Reduced-order DC models

Before an air-cooled data center (DC) is put into operation, the air flows and temperature distribution in the DC room have to be optimized to reduce the risk of overheating servers and to minimize the energy consumption of the cooling system. This can be achieved by using IBM's mobile measurement technology (MMT) and supporting CFD simulations. However, the performed optimization is valid only for the thermal conditions in the DC at the time of measurement. Owing to variations of environmental conditions, dynamic changes of the server's workload, and upgrading of computing IT equipment, the air flow and thermal conditions in the DC will change over time. To keep the thermal conditions close to the optimal operating point, advanced cooling-control concepts, thermally aware workload distribution techniques and upgrade strategies have to be applied.

We are developing and evaluating various reduced-order models that can characterize and predict the thermal behavior of a data center from temperature information provided by a few sensors deployed at judiciously chosen locations in the DC room. We derived a simple model that is based on the energy balance of heat flows in the DC, which describes the air recirculation between servers in the DC room as cross interference. The parameters of the model can be identified by performing MMT measurement campaigns in the DC under investigation and CFD simulations. Using this model, we have developed a control strategy that keeps the system close to the optimal operating point by adjusting the air flow rates through the racks and allocating incoming jobs to individual servers based on the sensed temperature information. Moreover, we also applied the model to determine the placement of the upgrading computing equipment in the racks of the DC so that the thermal impact on the existing cooling environment is minimized. The validation of these concepts in field tests and their implementation in a production DC are research challenges that we will address in the near future.

1. T. Scherer, "Modeling and Control for Energy Efficient Data Centers," Master Thesis, ETH Zurich, Zurich, Switzerland, August 2009.
2. N. Vasic, T. Scherer, and W. Schott, "Thermal-Aware Workload Scheduling for Energy-Efficient Data Centers," in Proc. of 7th IEEE/ACM Conference on Autonomic Computing and Communications (ICAC 2010), Washington DC, US, June 2010.
3. P. Biller, P. Chevillat, F. de Lorenzi, T. Scherer, W. Schott, R. Ullmann, and C. Vömel, "Efficient Cooling of Data Centers," in Proc. of World Engineers' Convention 2011, Geneva, September 2011.
4. J. Siriwardana, S.K. Halgamuge, T. Scherer, W. Schott, "Minimizing the Thermal Impact of Computing Equipment Upgrades in Data Centers," submitted to Energy and Buildings Journal, Elsevier, November 2011.