| In the cloud computing environment, large-scale virtual disk image files need tobe deployed reasonably in order to meet the customers’ increasing needs ofmaintenance, service, and deployment. This requires a rapid and dynamic adjustmentof resources from the data center, so to reduce operating costs, improve resourceutilization, and shorten customer services on-line time. At present, the singularityindex is often used to choose the target host in the cloud computing environment.However, the fact that virtual machines which have different business rely on differentresources is ignored. Also, real-time and effective prediction on the target host machineis insufficient. These lead to excessive migration resulting from unreasonableallocation of resources.Researching the deployment technology of virtual disk image files is aimed at inthe cloud computing environment. Firstly, it analyzes the cloud computing platformarchitecture and load indicators which affect the host’s computing power. ALoad-factor deployment algorithm will be proposed. Secondly, the thesis analyzes theBP neural network model forecasting techniques which based on time series, since thedeployment mechanism requires the hosts’ load indicators forecasting data, and the BPneural network model for time series forecasting techniques has a good practice.Besides, the BP neural network model will be improved and be used to predict thehost’s load. Thirdly, the prediction of BP neural network will be compared with otherforecasting techniques to prove that the improved model has a great advantage offorecasting over others. Finally, the thesis combines the weighted-deploymentselection mechanism and the forecasting data to deploy virtual machines. The virtualmachine migration time will be collected. The experiment proves the availability of the weighted-deployment selection mechanism to postpone the virtual machine migrationand its great importance to improve the cloud platform’s stability and resourceutilization. |
PDF Full Text Request