Researches In Virtualization Resource Allocation Algorithm In Green Data Centers

With the rapid advancements in information technology and the Internet,the demands of many industries and enterprises for the services of data centers continue to grow,which enables the quantity and scale of data centers have been expanded exponentially.The rise of cloud computing,has changed the traditional usage method of IT infrastructures and deployment pattern of software,and brought a new upsurge of building large cloud data centers.To maximize the energy utilization efficiency,the primary goal of data center is to realize green and energy-saving,and it has great significance to build green data centers.The most direct and efficient way to realize energy saving is,to ensure that resource in data centers be reasonably integrated and assigned.Virtualization technology,an effective technology to improve usage of resources,is widely adopted to realize resource sharing in cloud data centers.Therefore,how to bring the superiority of virtualization into full play to explore energy-efficient virtual resource allocation schemes,has attracted increasing attention.Our work in this paper is mainly focused on virtual machine placement(VMP)and virtual machine migration(VMM).Most policies try to concentrate VMs from all requests on as fewer physical machines as possible in order to minimize the number of activated physical machines,while ignoring the impact on the service of cloud tenant caused by a server failure described.Aiming at the above problem,this paper designs a VMP strategy named Dependability based Distributed Virtual Machine Placement algorithm(D2VMP),which limits the upper number of virtual machines from the same request assigned on each physical machine by setting a failure tolerance value for each request,to ensure the dependability of user service.Moreover,to ensure improving resource utilization,D2VMP divides the data center into several clusters,and thus,the number of activated clusters can be adjusted dynamically according to application scale,which enables the requests required by different tenants to reuse power-on physical machines.Extensive simulation results demonstrate that D2VMP performs well,both in terms of the cost of cloud service providers and the experience of cloud tenants.Furthermore,aiming at the virtual machine migration problem in the dynamic environment based on the variable resource requirement of applications in data centers,this paper proposes a MinCost Virtual Machine Migration(MinCost)algorithm.This algorithm presents a migration trigger strategy,which forecasts the resource demand of each VM according to its historical data by using auto-regression(AR)model firstly,and then to decide whether a migration should be performed by predicting the changing trend of server resource utilization.Moreover,traditional virtual machine migration policies select VMs to be migrated firstly and then determine where to migrate,which cannot obtain the global optimal migration solution.Therefore,this paper introduces the migration cost matrix,and an optimal migration pair<VM to be migrated,PM selected as destination>would to be recommended from an overall point of view through searching the matrix.Our simulation results reveal that the proposed algorithm is superior in migration performance energy efficiency to algorithms compared.