Research And Implementation Of Live Migration Based On KVM Virtual Machine

Cloud computing has more computing and storage capacity compared with the traditional Internet mode,data era has given rise to the development of cloud computing,as one of the key technology of cloud computing,virtualization can isolate the physical resources effectively,allocate resources to users according to user's requirements,and has low cost,high reliability,easy extension,and the advantages of high security.But because the cloud platform has the characteristics of complex structure and can provide uninterrupted service,the applications in the cloud platform are flexible,the load of each virtual machine is in constant change,so when the server cluster needs to upgrade or cluster load is imbalance,the live migration of virtual machines is needed.The live migration of virtual machine refers to migrate the source VM to another physical host on the premise of providing uninterrupted service,so When to trigger the migration and where to migrate are always two of the emphases for research.To solve the above problems,this thesis proposed a mechanism based on live migration of virtual machine,set up the environment based on NFS shared storage.This thesis mainly includes four aspects,first of all,designed a system of virtual machine live migration based on KVM virtualization technology,this system includes five parts-resource monitoring module,migration operation module,freeze module,prediction module and wakeup module,among these,resource listener module is used to listen to server and virtual machine information,such as memory,CPU,network,10,etc,and according to these information to determine the destination of the host,prediction module can predict the situation when the information exceed the threshold in the near future according to these observed information.Then this thesis proposed a predict migration mechanism based on grey prediction,the main ideas of this mechanism is to predict the probability that the server resource utilization may exceed the threshold in the future according to the information observed by resource monitoring module.Here we set a dual-threshold,one is the upper threshold,and one is the lower threshold,when the predicted value exceeds the upper threshold,indicating that the server is overloaded,the VM on the server needs to be migrated to another idle server,this can make the cluster load more balanced.When the predicted value is lower than the lower threshold,all the VMs on the server need to be migrated out,turn off the idle server to save energy.The predictive trigger mechanism has the following benefits,firstly,avoid the frequent migration of virtual machines caused by the instantaneous shock of server resources,secondly,the live migration of VMs itself will consume resources such as CPU,memory,network and so on,in the traditional migration algorithm,if the upper threshold is determined to be too high,it may lead to server downtime,and the determination of the threshold in the prediction migration algorithm is more flexible,thirdly,determining trigger time based on comprehensive resource utilization takes into account the impact of physical hardware on the system.After that,this thesis proposed a virtual machine selection mechanism to be migrated,designed correlation function between the virtual machine and the server,selecting the virtual machine to be migrated according to the relevance can effectively improve the resource utilization of the cluster.Then in order to choose destination host preferably,this thesis proposed a target node selection mechanism based on multiple attribute decision making,Through resources monitoring module to monitor the server memory,CPU,network bandwidth and I/O usage,each resource attribute as a coordinate axis,it will form a N d attribute coordinate space,and the destination host to be evaluated is one of the point in the n-dimensional space,sort the destination host for selection according to the proximity between destination host and idealized host with TOPSIS method.Lastly,we test the target node selection mechanism proposed in this thesis through the CloudSim cloud platform simulation tool,and tests triggering mechanism based on grey prediction on real host migration.Experiment shows that the proposed algorithm effectively improves the resource utilization of the cluster,and effectively improve the cloud center load imbalance phenomenon.