| Recent traffic measurements of existing packet networks and their applications (including Ethernet LANs, WANs, WWW, and VBR video over ATM) show that packet traffic exhibits the property of bursting over multiple time scales, namely the long-range-dependent (LRD) property. The conventional Markov (i.e., short-range-dependent or SRD) traffic models, however, cannot adequately capture the LRD property, and network design or configuration based on the Markov assumption may result in unsatisfactory network performance. To properly design or configure a network which meets its performance requirements, it is important to study the impact of LRD traffic on network performance. This dissertation develops new performance evaluation methods which are effective for analyzing networks with LRD traffic, and this dissertation makes three significant contributions to the study of the impact of LRD traffic on network performance. First, we develop performance evaluation methods which are based on nonasymptotic analysis, and we show that existing asymptotic analysis significantly underestimates the impact of LRD traffic when the buffer size is small. Second, we develop a performance evaluation method which assumes a buffer with finite capacity, and we investigate the significance of the finite buffer assumption. We show that existing analysis assuming an infinite buffer significantly underestimates the network performance, and designing networks using such analysis will cause inefficient network utilization. Third, we develop an on-line performance evaluation method which determines performance sensitivities with respect to network parameters in real time (i.e., establishes how performance measures vary with the changes in network parameters in real time). The method is particularly useful for determining performance sensitivities in queueing systems (e.g., the feedback controlled queueing system with LRD input traffic studied in this dissertation) which are not amenable to existing on-line performance evaluation methods. The performance evaluation methods developed in this dissertation effectively assess network performance and enable network performance optimization for networks with LRD traffic. |
