Accounting architecture and multicast cost allocation in differentiated services networks.

摘要

This dissertation addresses issues in architectural support in networks that use economic principles to manage resources allocation. As the Internet increasingly extends its reach to new users, and emerging applications set the standard for stringent QoS (Quality of Service) requirements, it becomes imperative to devise effective methods for efficient distributed resources allocation. However, the multiple ownership of the networks making up the Internet challenges traditional approaches and compounds the complexity of the issues involved. The variety of user needs call for more individual performance objective enactment. Network economics purport to contribute flexible and scalable solutions to these issues, by emulating human markets that are self-regulated by basic exchange principles local to producers and consumers. The fundamental idea is the use of frameworks based on Game Theory and Pricing Theory that require the resource owner (producer) to set the “rules of the game” and leave to each user (consumer) the selection of an algorithmic behavior that optimizes his utility. Many proposals have been put forth, ranging from optimum pricing to auction models; a central need of these endeavors lies in the design of efficient accounting architecture for collecting, storing, processing and communicating relevant technical information to parties involved in transactions.; In this research, we first develop a taxonomy of the major accounting characteristics with the aim of highlighting important features to optimize for building effective architectures. We then go on to propose a scalable and efficient architecture “TOKENAC: token-based accounting”, that leverages the principle of leaky bucket to transparently monitor traffic and perform accounting functions, thus minimizing overhead and complexity.; Next, we consider the extension of TOKENAC to multicast, where it is important to determine the incidence of each user participation to the overall cost of a joint session, and compute individual shares accordingly. We investigate the general principles of viable allocation methods from cooperative game theory and propose a cascading decomposition of the multicast cost tree down to domain, and then to users. In our framework, all inter-domain interactions are covered by interconnection agreements. Hence, accounting becomes a local issue to each domain with all the flexibility and scalability that entail.