Performance and energy modeling for live migration of virtual machines
Proceedings of the 20th international symposium on High performance distributed computing
Performance and energy modeling for live migration of virtual machines
Cluster Computing
Hi-index | 0.00 |
Power consumption has become one of the most important design considerations for modern high density servers. To avoid system failures caused by power capacity overload or overheating, system-level power management is required. This kind of management needs to control power consumption precisely. Conventional solutions to this problem mostly rely on feedback controllers which only concern the power itself, known as black-box approaches. They may not respond to the variation of system quickly. This paper presents a gray-box strategy to design a model-predictive feedback controller based on a pre-built power model and a performance prediction model to constraint the peak power consumption of a server. In contrast to the existing strategies, this gray-box approach uses the performance events, which bring more insights of the behaviors and power consumption of a system, for the purpose of model prediction. We implemented a prototype of this controller and evaluated it using SPECweb2005 benchmark on a web server. This controller can settle the power consumption below the power cap within 2 control periods for more than 75\% of the power overloading regardless of workload variations, outperforming black-box approaches. Meanwhile, the performance of application can be maximized with this controller.