Non-Cooperative and Cooperative Game Approaches for Load Balancing in Distributed Systems

Distributed system is a system consisting of groups of servers and clients where they are connected together to do specific tasks. In order to optimize the system, a load balancer can be used to mitigate the effect of imbalance of the number of clients and servers. Without the load balancer, the servers might be overwhelmed due to response many requests of clients. To implement a load balancer, several methods or algorithms, such as Round-Robin, Least Connection, and Response Time have been applied. In this paper, we focus on game-theoretic load balancing algorithms in a cloud system. We formulate non-cooperative and cooperative game-theoretic models to be used as load balancing algorithms. Clients are the players in each game and the players' strategies are based on the current status of resource utilization of the servers. To obtain the solution, iterative algorithms combined with Discrete-Time Markov Chain (DTMC) are presented. Moreover, the comparison between non-cooperative and cooperative approaches is presented. The evaluation results of both the games are compared with the ones from common techniques, such as round-robin, randomization method, and least connection for effectiveness in terms of request completion time and balancing time.