Summary: | End-to-end support for Quality-of-Service (QoS) guarantees is a central and critical issue in the design of the future integrated services multimedia networks. Due to the bursty nature of the network traffic and the statistical multiplexing mechanism offered by the underlying network architecture, how to provide statistical services is especially important in improving network utilization. In this dissertation, we address several aspects of this problem from both the network and the application/end host system perspectives. In the first part of the dissertation, we study the problem of network support for statistical QoS guarantees using the Generalized Processor Sharing (GPS) scheduling discipline, a very important scheduling mechanism because of its ability to provide both session/class isolation and controlled bandwidth sharing. In understanding the bandwidth sharing mechanism of GPS, we develop several key notions and techniques, and present a thorough and comprehensive analysis of GPS under various stochastic settings. We establish performance bounds for both a single GPS server in isolation and a network of GPS servers. Based on these analytical results, we lay out an asymptotic framework for performing call admission control with multiple statistical QoS guarantees for a network employing GPS scheduling and design various optimal and sub-optimal call admission control schemes with varying computational complexity for both session-based and class-based network service models. In the second part of the dissertation, we turn our attention to an important class of applications--real-time transport of stored video. We investigate the impact of video smoothing performed at a video server on traffic characterization and network resource management in the context of real-time transport of stored video using statistical service. We evaluate the effect of video smoothing on statistical multiplexing gains, and design a simple call admission control scheme which employs a parsimonious traffic model. Our work supports the contention that by explicitly exploiting multiplexing gains, network service with statistical QoS guarantees can provide a viable, resource-efficient way to support real-time transport for stored video.
|