Stochastic Game between Cloud Broker and Cloudlet for Mobile Cloud Computing

Abstract

Offloading computational intensive applications to cloud broker is a promising solution for overcoming the limitation of computational resources and energy of mobile devices. The cloud broker sends the jobs to the nearest local cloudlet to guarantee the least delay for the applications. Most of the existing studies consider a fixed amount of virtual machines (VMs) that the local cloudlet would like to rent. In reality, the jobs arriving at the cloud broker and the jobs submitted by the interior users to the cloudlet are both dynamic. In order to address this issue, we use the stochastic game approach in this paper to improve the long-term revenue of cloud broker and local cloudlet. We model the jobs arriving at the queue as a stochastic progress. Then, the cloud broker and the local cloudlet sent a number of jobs to the virtual machine pool in each time slot. Both of them ensure that the submitted jobs are completed, and the cloudlet guarantees that each of the jobs submitted by interior users is executed. We prove the existence of the equilibrium in this game and develop an algorithm to calculate the ߚNash equilibrium. The proposed approach is carefully evaluated based on usage the price information of Amazon EC2. The simulation results indicate that the stochastic game approach can improve the revenue for both cloud broker and local cloudlet.