From chip to cooling tower data center modeling: Validation of a Multi-Scale Energy Management model

Abstract

Novel modeling techniques have become prominent tools in the design of data center facilities and optimization of their operating profiles. These tools provide a method for extensive analysis of data center facilities by enabling rapid assessment of numerous what-if scenarios. As the data center industry moves toward lower PUE, further efficiency gains can be found only in detailed understanding of the integrated IT-facility data center infrastructure. The reliance on rule-of-thumb guidelines for system operations are being superseded by more physics and empirically based models. The Multi-Scale Energy Management (MSEM) tool has hence been developed based on a compact modeling approach that maintains much of the physics of the cooling architecture. However, extensive validation of such models in the live data center environment is essential to deploy a reliable tool for data center energy optimization. Previous work by the authors in experimental assessment and validation of the MSEM model highlighted a number of lessons to be taken into further experimental validation phases. In this paper, a more rigorous validation of the Multi-Scale Energy Management model in the live data center is explored integrating both the IT and facilities infrastructure. The methodology for data center experimental analysis is presented with consideration of steady-state evaluation in a data center facility which employs an extensive uncertainty characterization of the data center's operating parameters. The experimental evaluation is verified against the compact models within the MSEM code. A chip to cooling tower validation of the MSEM model is then undertaken with this new empirical data from the test data center.