Supporting quality of service for internet applications

University of Technology, Sydney. Faculty of Information Technology.Regarding the dominance of IP applications and the requirement of providing quality of service for users, it is critical to provide an scalable network architecture capable of supporting sufficient Quality of Service (QoS). Of the two network models (Integrated Services and Differentiated Services) approved by the Internet Engineering Task Force (IETF) [1, 2], the differentiated service model has gained wider acceptance because of its scalability. Differentiated Services (DiffServ) QoS architecture is scalable but inadequate to deal with network congestion and unable to provide fairness among its traffic aggregates. Recently, IETF has recommended additional functions including admission control and resource discovery to enhance the original DiffServ [2]. In this thesis, we propose a new framework based on DiffServ. The new architecture, called Fair Intelligent Congestion Control DiffServ (FICC- DiffServ), applies the FICC algorithm and control loop to provide fairness among traffic aggregates and control congestion inside DiffServ networks. The augmented architecture is realisable within the existing IP network infrastructures. Simulation results show that the FICC-DiffServ performs excellently in terms of guaranteed fairness, minimised packet delay and jitter, as well as being robust to traffic attributes, and being simple to implement. Moreover, providing end-to-end QoS for Internet applications presents difficult problems, because the Internet is composed of many independently administrative domains called Autonomous Systems. Enabling end-to-end QoS, negotiations between domains is then crucial. As a means of negotiations, inter- autonomous system QoS routings play an important role in advertising the available network resources between domains. In this thesis, the Border Gateway Protocol (BGP) is extended to provide end-to-end QoS. The BGP is selected for two reasons: (1) BGP is an inter-domain routing protocol widely used on the Internet and (2) the use of attributes attached to routes makes BGP be a powerful and scalable inter-domain routing protocol. For end-to-end QoS, a completed framework includes a FICC-DiffServ in each domain, an extended BGP between domains and an admission control at the edge router. Via simulation, we demonstrate the reliability of the BGP-extended architecture, including route selection policy and overhead reduction issues