Research On Virtual Machine Scheduling Strategy In Cloud Computing Environment

With the widespread application of cloud computing,the scale of data centers is gradually expanding.The contradiction between the growing demand for performance and the inefficient utilization of resources makes high energy consumption and low quality of service(QoS)an urgent problem to be solved at present.Virtualization technology is an effective resource management technology in the cloud computing environment.Through the appropriate application of the virtual machine(VM)scheduling strategy,the goals of reducing energy consumption,improving resource utilization,and ensuring service quality can be achieved.In this thesis,the physical machine(PM)load state classification,VM selection to be migrated and VM placement strategy of the VM scheduling process in the cloud computing environment are studied as follows:(1)For the study of the PM load state classification strategy:a prediction-based PM load anomaly classification strategy is proposed to address the problems of lag in the determination of load state and continuity in the transfer of load state,which increases the cost of unnecessary VM migration.The strategy introduces a host-independent threshold mechanism and proposes a load prediction algorithm based on exponential smoothing and the Elman neural network to predict the future load values of PMs.Based on this,a load state classification algorithm based on load abnormality degree is designed to comprehensively evaluate the degree of PMs tending to load abnormality in order to identify the source PMs to be executed for VM migration,so as to reduce the number of unnecessary VM migrations and ensure QoS.(2)For the study of the selection strategy of VMs to be migrated:to address the problem that the existing VM selection strategy is prone to waste of non-overloaded resources and contention of overloaded resources of source PMs,a resource feature-aware VM selection and matching strategy is proposed.The strategy sets the weights of different types of resources according to the multi-dimensional resource utilization of VMs and host PMs,and designs the objective function through the multidimensional resource weight model to reasonably select the VMs to be migrated.For the selected VMs,complementary pairing is performed between individual VMs with low similarity according to the VM load characteristics to serve as the basic VM units in the VM scheduling process,so as to achieve a multi-dimensional load-balanced state and reduce resource waste.(3)For the study of VM placement strategy:in the process of VM placement,the deployment of VMs usually affects the load balancing state of the target PM and generates resource waste,resulting in high energy consumption and low resource utilization,a VM placement strategy based on Adaptive Mutation Particle Swarm Optimization(AMPSO)algorithm is proposed.The strategy takes reducing energy consumption and achieving multi-dimensional load balancing as the optimization objectives,redefines the computational scheme in the Particle Swarm Optimization(PSO)algorithm in a discrete way for the actual scenario of the VM placement problem,and triggers the mutation process of particles with a certain probability when the variance of the population fitness of particles is less than the expected value according to the degree of merit seeking,so as to enhance the ability of global merit seeking of particles to determine a better VM placement scheme and achieve the goal of reducing energy consumption and improving average resource utilization.