The Study On Selfishness Of Nodes In P2P Networks Based On Game Theory

With the rapid development of the Internet, there emerged file sharing, live video, distance education and some other large scale multi-user network applications, which brought a huge data transmission problem. P2P is a distributed computing method which can solve the waste of bandwidth and the network congestion problems caused by the huge data transmission in traditional unicast and broadcast mechanisms. In this context, P2P gains considerable development and improvement in recent years.In P2P networks, each user is an independent user that can make its own strategy and there exist some selfish users who want to gain high quality of service with less forwarding load. As we known, existing data transfer methods are equal to all users, therefore, the behavior of selfish users is harmful to the interests of other users, and also a serious impact on network performance and data transmission efficiency. IETF gives high attention to the selfish behavior in P2P networks and have already proposed a variety of mechanisms to curb such behavior.This article uses the ideas of Game Theory to study the behavior of users in P2P file-sharing and P2P live streaming applications. As a result, we proposed two mechanisms to solve the selfish behavior in these two P2P applications:an incentive mechanism based on the Bayes Equilibrium of Game Theory in P2P file sharing system and a Game Theory based incentive mechanism in P2P streaming system.1. We propose an incentive based on Game Theory for the selfish behavior in P2P file sharing system. Unlike the previous complete information game based mechanisms, we define the game in P2P file sharing system as an incomplete information game, in which the resources shared by the nodes is their private information known only by themselves. We define the amount of resources a node shared as its type, the judgment of other nodes’type as its beliefs about other nodes. Each node determines the strategy of its service action based on the resources its type and beliefs. An incentive reputation is proposed to incentivize peers to share their resources and provide services to others, for the services available to the peer are directly depend on its current incentive reputation. A peer can only enhance its reputation by increasing its contribution. Simulation results show that the mechanism can effectively inhibit the selfish behavior of nodes and promote the node to provide better services for others, the number of selfish nodes in the system will converge to a smaller proportion. In the simulation about the response probability of service request changes over time, the performance of the mechanism is superior to the traditional transaction-based model and the Nash equilibrium game-based model. The statistics of the service show that the mechanism can effectively improve the quality of service in the whole system.2. We first construct a multi-tree structure based P2P streaming system in which the streaming media data will be cut into multiple segments and forward in the multicast trees build for them own, effectively solving the uneven load problem and the maintenance of multicast tree.In the forwarding process, each node receives download bandwidth allocated by its parent nodes and allocates download bandwidth to its child nodes. There may exists some selfish nodes that reject or allocate very little download bandwidth to their child nodes in order to save their own resources. We proposed a game theory-based incentive mechanism in which an reputation is used to inhibit the selfish behavior of nodes for the download bandwidth allocated by the parent node is directly depend on the reputation of the request node. By periodically reconstructing the multicast tree, the mechanism will be applied fully and fairly to each node.Simulation results show that the mechanism can effectively inhibit the selfish behavior of nodes and promote the node to allocate more download bandwidth for the child node, so that the average output bandwidth of the node will maintain a good level. Comparing with the scoring-based model and the historical reputation game-based model, the proportion of selfish nodes will converges to a lower level. The statistics of the service show that the mechanism can effectively improve the quality of service in the whole system.