Real-time communication in integrated services packet networks.

One of the main issues in Integrated Services Packet Networks (ISPNs) is providing per-connection Quality-of-Service (QoS) in the heterogeneous traffic environment. To this end, many technical problems such as packet scheduling, resource reservation, traffic policing, and QoS routing must be resolved. In this dissertation, we provide solutions to these problems. We propose a new cell-multiplexing scheme, named TCRM, which provides per-connection delay guarantees in an ATM network. By guaranteeing a minimum throughput, TCRM can provide deterministic delay guarantees under the leaky bucket constraint of input traffic. A scalable implementation of TCRM is presented. In addition to deterministic delay guarantees, TCRM can be used to provide statistical delay guarantees by statistically multiplexing different real-time connections onto a single virtual channel. Such statistical multiplexing enhances network utilization. Using a histogram-based source model, we investigate the cell loss/delay characteristics of this approach. Since a large portion of the end users are connected to ISPNs via Ethernets, providing QoS guarantees over Ethernet is important. To this end, we model the MAC protocol of Ethernet with its collision resolution protocol using a Semi-Markov Poisson process. We derive a connection admission control for statistically bounding the medium access time. Video-on-Demand service requires delivery delay guarantees unlike non-real-time applications, but its delay requirement is not as tight as real-time applications. We propose a new service category, called the semi-real-time class for such applications, and present the network control functions necessary for it. For the source traffic characterization necessary to support the semi-real-time class, we define and derive a statistical traffic envelope. For efficient QoS routing, we propose a flooding-based QoS route selection algorithm. It not only dramatically reduces message overhead and operational cost, but also provides good performance compared to existing QoS-routing schemes based on either request flooding or link-state dissemination. The overhead is reduced by limiting the search area within which connection-request messages are flooded using a topological information kept at local nodes.PhDApplied SciencesElectrical engineeringSystems scienceUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/131459/2/9909920.pd