Optimizing providers' profit in peer networks applying automatic pricing and game theory.

摘要

This research exploits the agility of game theory by synthesizing economic theories and Internet traffic engineering techniques to optimize the profit of Internet Service Providers (ISP), and to meet the customer desire of automatic subscription from any provider that offers the lowest price.; We propose a new Automatic Price Transaction-based One-to-Many Peer Network architecture that facilitates customers' options for subscribing to services from providers based on the negotiated price. This model is for enterprise-provider IP peer networks or customer-provider wireless networks. In this model, customers and providers perform simultaneous price negotiations by a Sealed-Bid-Reverse auction protocol. We suggest Session Initiation Protocol (SIP) entities and call flow to implement the mechanism. Our model extends the one-to-one IP peering architecture (IP Network-Network-Interface) of the Alliance for Telecommunications and Industry Solutions (ATIS). Our model also extends the one-to-one Online Charging architecture of the Third Generation Partnership Project (3GPP).; Implementation of the architecture causes strategic interaction among the providers; thus, a game theory model is required to compute the service price and to optimize the providers' profit.; We propose a new game theory model---the Providers Optimized Game in Internet Traffic---to optimize providers' profit in the proposed architecture subject to constraints of network architecture, traffic pattern, and game strategies. This model determines strategic price using a myopic Markovian-Bayesian game of incomplete information and an extension of previous work based on the Bertrand oligopoly model. Our model is sensitive to the dynamic Internet traffic demand, the congestion in networks, and the service class. Selecting a strategically appropriate price is one of our methods to optimize profit; the others are minimizing the network congestion sensitive cost and optimizing routes. The model associates a congestion indicator---the mean IP packet count in a network queue system---with the service cost. An M/M/1 queuing analysis determines the mean packet count. The model applies two well-known non-linear programming techniques, the Gradient Projection algorithm and the Golden section line search, to minimize the mean packet count and to optimize routes in providers' networks.; This dissertation presents the novel models, validates the models by analyses and simulations, evaluates advantages of the models, determines providers' the best strategies for optimizing their profit, and introduces traffic-engineering applications. (Abstract shortened by UMI.)