| As the base of digital communication,Internet has been widely studied since its birth.Game theory,as a method of analyzing network behaviors,assuming that communication nodes or data are selfish when competing for network resources,aims to analyze the influence of these selfish behaviors of the players on the network performance.Non-equilibrium of a game means that players can profit by unilaterally changing their strategies,and in this situation,the selfishness of players will affect the entire network.Therefore,analyzing network games can provide ideas for improving Internet performance.In a review of the development of network games,from the early theoretical analysis of abstract networks to the applications of game theory in edge computing and other fields in recent years,it is found that using game theory to analyze Internet network behaviors still requires a lot of research.This dissertation studies the Internet games under TCP/IP protocols,especially the routing game under IP routing protocol and the congestion control game under TCP protocol.The main research contents include the following three parts:(1)Aiming at the existence and quality of equilibrium solutions in Internet games,it is proved that Internet games can be abstracted into atomic games,and there is no equilibrium solution under certain circumstances.When an Internet game has an equilibrium solution,it is proved that the Price of Anarchy of the game is bounded under certain circumstances.(2)For the problem of analyzing IP routing behaviors,a static game model with complete information is proposed.Based on this model,it is analyzed that the main reason for non-equilibrium in a routing game is players collectively avoiding the links with high costs in the previous round.Inspired by this reason,a method to improve the routing protocols is proposed,which appropriately reduces the excessive link cost reference values to improve the link utilization.The experimental results show that IP routing protocols cannot guarantee the equilibrium of the routing games.Even in the load-sensitive routing games,the proportion of packets with non-minimized cost can be as high as 80%,and the actual cost of a single packet can even be 2.58 times its optimal cost.The improvement method can reduce this ratio by 20%and reduce the actual cost by 36%.(3)For the problem of analyzing TCP congestion control behaviors,a dynamic game model with incomplete information is proposed.Based on this model,probabilistic analysis is used to deduce the root of non-equilibrium in congestion control games,that is,the impacts of players unilaterally changing congestion control strategies on their transmission rates are greater than the impacts on the network congestion probability.The experimental results show that TCP congestion control algorithms cannot guarantee the equilibrium of the congestion control games.Players unilaterally changing congestion control algorithms can increase their profits by 43%,but reduce the global profit by 2%.As the transmission time prolongs,the impacts of the players unilaterally changing strategies on their profits will gradually decrease,and the impacts on the global profit will fluctuate around 0. |
