A Game-Theoretic Approach Towards Congestion Control

In recent years, there have been explosive applications in the Internet, through which multiple types of data can be transmitted. It has been found that the TCP congestion control mechanism is not enough.Furthermore, in massive information system in network environment, because the system is complicated and the topology and resources of network are changing, survivability such as connectivity, resource distribution becomes serious more and more. In fact, it will be more effective for congestion control if the routers perform queue scheduling disciplines and queue management strategies. Congestion control schemes in the routers have been discussed widely, and new methods based on game theory are investigated. It regards competition of resources between users in the complicated network as n-user uncooperative network game. It is a new field of the research on congestion control.This thesis begins with the introduction of the end-to-end congestion control mechanism of TCP. TCP critically depends on end-users' cooperation, and in the presence of uncooperative users, which will inevitably lead to a congestion collapse. After that the approaches of congestion control in the routers are described. Then, a game-theoretic approach in the routers towards congestion control is presented, which is named MPS (Misbehaving-user Published Scheduling). MPS rewards congestion avoiding users and punishes misbehaving users. With MPS approach deployed in the network, a Nash Equilibrium could be reached. The Problem of excessive congestion due to unresponsive flows is resolved. We also propose three MPS samples (PIS, RIS, PIS) with different weight punishing functions. Simulations illustrate that MPS is effective, fair and stable and has better performance than other congestion control algorithms in the routers.