Quality of service management for multi-service networks

Applications such as distributed virtual simulation, video-on-demand, distance learning, and tele-medicine often have critical requirements for network performance, such as guaranteed throughput and bounded delay. The metrics used to characterize network performance are called quality of service (QoS) parameters. In addition to receiving predictable QoS, many applications would also greatly benefit from special network services. For instance, all multi-party distributed applications, such as virtual simulation, naturally use group communication, which is best supported by network-level multicast. The current Internet is only starting to move towards providing predictable QoS support and at best offers limited multicast service. One of the major obstacles to achieving these goals is the fact that the Internet is actually a collection of independently administered domains. My dissertation addresses the issues of achieving end-to-end QoS and group communication across heterogeneous Internet domains.; One approach to addressing the problem of coordinating network services across domains is to define Service Overlay Networks (SONs). A SON can simplify the process to provide end-to-end QoS by producing end-to-end QoS guarantees with a series of bilateral service level agreements (SLAs) between domains. Specifically, my dissertation addresses two fundamental problems: first, providing end-to-end QoS in an environment where network resources must be shared between real-time and best-effort traffic, and second, efficiently supporting group communication within a network that only offers unicast service.; As a solution for the first problem, I develop a congestion-based pricing model combining with the admission rules for QoS traffic for dynamic resource allocation in Multi-Service Overlay Networks (MSON ), and an analytical model to evaluate the performance of MSON employing the pricing scheme. Experimental results show that the pricing scheme results in improved resource allocation policies between QoS and best-effort traffic as compared to other approaches. Via numerical simulation, I also show that my analytical model accurately predicts network performance under a variety of workloads.; For the second problem, I propose a general-purpose agent-based architecture for the network support of distributed and composable simulation systems. In my approach, group communication between heterogeneous multicast group members is achieved using application layer multicast ( ALM). I develop a set of agent interconnection algorithms to efficiently use ALM. Simulation results show that it is feasible to achieve end-to-end system performance through End-Host Multicast comparable to native network layer multicast.