Research On The Cloud Platform Self-adaptive Resource Allocation Method Based On Mixed Prediction

With the development of cloud computing, the pay-on-demand usage pattern has gradually become a trend. In this pattern, cloud computing platform must be capable of self-adaptive to resource allocation at the request of users to improve economic efficiency.Self-adaptive resource allocation technology of cloud platform that is based on prediction can forecast future resource requirements according to its applications' historical operating data so as to make accurate dynamic adjustments to the application resources. At present, the researches on the demand forecasting of application resources of cloud platform mainly focus on a single prediction model or method. But a lack of classification of predicted samples leads to the results that is not accurate enough. Researches on self-adaptive resource allocation lay emphasis on self-adaptive resource allocation of virtual machine, without combining the resource adjustment and the reset of virtual machines. Hereby this paper proposes a method for self-adaptive resource allocation of cloud platform based on mixed prediction. Higher resource utilization efficiency of cloud platform could be reached through mixed prediction models and multi-granularity self-adaptive resource allocation, and the user's SLA(Service-Level Agreement) could be guaranteed as well.Firstly, this paper mainly studies and analyses the current status of the resource allocation of cloud platform, on which base continue to study the modeling method of the placement problems and cost analysis method of reset problems of virtual machines. Secondly through analytical researches on existing prediction models, prediction methods of Markov Chain and FFT that are more suitable for the characteristics of the current cloud application are selected as the basis for a mixed prediction model. Meanwhile this paper also discusses the method of combining these two algorithms to predict the demands of applications'resources.The proposed self-adaptive resource allocation methods of cloud platform based on mixed prediction are classified in accordance with the periodic characteristics of the changing demands of application resources, and different prediction models will be adopted for periodic or non-periodic application. Based on the prediction results, three strategies including resource dynamic allocation strategy for virtual machine , online transfer strategy for virtual machine , and the dynamic reset strategy are respectively adopted to make multi-granularity and self-adaptive resource allocation of cloud platform, thus effectively adaptative to the changing application requirements, reducing the transfer times of virtual machine and the probability of SLA violation, and lessening the number of occupied physical machine for virtual machine as well, all for the ultimate goal of improving the resource utilization efficiency of cloud platform system.In the end, the experiments prove that the cloud platform self-adaptive resource allocation method based on mixed prediction can conduct effective forecast about application resource demands and self-adapt to the resource allocation. Desired effects could be achieved in improving resource utilization efficiency of virtual machine and reducing the occupation of physical machine and SLA.