Improving the Energy-Efficiency of Virtual Machines by I/O Compensation

With the development and application of cloud computing, virtualization technology has been widely applied in modern data centers to provide performance isolation and elastic resource provision. Although the performance overheads of resource virtualization have been extensively studied, the energy-efficiency loss caused in virtualization layer are seldom mentioned in previous studies. In this paper, we firstly analyze the VM scheduling model and I/O virtualization model from the perspective of energy-efficiency optimization; then, we propose a power-fairness credit scheduling model, which introduces a novel I/O compensation mechanism for achieving quick I/O response and better energy-efficiency at the same time. To investigate the performance and effectiveness of the proposed scheduling model, extensive experiments are conducted by using variety of benchmarks as workloads. The experimental results show that the proposed power-fairness credit scheduling model can significantly reduce the energy-efficiency loss caused by scheduling delay and I/O virtualization. In addition, it also exhibits better robustness and adaptiveness when a virtualized system is facing I/O-intensive workloads or mixed workloads.