| Cloud computing has become the most promising research direction in the field ofdistributed systems. The VM (virtual machine) technology is the most crucial to cloudcomputing for achieving high flexibility and scalability. Live VM migration technology is thesignificant application and embody of the flexibility and availability of VM technology.Currently, live migration is widely used for the maintenance management in virtualized cloudcomputing data centers. Live VM migration technology is composed of live VM migrationmechanism and live VM migration policy. The live VM migration mechanism has focused onthese approaches which can achieve the live migration of VMs while the concern of the liveVM migration policy is the location selection problem of target host of live VM migration. It isthe migration policy of live VM migration that this paper is focusing on.The research of live VM migration policy has been involving the location selectionproblem of target host of live VM migration. Since it is a selection optimization problem, oneor more evaluation criterions is necessary. At present, green cloud data center has become aresearch hotspot of virtualized cloud computing architecture. And load balancing has also beenone of the most important goals in cloud data centers. Thus, in general, power saving or loadbalancing is chosen as the criterion and the goal in these research fields relating to live VMmigration policy. However, this paper has focused on the live VM migration policy for powersaving and load balancing. It is aiming to find out a migration policy, such that the totalincremental power consumption caused by the migrated VMs onto hosts is minimized and aftermigrating the load is balanced as much as possible, while maximizing the performance byfulfilling the resource requirements of maximum number of VMs. So, this paper has convertedthe location selection problem of live VM migration policy for power saving and load balancingto a constrained multiple objective combinational optimization problem. To formalize theproblem, this paper has designed and employed the standard deviation of the residual load ratesof the hosts to measure the load balancing degree after migrating while employing theprediction algorithm to calculate the total incremental power consumption of the cloud datacenter after migrating. And the performance constraint is achieved and denoted by the numberof all successful live VM migration events. The number is obtained by a well-designed formula. To address the multiple objective combinational optimization problem, this paper has proposeda novel heuristic approach based on improved Genetic Algorithm, and on this basis, built a liveVM migration policy model. Specific tasks of this paper are as follows:(1)The research significance and the starting point of this paper is introduced. The currentresearches of live VM migration mechanisms and live VM migration policies are presented athome and abroad. The research progress in the field of live VM migration policy for powersaving or load balancing is introduced primarily and the advantages and deficiencies ofthe relevant researches are analyzed.(2)Firstly, the basic concepts and theories of cloud computing are introduced. Secondly,VM technologies and the field of live VM migration policy are presented in detail. Thirdly,multiple objective combinational optimization problems and the concept of Pareto optimalsolution are aroused and introduced. Finally, Genetic Algorithm and Simulated Annealing ideautilized by this paper are presented briefly.(3)This paper has provided a live VM migration policy model and presented itsarchitecture and logic execution flow. Subsequently, the relevant contents of the predictionalgorithm used in live VM migration policy of this paper are presented. This paper proposes anovel migration policy MOGA-LS (Multiple Objective Genetic Algorithm-Location Selection)for power saving and load balancing, which is a heuristic and self-adaptive multi-objectiveoptimization algorithm based on the improved GA (genetic algorithm). This paper has presentedthe specific design and implementation of MOGA-LS such as the design of the geneticoperators, fitness values and elitism etc. This paper has introduced the Pareto dominance theoryand the SA (simulated annealing) idea into MOGA-LS as well as has presented the specificprocess to get the final solution. And thus the whole approach achieves a long-term efficientoptimization for power saving and load balancing.(4)This paper has evaluated and verified the proposed MOGA-LS algorithm. The finalexperimental results demonstrate that MOGA-LS evidently reduces the total incremental powerconsumption and better protects the performance of VM migration as well as achieves thebalancing of system load compared with the existing research. It makes the result of live VMmigration more high-effective and meaningful. |
PDF Full Text Request