ENERGY EFFICIENT THERMAL MANAGEMENT OF DATA CENTERS VIA OPEN MULTI-SCALE DESIGN:A REVIEW OF RESEARCH QUESTIONS AND APPROACHES

摘要

Currently, the airflow and heat transfer characteristics of data centers are simulated through computational fluid dynamics and heat transfer (CFD/HT) modeling. The degrees of freedom of these models are too large to be efficiently used with methodologies for effective design around the significant parameters of interest. Effective modeling and design of a data center is complicated due to its multi-scale nature, where length scales span four orders of magnitude from the room and plenum to the rack, to the server, and to the chip levels. Multi-scale resolution in simulation and design methodologies is important because the system design objectives (e.g., keeping the chip temperature in a desirable range) and the heat generation sources are located at the chip scale, yet the design solution (cold air from cooling units) spans the plenum and room scales. Using a complete CFD/HT model involving all of these length scales is infeasible for design. This paper begins with a review of the recent studies on energy usage trends and the need for energy efficient thermal management of data centers. Reduced order modeling of the thermal/fluid phenomena, by coupling the transport at different scales of the data center, is reviewed as a method that can be easily incorporated in design methodologies to find the optimum design variables involved at all scales. The energy-efficient design of data centers is a multi-objective problem. This design problem and possible multi-objective methodologies are reviewed. Incorporation of the coupled multi-scale model within established multi-objective design frameworks potentially allows for energy efficient open or scalable design of new facilities as well as retro-fits of current ones. These concepts are illustrated with selected examples.