In the best-effort service model of today’s Internet, there is an increasing demand for replacing tho current network with a model in which some packets and applications can be treatcd differently from others based on their requirements. The Differentiated Servicc (DiffServ) architecture aims to preserve the statistical multiplexing nature of the current, Internet, while using scalable, flexible mechanisms to provide a wide rang of Quality of Service (QoS) requirements....[ Read more ]

In the best-effort service model of today’s Internet, there is an increasing demand for replacing tho current network with a model in which some packets and applications can be treatcd differently from others based on their requirements. The Differentiated Servicc (DiffServ) architecture aims to preserve the statistical multiplexing nature of the current, Internet, while using scalable, flexible mechanisms to provide a wide rang of Quality of Service (QoS) requirements.

In this thesis, we first examine the structure of an interior node that support Assured Forwarding (AF) and Expedited Forwarding (EF) Per-Hop Behaviors (PHBs) defined in the Internet Engineering Task Force (IETF). The proposed stochastic model studies the nodo that shares a link capacity between a Premium service representing the EF PHB, and two classes of Assured service, that represent two classes of the AF PHB. In our study, the traffic sources are modeled as Markov modulated fluid flows. We focus on the system where out of profile traffic is dropped at the edge of the networks by the boundary nodes. Thus, each AF queue supports only one precedence.

The study addresses the priority queueing and General Processor Sharing scheduling in the model. Using the decomposition approach, approximations and the spectral analysis, we are able to derive upper and lower bounds on the tail distribution of the buffer content for both AF classes. Numerical investigations show that the lower bound and the upper bound approximation are very tight. Hence, our work greatly improves tho earlier results on DiffServ performance with AF classes services and Poisson traffic sources. We are able to explore the interaction among classes arid the level of differentiation that can achieve.

Such approximation analysis is helpful in many practical design issues, such as bandwidth brokers that perform dynamic admission control of traffic between adjacent DiffServ domains. The model can be extended to study the interconnections between interior nodes and edge marking schemes at boundary nodes to view the architecture as a whole.