Improving Resource Utilization via Virtual Machine Placement in Data Center Networks

The resource utilization of servers (such as CPU, memory) is an important performance metric in data center networks (DCNs). The cloud platform supported by DCNs aims to achieve high average resource utilization while guaranteeing the quality of cloud services. Previous papers designed various efficient virtual machine placement schemes to increase the average resource utilization and decrease the server overload ratio. Unfortunately, most of virtual machine placement schemes did not contain the service level agreements (SLAs) and statistical methods. In this paper, we propose a correlation-aware virtual machine placement scheme that effectively places virtual machines on physical machines. First, we employ neural networks model and factor model to forecast the resource utilization trend data according to the historical resource utilization data. Second, we design three correlation-aware virtual machine placement algorithms to enhance resource utilization while meeting the user-defined SLAs. The simulation results show that the efficiency of our virtual machine placement algorithms outperforms the generic algorithm and constant variance algorithm by about 15%-30%.