Computer Experiment Modeling for an Energy-Efficient Data Center

• Activity Number: 501
• Type: Contributed
• Date/Time: Wednesday, August 12, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): Yufan Liu* and Ying Hung
• Companies: and Rutgers University
• Keywords: Asymmetric Laplace Process ; Computer Experiment ; Gaussian Process Model ; Kriging ; Model selection ; Quantile regression

Abstract

Data centers are widely used to provide services or management for various data processing applications. An important and challenging issue for a data center is to specify an efficient cooling mechanism. We propose an unified framework to tackle this problem, including a new parametric quantile regression model for the analysis of computer experiments and a variable selection procedure. This framework provides estimation and prediction for different quantiles and simultaneously identifies different sets of factors that have significant effects on those quantiles. As a result, an energy efficient data center can be constructed by optimizing the settings of these factors so that the lower quantile temperature is maximized to reduce unnecessary energy consumption. Meanwhile, the higher quantile temperature is maintained below certain value for reliable system operation. Asymptotic properties of the proposed methods are developed and the finite sample performance is demonstrated by simulations. The framework is successfully applied to a thermal management study in a real data center based on computer experiments built on computational fluid dynamic simulations.