Market-based Control for Quality of Services in Network Applications List of Figures

Acknowledgments I would like to express my sincere gratitude to my thesis supervisor, Professor Toru Ishida at Kyoto University, for helpful guidance, valuable advices and worthy discussions. I would also like to thank my other reading comittee members at Kyoto University, Professor Yahiko Kambayashi and Professor Tetsuro Sakai for their valuable comments. I must express my special thanks to the cooperation of Associate Professor Michael P. Wellman at the University of Michigan, who provided the walras market-oriented programming environment to us and took an important role in the design of FreeWalk markets. I am also thankful to his laboratory members for their discussions. I want to thank Professor Tatsuya Suda at the University of Cal-ifornia, Irvine. The stay at his laboratory was a great experience for me. I appreciate the constructive discussions and the kindly advices from Tadashi Murakami and Kouichi Hayashi, for their devotion to the project, hard work and precious discussions. I am grateful to Hideyuki Nakanishi, the primary author of the FreeWalk network conferencing environment, for providing his product and kindly advices. Without FreeWalk, this project should have not been even started. At last, I would like to extend many thanks to all the members of Ishida Laboratory at Kyoto University for their help and fruitful discussions. Abstract Modern networks such as the Internet are shared by humans with heterogeneous objectives which makes the resource allocation non trivial, in that there usually are multiple objective functions to optimize which often conflict with each other. In such an environment, the criteria to evaluate the system performance varies person to person because of their different usage of network application programs, which makes the situation quite complex. We simplify this situation by incorporating a concept called application QoS, which is the quality of the services that are provided by the application programs and enjoyed by their users. We consider that the allocation of application QoS to each user determines the overall system performance, and it is required to establish a method to achieve efficient resource allocation allowing every user to maximize their own utility under the shared environment. Our approach here is to use a computational market, where agents that represent users and network application programs buy/sell goods that represent network resource and application QoS, to achieve efficient resource allocation through price adjustment mechanism. This research explores a decentralized approach that allocates QoS through a dynamic market. In our approach, …