iMeter: An Integrated VM Power Model Based on Performance Profiling
Hailong Yang, Qi Zhao, Zhongzhi Luan, Depei Qian, Ming Xie, Jason Mars, Lingjia Tang · 2013
Power capping and VM migration have become the most widely adopted power saving solutions in cloud service data centers. However, without the capability to track the VM power usage precisely and portably, the combined-effect of the above two techniques could cause severe performance degradation to the consolidated VMs, and thus violating the users' service level agreement. Previous studies resort to vendor specific performance counters and segregated power models for each server component to estimate the power consumption of VMs. However, the lack of portability with such an approach leads to limited applicability for heterogeneous platforms commonly found in modern data-centers. In addition, segregated modeling approach neglects the correlation and interaction among different performance counters, leading to inaccurate prediction. In this paper, we propose an integrated VM power model called iMeter, which achieves high portability and VM level accuracy. Our technique uses only abstracted kernel-based performance events that provide accurate performance statistics as well as high portability across heterogeneous platforms to build the VM power model. Principal component analysis is applied to identify independent performance events that show strong impact on the VM power consumption with mathematical confidence. We also present a brief interpretation of the first two selected principal components on their indications of VM power consumption. We utilize the support vector regression to build the VM power model predicting the power consumption of both a single VM and multiple consolidated VMs running extensive workloads. We demonstrate that our approach is highly portable across heterogeneous platforms while providing precise predictions of the instantaneous VM power usage with an average error less than 2.7% against the actual power measurement.