techniques for Asynchronous Transfer Mode (ATM) networks are limited to simple traffic assumptions, topologies and switch architectures. This is because anything but the simplest of traffic or network models can make simulations too time-consuming or queueing analysis intractable. Consequently, current research focuses on expanding the scope of traffic and network conditions that yield to efficient analysis. This thesis contributes to that goal with a framework for computationally efficient end-to-end queueing analysis of ATM networks. Specifically, this dissertation describes techniques for obtaining steady-state Quality of Service (QOS) measures for simple models of integrated bursty (VBR), constant bit-rate (CBR), and available bit-rate (ABR) traffic. Novel approaches are presented for the analysis of networks of output queueing switches with arbitrary speed-up and buffer size, and with several allowable buffer sharing strategies. Analytic techniques are also developed for flow control mechanisms such as backpressure, rate-based control for ABR traffic, and traffic shaping. An analysis tool with graphical interface has been built to make this work easily accessible.; The traffic modeling and queueing analysis in this work make use of the fluid flow approximation which takes advantage of the small size of the ATM cell compared to buffer sizes. The benefit of this approach is independence of computational complexity from buffer size, making the analysis more flexible than many other approaches. Specific contributions of the work include the following: (1) complex buffer sharing strategies are analyzed with a novel iterative approach that approximates the shared buffer with a simpler to analyze partitioned buffer; (2) queueing analysis is developed for reactive traffic shaping mechanisms in which arrival processes depend on queue length; (3) simple fluid-based edge and inter-node traffic models are introduced for VBR, CBR and ABR traffic streams; and (4) tractable models of traffic mixing processes and ABR traffic streams are developed which yield accurate QOS measurements.
展开▼
