Inverse distance weighting and kriging spatial interpolation for data center thermal monitoring

Abstract

Studies have shown that data center performance is also influenced by its environmental conditions, one of them is thermal state. If thermal information inside a data center, such as temperature and humidity are not well-monitored, then the data center might experience overheat or overcool state, resulting in downtimes or other performance issues. However, for more accurate thermal monitoring, it is better to collect temperature readings from many sensors, which is not really feasible in reality. Spatial interpolation methods have been adopted as a means to predict spatial information at certain locations without adding sensors continuously. However, most of the previous works utilize spatial interpolation concept inside an environment with medium to large scale sensor nodes. In this study, we first customize a model of real server room with 5 temperature and humidity sensors in order for it to be fitted in spatial interpolation concepts. We then apply and evaluate two most commonly used spatial interpolators, i.e. Inverse Distance Weighting (IDW) and Kriging with regard to our customized server room model. Our results from 30 measurements followed by significance test demonstrate that IDW gives higher accuracy than Kriging when it is implemented inside an environment involving small-scale sensors.