Thermal Mass Characterization of a Server at Different Fan Speeds Mahmoud Ibrahim, Saurabh Shrivastava, Bahgat Sammakia, and Kanad Ghose Panduit Corporation 6200 W. 175 th Street Tinley Park, IL 60477 Phone: (708) 532-1800 Fax: (708) 614-8345 Email: mibrahi1@binghamton.edu ABSTRACT Dynamic cooling has been proposed as one approach for enhancing the energy efficiency of data center facilities. It involves using sensors to closely monitor the data center environment with time and making real time decisions on how to allocate the cooling resources based on the location of hotspots and concentration of workloads. In order to effectively implement this approach, it is good to know the transient thermal response of the various systems comprising the data center must be determined, which is a function of thermal mass. Not only is thermal mass important in dynamic cooling, it also plays a major role in the temperature rise of a data center during power failure. A previous study concentrated on characterizing the thermal mass of a 2 RU server by running the server at different powers and a fixed fan speed. The fixed fan speed corresponds to one specific heat transfer coefficient value. This study is a continuation to the previous work, where the server fan speed is varied to deduce the change in heat transfer coefficient at different airflow rates. As expected, the heat transfer coefficient increases as the server airflow rate increases. The average thermal mass value obtained for the 2 RU server in this study was 12 kJ/K. A method of adopting the compact model developed in this study into a Computational Fluid Dynamics (CFD) code is proposed to cut down on the computational time of transient analysis. KEY WORDS: Data centers, dynamic, transient, thermal mass, compact models, heat transfer coefficient. INTRODUCTION Data centers are mission critical facilities that house a large number of electronic equipment, typically servers, switches, and routers which require a tightly controlled environment to ensure their safe operation. The importance of these electronic equipment to any business application sometimes drives data center operators to unnecessarily overcool their data centers fearing the presence of hotspots in any region of the facility. This explains why in the 2007 EPA report to Congress it was estimated that U.S. data centers used 61 billion kWh of electricity in 2006, representing 1.5% of all U.S. electricity consumption and double the amount consumed in 2000 [1]. This has driven the data center community to find ways in reducing the energy consumption of data centers [2- 7]. Knowing how highly dynamic the data center environment is, many are considering the option of dynamic cooling to maximize the energy efficiency of the cooling infrastructure supporting a data center [8-11]. This requires close monitoring of the facility environment with time using sensors, and making real time decisions on how to allocate the cooling resources based on the location of hotspots and concentration of workloads. The issue with these mission critical facilities is not only maintaining a suitable steady environment with maximized energy efficiency, but also the security of electrical power supply to the electronic equipment is highly crucial. A survey conducted by Ponemon Institute on data center outages [12] stated that 88% out of 453 surveyed data center operators experienced a loss of primary utility power and hence the power outage of their data centers, with an average of 5.12 complete data center outages every two years. Unfortunately, data centers cannot afford shutdown due to the increase in inlet temperatures associated with the cooling equipment failure which in turn poses the shutdown of the electronic equipment that are the core of the customer-business confidence. A number of studies have addressed this issue and developed simple energy based models to predict the data center room temperature rise during power failure [13]. Khankari [14] looked at the effect of rack weights, number of rack rows, number of racks, and total heat load on the room temperature rise during power failure. Zhang and VanGilder [15] discussed different design strategies for power backup and whether having some of the cooling equipment on UPS (Uninterruptable Power Systems) is useful. While such models may provide useful information, they are based on combining every parameter of the data center into one, giving just an overall sense of the response. These models however are not capable of providing a detailed view of the data center during power failure, and which regions are most critical. The availability of tools such as Computational Fluid Dynamics (CFD) that can accurately analyze, in transient, the different dynamic control methods and look at different failure scenarios in data centers can be very useful. However, the downside to conducting transient analysis in CFD is the computational time, and the lack of knowledge in modeling the equipment inside the data center to accurately account for the thermal mass effects which were shown to strongly affect the transient analysis [14, 16]. There were a few efforts looking at transient analysis of data centers such as the work by Stahl and Belady [17], where they collected empirical data at different heat loads to compare with theoretical calculations. Beitelmal et al. [18] performed transient simulations to study the impact of CRAC failure on the temperature variations 978-1-4244-9532-0/12/$31.00 ©2012 IEEE 457 13th IEEE ITHERM Conference