End to End Modeling, Analysis and Parallel Simulation of Integrated Network Architectures

点击翻译按钮获取中文摘要

The research focuses on an important, so-far, unresolved issue of end-to-end modeling, analysis, and parallel simulation of Broadband Integrated Service Digital Network (B-ISDN) architectures. ATM (Asynchronous Transfer Mode) is the multiplexing and switching techniques for the B- ISDN architecture. The end-to-end models to predict performance measures such as delay and throughput in order to design an accurate admission control protocol for ATM networks are unsatisfactory. They are either computationally too expensive (And thus impossible to perform on-line in real- time) or too simplistic and inaccurate. There is considerable room for improvement in accuracy, without prohibitive complexity. Also the concept of `Quality Service` (QoS) has been trivialized in most analytical results to be just the cell loss probability. Consequently, results, e.g., in call admission and call policing, are primarily judged by their effectiveness concerning cell loss probability. The introduction of additional measures (e.g., delay and delay jitter bounds) will further restrict the set of possible admissible traffic mixtures, and as a result, the often-used linear call admission method (and the related equivalent bandwidth definition) becomes insufficient. Hence, other call admission techniques may very well be needed. The effectiveness of the call admission algorithms can be investigated while utilizing results from end-to-end models. There is a need to exploit the special control features that can be provided by an ATM network, such as the allocation of buffers to connections in intermediate nodes, or the re- negotiation of the traffic descriptor or of the QoS during the lifetime of a connection. In this project an analytical method based on Pade Approximation is used to analyze the end-to-end model for ATM networks. Also a fast parallel simulation approach is used as a complementary technique. Note that the methods can also be used as general purpose performance evaluation tools and can be applied to a larger set of modeling problems.