Author(s): CHEN Pengfei, LI Xinyi, QI Yong, ZHANG Xiaohui, School of Electronics and Information Engineering, Xi’an Jiaotong University
Pages: 100-107
Year: 2016 Issue:  1
Journal: Journal of Xi'an Jiaotong University
Keyword:  cloud computing; availability; VM migration; restless multi-armed bandit;
Abstract: Based on the restless multi-arm bandit(RMAB)approach,a multiplexing strategy of backup virtual machines(VMs)is proposed to resolve the problem of low utilization of backup VMs in the cloud environment,and the optimal condition is given.This strategy regards an individual backup VM as a Markov process with two states,namely"idle"(1)and"backup"(0),and models the scheduling of multiple backup VMs as a Markov decision problem(MDP)consisting of multiple Markov processes.The goal of this strategy is to maximize the utilization of backup VMs without obvious reduction in the system availability under the constraint of limited backup VMs.However,this problem is computationally intractable with traditional dynamic programming methods due to the curse of dimensionality.Therefore,this paper transforms the original MDP problem to a RMAB problem and adopts a simple single-step heuristic policy to resolve it.By calculating the single-step optimal solution,the long-term optimal solution can be obtained.Under specific conditions,the optimal solution of this strategyis guaranteed.The results of simulation experiments show that the proposed policy can achieve the goal of extending the backup ratio between backup VMs to service VMs from 1∶1to 1∶M(M 1)while the failed VM assurance rate is no lower than 96%.Correspondingly,the utilization of backup resources is significantly enhanced.When the failure rate of service VM is low,the utilization of backup resources can be raised 89% compared with the 1∶1backup.The building and operation costs of a cloud platform can be reduced with the help of this backup VM scheduling strategy.