A game-theoretic approach for cost-aware load balancing in distributed systems

In this paper, we consider a load balancing problem in distributed systems, that has two conflicting objectives: (i) minimizing the users' expected response time and (ii) minimizing the total monetary cost incurred by each user. To satisfy both the objectives simultaneously, we consider the objectives in an integrated manner and formulate the problem as an optimization problem. We then cast it into a game-theoretic setting and model the load balancing problem as a non-cooperative game. To solve the game, we characterize the best response strategy for each player, and derive a decentralized algorithm called Cost-Aware Load Balancing Algorithm (CALBA). We conduct a rigorous experimental study to demonstrate the effectiveness of CALBA. Further, to establish the effectiveness of CALBA, we compare it with three other load balancing schemes, i.e., MinRT, MinCost, and GPMS, using various system configurations such as varying system size, varying system utilization, and system heterogeneity, across multiple performance indicators. The computational results show that textitCALBA outperforms the competitive schemes by reducing the response time and cost, and unlike others, CALBA produces an allocation of load which guarantees fairness (in terms of response time) between the users. In a nutshell, the results demonstrate the suitability of CALBA in realistic scenarios as it is an adaptable and feasible approach to get a cost-aware load balancing solution. (C) 2020 Elsevier B.V. All rights reserved.