A Pure Nash Equilibrium-Based Game Theoretical Method for Data Replication across Multiple Servers

This paper proposes a noncooperative game theoretical technique to replicate data objects across a system of multiple servers in order to reduce user-perceived Web access delays. The proposed technique uses computational agents that compete with each other to optimize the performance of the servers that they represent. In a large multiagent system, agents may act in a self-interested manner because of their local and limited knowledge, which may negatively impact the systemwide performance. The optimality of a noncooperative game is typically described by Nash equilibrium, which is based on spontaneous and nondeterministic strategies. However, Nash equilibrium may or may not guarantee systemwide performance. Furthermore, there can be multiple Nash equilibria, making it difficult to decide which one is the best. In contrast, the proposed technique uses the notion of pure Nash equilibrium, which, if achieved, guarantees stable optimal performance. In the proposed technique, agents use deterministic strategies that work in conjunction with their self-interested nature but ensure systemwide performance enhancement. In general, the existence of a pure Nash equilibrium is hard to achieve, but we prove the existence of such equilibrium in the proposed technique. The proposed technique is also experimentally compared against some well-known conventional replica allocation methods such as branch-and-bound, greedy, and genetic algorithms. The experimental setup incorporates GT-ITM and Inet network topology generators and 1998 Soccer World Cup access logs to closely mimic the Web in its infrastructure and user access patterns.