Comparison of Nash Bargaining and Myopic Equilibrium for Resources Allocation in Cloud Computing

摘要

Distributed (cloud, cluster, grid) computing systems are becoming popular due to the huge amount of data available nowadays and the complexity of the computations required to handle them. An efficient allocation of computational resource is key to guarantee service quality in terms of execution time and cost. However, the inherent distributed character of these scenarios prevents them from adopting centralized allocation strategies and suggests that approaches inspired or related to game theory can be used instead. However, most solutions available in the literature propose simple techniques based on static allocation scenarios subsequently finding their outcome as a plain Nash equilibrium, which seems to leave some room for improvement. In this paper, we address this issue by considering instead a Nash bargaining solution obtaining a Pareto optimal solution of the allocation problem. We compare the results of this approach with those of a "myopic" strategy that pursues a Nash equilibrium, and we determine that, while both allocation strategies fully utilize the entire system capacity, a Nash bargaining achieves significantly better performance in terms of time spent by the users in the system. This gives evidence for a high Price of Anarchy of the myopic allocation and points out the need for a better allocation policy that makes a more efficient use of the available resources.