The success of the emerging ATM networks depends both on switch performance at the cell level and routing strategies at the call level. In this dissertation, we address both issues. At the cell level, we propose a measure that captures cell loss behavior and analyze ATM switch performance by computing the distribution of consecutive cell losses. The extremely low loss probability requirements of an ATM switch preclude the use of simulation, calling for the use of analytic and numerical methods. The latter methods involve the construction and solution of the underlying stochastic processes associated with the switch and workload. Since the detailed stochastic process representations of the above are on the order of tens to hundreds of thousands of states, we use a tool called UltraSAN, which allows for the automatic construction and solution of these detailed stochastic processes. We also compute the distribution of the queue length rather than just the average queue length. At the call level, we propose two restoration schemes and a method to analyze their performance in the case of failures in ATM transport networks. The proposed restoration strategies utilize the existing portions of the network after a link or switch failure rather than relying on redundancy while restoring the affected calls. We also propose an efficient routing scheme for multi-class traffic with widely differing call characteristics. We develop as approach based on Markov decision theory and propose an adaptive band-width protection strategy to prevent any specific application type from monopolizing the link resources.
展开▼
