Research On Reliable Resource Scheduling Mechanism In Cloud Datacenters

Recent years have witnessed the large-scale applications of cloud com-puting with more and more deployments of cloud applications in datacenters.Moreover,the rapid development of virtualization technology makes cloud applications more efficiently share the physical resources of datacenters.By virtualization technology,cloud applications are commonly deployed in the vir-tual clusters(VCs).A VC contains multiple virtual machines(VMs),and those VMs need to be placed on physical machines(PMs)in the datacenter.How to assign and manage the physical resources for cloud application has already been one of the hot issues in industry and academia,as the virtualization-based resource scheduling problem.Virtual Machine Placement(VMP)and Virtual Cluster Allocation(VCA)are two kinds of representative problems thereby.However,the failures of equipment in datacenters will affect cloud applications’ reliability,which is a very important aspect of Quality of Service(QoS)and receives much attention on resource scheduling issues.Therefore,reliability-related resource scheduling problems belongs to a significant branch problems of resource scheduling issues.The problems of VMP or VCA,and reliability-aware resource scheduling have received widespread attentions and have obtained many valuable research results.However,there exist some shortages in considering optimization of traffic,energy consumption,and resource on this basis.Hence,this paper focuses reliability-aware resource scheduling upon VMP and VCA problems,and makes the following contributions:1.This paper proposes an efficient approach for reliability-guaranteed and traffic-aware VMP problems in the datacenter.This approach considers the failures of PMs,and builds a reliability-related model from the aspect of the VM-loss.This model is formulated as the reliable VMs ratio by Infinite Norm.This approach indicates that loose placement schemes tend to enhance the reliability of VM-based applications,but will lead to much traffic consumption.For balancing the reliability and the traffic consump-tion,this approach formulates the constrained single-objective optimiza-tion,and analyzes the complexity of the formulated problem.Based on the minimum k-cuts of graph,this approach uses a VMs-partition-cluster ap-proximation algorithm to guarantee the reliability constraint and optimize the traffic consumption in the datacenter.Experimental results shows that,when the reliability parameter requirements are between 45%and 75%,the proposed method can reduce the bandwidth consumption more than the comparative methods.2.This paper proposes an optimization approach for enhancing the reli-ability and minimizing the bandwidth usage for VCA problems.This approach considers the failure rates of switches/PMs,and the impacts on the VM-loss,to model the reliability-related risk cost.Besides,the above mentioned reliable-VMs ratio is used to enhance the reliability of the VC.This approach also analyzes the impact of different topologies of datacen-ters on the traffic consumption and builds the corresponding model.Based on these models,this approach formulates the constrained joint-objectives optimization problem.An efficient evolutionary algorithm improved on Biogeography-based Optimization(BBO)algorithm is raised for solving the optimization problem.Experimental results shows that,compared with the efficient evolutionary computing method,the proposed method optimizes the reliability and bandwidth consumption by 6.80%and 1.18%,respectively.3.This paper proposes an efficient approach for reliability-aware and energy-aware multi-objective VCA optimization problems.This multi-objective optimization approach aims to obtain the trade-off among reliability,en-ergy consumption,average resource utilization,and resource load balance.This approach indicates that reliability and energy consumption performs conflictly on optimization of VCA.The existing researches consider these two objectives,but there is still a lot of optimization spaces in average re-source utilization and resource load balance.Based on the corresponding models,the constrained multi-objective optimization problem is formu-lated.This approach also presents an improved evolutionary algorithm based on BBO for solving the multi-objective optimization problem.The proposed algorithm uses two kinds thresholds and the related mechanisms to perform more effectively and efficiently.Experimental results shows that,compared with the highly efficient evolutionary computing algorithm,the average multi-objective improvement rate of the proposed method is 47.63%.