A service curve approach for quality of service management in integrated services networks

This dissertation examines some of the fundamental problems related to quality of service management in integrated services networks. Specifically, the issues of end-to-end delay bounds, efficient resource allocation, dynamic connection control, traffic shaping and pricing are addressed.; The end-to-end delay bounds associated with the current proposals for the provision of deterministic quality of service guarantees in integrated services networks can be overly conservative, limiting their utility to guide efficient resource allocation. We first develop a scheduling algorithm, called the Subnet Service Curve based Earliest Deadline first (SSCED) algorithm, which guarantees subnet service curves to virtual connections. This algorithm provides tight deterministic end-to-end delay bounds and controls delay jitter for the guaranteed traffic. We propose a scheme by which it is possible to reduce average delay of the best effort traffic without affecting the end-to-end delay bounds for the guaranteed traffic. We derive this algorithm and present appropriate admission control conditions in a static as well as a dynamic framework.; For many applications, it is desirable that a connection be able to renegotiate its service curve parameters even after the connection establishment phase. We develop a dynamic connection management framework for this purpose. We present some definitions for which it is possible for a connection to renegotiate its arrival and service curves, and control the delay bounds appropriately.; In order to manage the flow of traffic to prevent traffic sources from monopolizing network resources, traffic shaping mechanisms are necessary so that traffic streams can be limited as appropriate. Traffic clippers are devices that discard packets from a traffic stream to make it conform to the required burstiness constraints. We present sample path results for traffic clippers which bound the packet loss rate of clippers in terms of the loss rate of standard single server queues. We also demonstrate that traffic clippers in tandem can be used to enforce burstiness constraints, but such an enforcement is done in a non-optimal manner. In the end, we present some pricing policies using which it is possible to satisfy the economic objectives of network provider and users.