Research Of Virtual Machine Scheduling Policy In The Cloud Environment

As a new computing model, cloud computing has been received plenty of attention in academiccircles and that is gradually moving from theoretical research to practical application. As afundamental and complex problem, virtual machine scheduling becomes a hot issue in cloudcomputing. Virtual machines scheduling in cloud computing mainly researchs in how to allocate theappropriate virtual machines to users and assign the virtual machines to physical resource to ensureusers’ quality of service and performance of system. This thesis has done a research on the virtualmachine scheduling in cloud computing. The major work in this thesis can be included from threeaspects:(1) Research status of cloud computing and virtual machine scheduling of cloud computing areanalyzed. The concept, characteristics, classificiation, application scenarios and key technologies ofcloud computing are summarized. And then the technology of virtual machine scheduling isanalyzed，which including scheduling model, scheduling features, scheduling targets and schedulingalgorithms. The research on virtual machine scheduling algorithm is introduced and the existingshortcomings is analyzed.(2) For the mapping relationship between tasks and virtual machines in virtual machinescheduling model, a hybrid genetic-bacterial foraging optimization (GBFO) algorithm is proposedto optimize virtual machine scheduling. First, in order to avoid genetic algorithm falling into localoptimal and solve the problem of slow evolution, the fitness variance is introduced in selectionoperator. And the best combination of two algorithms is desigened based on fitness variance. Thebacterial chemotaxis operation based on mutation idea is proposed can not only to adapt a unifiedcoding rule, but also to enhance the bacteria themselves feedback. The bacterial replicationoperation based on simple crossover idea is proposed to enhance the communication betweenbacteria. Finally, GBFO is uesd in virtual machine scheduling. The simulation results show thatGBFO has a faster convergence speed and higher solution accuracy than genetic algorithm andbacterial foraging algorithm. Compared with other scheduling algorithms which have the similargoals, scheduling algorithm based on GBFO has obvious advantages.(3) For the mapping relationship between physical machines and virtual machines in virtualmachine scheduling model, a mulit-objective algorithm is proposed to optimize virtual machinescheduling. First, this thesis establishes bin packing problem model based on virtual machine deployment and solves the model through transforming it into a multi-objective optimizationproblem. The objectives respectively are load-balancing, to improve task-execution efficiency andto reduce energy consumption. Then, non-dominated sorting genetic algorithm is improved on thisthesis. The pruning function in the backtracking is used to confirm the optimal initial population.The normal distribution density function is introduced to restrict elite. The idea of two-dimensionalmatrix is adopted to implement crossover operation. Finally, the improved NSGA II is uesd invirtual machine deployment. The results of simulation show that the virtual machine deploymentalgorithm based on improved NSGA II has better performance on the aspects of task execution time,load-balancing and energy consumption.