Research On The Server Consolidation Plan For Power-minimizing

Virtualization technology combined with cloud computing built more convenient, reliable and larger new data centers for cloud providers. As the number of servers increasing, the workload on some servers will be low. Research shows that though the workload on it is low, even only 10%, the server still has high power consumption up to more than 50% of that when the load is full. Thus, it is a great waste of power and resources. The traditional server consolidation makes virtual machines encapsulate applications running on multiple physical servers in the cluster, then consolidates them into a small number of servers, and closes the idle servers to minimize the power consumption. But there are three problems. The first is that the evaluation is rough. Then, they did not consider the cost problem. Finally, the power consumption was not calculated precisely before and after server consolidation.On the basis of the analysis of traditional server consolidation method, a problem of server consolidation for power-minimizing is proposed in this thesis, and the solution plan is designed. Choosing the consolidated and destination servers and calculating the power consumption precisely before and after server consolidation are the key in this plan. According to the resources and performance data of servers and the virtual machines, the power consumption can be calculated precisely. According to the evaluation of the server, the consolidated server can be selected. According to the resources of servers and virtual machines, the dynamic virtual machine packing algorithm can be used to reduce the number of servers from the perspective of global optimization. So the process of this thesis is as below. Firstly, determines the server consolidation time according to the CPU utilization. Secondly, uses the principal component analysis (PCA) to choose the principal component related to the power consumption, and then builds system power model and optimize the model to calculate power consumption precisely. Thirdly, uses the multi-objective decision algorithm to select the consolidated server and propose a strategy of global dynamic virtual machine packing, GDVMP, which can significantly reduce the number of servers from the perspective of global optimization, solving the problem of virtual machine packing. According to these, closes the idle server, completes server consolidation plan for minimum power consumption. Finally, through experiment to verify the plan is feasible and effective.Experiments prove that the system power model proposed in this thesis can predict the power consumption precisely, and the method of server consolidation for power-minimizing is feasible and effective. Evaluations of the algorithm demonstrate that the plan proposed in the thesis could increase resource utilization and minimum power consumption compared with the traditional method, and verify the plan practicability and rationality.