Some aspects of traffic control and performance evaluation of ATM networks

摘要: The emerging high-speed Asynchronous Transfer Mode (ATM) networks are expected to integrate through statistical multiplexing large numbers of traffic sources having a broad range characteristics and different Quality Service (QOS) requirements. To achieve high utilisation network resources while maintaining the QOS, efficient management strategies have be developed. This thesis considers problem control for ATM networks. studies application neural various issues such as feedback congestion control, characterization, bandwidth estimation, Call Admission Control (CAC). A novel adaptive approach based on that uses reinforcement learning is It shown controller very effective in providing general QOS control. Finite Impulse Response (FIR) proposed adaptively predict arrival process by relationship between past future variations. On basis this prediction, flow scheme at input access nodes presented. Simulation results demonstrate significant performance improvement over conventional mechanisms. In addition, an accurate yet computationally estimation multiplexed connections investigated. method, feed forward employed model nonlinear situations measure. Applications admission allocation dynamic routing also discussed. detailed investigation has indicated CAC schemes approximation can conservative prevent optimal use resources. modified therefore overcome drawback methods. Considering sources, we directly calculate aggregate which modelled two-state Markov modulated Poisson via matching four important statistics. We theory deviations provide unified description bandwidths associated multiplexer queueing approximations, illustrating their strengths limitations. more method parameters Bahadur-Rao theorem proposed, refinement original lead higher link utilisation.