Call admission control in high speed networks: Modeling and analysis

This dissertation presents the development, modeling, analysis and performance evaluation of three new call admission control (CAC) protocols for Asynchronous Transfer Mode (ATM) based high-speed Broadband Integrated Services Digital Networks (BISDN). The protocols presented are suitable for "On-line" application and provide efficient network resource (buffer space and link capacity) utilization. These protocols simultaneously satisfy two QoS parameters, namely cell loss ratio (CLR) and delay. Two of these protocols are for ON-OFF sources with Poisson output when ON, and the other is for bursty sources.Numerical Optimization Method and Neural Network based method. Numerical optimization method is optimal in the sense that it allows admission of the maximum possible input data rates for given buffer size B and link capacity C at the node and desired CLR and delay. However, it takes time that may be too long for some applications. Hence, the neural network based method is developed. This method gives similar maximum acceptable input data rate as given by the numerical optimization method. However, it takes shorter time to make a CAC decision. Thus the neural network method is optimal and fast. Detailed steps of applying these protocols to an example node and the corresponding results are also presented. The performance of these CAC protocols are evaluated and compared with a two level CAC algorithm and between them. A CAC protocol for bursty sources is also developed and evaluated by using queueing models.Call Admission Control is an important function of high-speed networks and hence these protocols should be useful to the design and operation of many real life networks.Analytical expressions at nodes for CLR and delay have been developed. For ON-OFF sources with Poisson output, M/D/1/K and MMPP/D/1/K queueing models are used. For the bursty sources also, analytical expressions for CLR and delay are developed. These analytical expressions are used to develop CAC protocols. For the ON-OFF Poisson sources and the MMPP node model, two protocols are developed:...