Research On Neural Network Based Virtual Machine’s Power Prediction Model

In recent years, due to the vigorous development of the Internet has been rapid increases in the size of the data center, which an increasing number and size of a single data center is also growing. With the popularity of cloud computing technology and mature, the vast majority of data centers began to use the services of external cloud computing cloud services, thereby becoming a cloud service provider. Now external data centers to provide cloud services are consuming a lot of energy every day and energy consumption costs has become an issue which cloud service provider can not be ignored. Therefore, how to save energy and reduce energy consumption has become a key issue to be solved to cloud service providers. Under the Infrastructure as a Service (IaaS) cloud service model, accurately predict the energy consumption of virtual machines(VM) among different physical machines (PM) scheduling strategy for migration and consolidation strategy to develop a virtual machine scheduling is important, at the same time can reduce energy consumption, which is good for the environment; but also conducive to the development of rational pricing strategies to further attract users.This paper studies the impact of PM’s workload to VM’s energy consumption in multiple VM environment. For the problem of considering the PM’s workload, and predicting the power of VM, a single RBF artificial neural networks is proposed to solve this problem. Current researches rarely use neural networks to solve this problem. Firstly, the parameters of VMs are chosen and normalized as the model inputs. Then, the model is trained using the real-world collected data. Finally, the trained model is used to predict the power of a single VM in the PM containing multiple VMs. Experiments show that the proposed RBF NN model is effective for VMs’power prediction and can achieve average errorless than2%, which is much smaller than those of comparative models.