Study on data center server management system for using server power estimation technology

The increasing power consumption in servers is a problem in data centers. One approach to address this is to optimize physical resource allocation for virtualized functions. For this approach, power consumption needs to be estimated, and many papers have proposed models to estimate server power consumption. However, applying such estimation techniques to data center servers presents an operational challenge. Specifically, it is difficult to reduce the cost of model building and data collection while maintaining estimation accuracy. To address this problem, we propose a management system with a modeling function and a data selection function. The modeling function builds accurate models in accordance with application requirements, and the data selection function selects data related to server power consumption and then instructs data collection to each processing server. Our main effort in this paper was to address the monitoring function issue to reduce server power consumption during data collection. We analyze the relationship between server performance data and server power consumption and reduce the type of data. Experimental results showed that our selected data are sufficient to construct a server power modeling. Furthermore, by limiting the amount of data collected, we were able to reduce server power consumption by about 3 W and monitoring overhead by about 30%.