Abstract
End-to-end support for quality of service guarantees is a centraland critical issue in broadband Internet. In this dissertation,we address several aspects of this problems from the network perspectives.In the first part of the dissertation,we apply matrix-analytic approach tothe examination of the loss behavior of a space priority queue.In addition to the evaluation of the long-termhigh-priority and low-priority packet loss probabilities,we examine the bursty nature of packet lossesby means of conditional statistics with respect tocritical and non-critical periods that occur in an alternatingmanner.These performance measures greatly assist the spacepriority mechanism for determining a proper threshold.The overall complexity of computing these performancemeasures is of the order O($K^2m_1^3m_2^3$), where$K$ is the buffer capacity and $m_1$, $m_2$ arethe numbers of phases of the underlying Markovian structuresfor the high-priority and low-priority packet arrivalprocesses respectively.Thus the results obtained are computationally tractable andnumerical results show that, by choosing a proper threshold,a space priority queue not only can maintain the qualityof service for the high-priority traffic but also canprovide the near-optimum utilization of the capacity forthe low-priority traffic.In the second part of the dissertation,a per-connection end-to-end call admission control (CAC) problemis solved to allocate network resourcesto an input sessionto guarantee its quality of service (Qos) requirements.In conjunction with the solution of the CAC problem,a traffic descriptor is proposed to describethe loss rate and the delay bound Qos requirements ofthe connection to be set upas well as the statistical characteristicsof the associated input traffic which is modeled asa linear mean functionplus a (zero-mean) fractional Brownian motion.The information in the traffic descriptor is sufficient todetermine the allocation of channel bandwidth and buffer spaceto the input trafficin a network which employs leaky bucket shapers and scheduling algorithmsto guarantee the Qos requirements.The CAC problem is solved byan iterative algorithm of whichthere are two stages in each iteration:one is responsible for the search of a candidate end-to-end routing pathand the other for the verification of the legitimacy of this candidate pathto meet the Qos requirementsand for the allocation of resources in such a legitimate path.In the third part of the dissertation,we turn our attentation to traffic modelingproblem.In considering self-similar network traffic transmission throughand statistical multiplexing on a high speed network, problemssuch as cell loss and delay must be well evaluated. Beforehandling network performance issues, an accurate statisticalcharacterization and modeling of self-similar network trafficsource should be done at first. Some performance measures of anetwork carrying self-similar network traffic are over-estimatedby conventional stochastic models due to their lack of long-rangedependence to characterize self-similar network traffic.In [75], Shih proposed a multiscaleon-off model with sufficient long-range dependence for MPEG codedVBR video traffic.This model was further investigated by Wei in [84].Different from the empirical justification in [75] and[84], we theoretically establish in this dissertation thelong-range dependency of the multiscale on-off model by provingthat the auto-covariance function of the multiscale on-off modelhas hyperbolic decay asymptotically when the number of layersapproaches to infinity.Queueing analysis of the multiscale on-off model is accomplishedby using the large deviation technique and stochastic fluid flowmethod.The effect of the long-range dependence of the multiscale on-offmodel to the hypothetical queue in a multiplexer is observed.