Repeated Leader Follower Game for Managing Cloud Networks With Limited Resources

Abstract

Virtual resource arbitration, adjustment, and migration are techniques used by infrastructure providers (InPs) to manage limited resources and maximize revenue by serving as many virtual network operators (VNOs) as possible. However, these techniques tend to violate service level agreements (SLAs) and require InPs to compensate VNOs, resulting in reduced benefits. As reported in the literature, no more than 20% of leased cloud resources are actually used. Therefore, InPs and VNOs can both benefit if some of these resources can be returned to the InPs. In this paper, we model a process for negotiating resource returns using a repeated leader follower game. InP (the leader) offers VNOs (the followers) compensation for returning resources with the objective of creating a pool of resources that can be allocated to other VNOs. In this manner, we can maximize the utility of InP and VNO. In particular, an InP increases utility by increasing its revenue, and VNOs increase their utility because their workload can be completed at a lower price owing to received compensations. We numerically evaluate these utilities using the three strategies that we propose and make comparisons against two conventional strategies. Our three strategies focus on revenue maximization of the InP, preferences of VNOs, and the previous negotiation histories. We observe that, after serving new requests using our first and third proposed strategy, the utility of InP increases by 5% and 13%, respectively, compared with using conventional strategies.