Towards Continuous Policy-driven Demand Response in Data.pdfVIP

Towards Continuous Policy-driven Demand Response in Data Centers David Irwin, Navin Sharma, and Prashant Shenoy University of Massachusetts, Amherst {irwin,nksharma,shenoy}@ ABSTRACT of all U.S. energy consumption specifically to data centers [11], Demand response (DR) is a technique for balancing electricity sup- with consumption estimated to double every 5 years [1]. Since ply and demand by regulating power consumption instead of gen- over 83% of electrical energy in the U.S. derives from burning eration. DR is a key technology for emerging smart electric grids dirty fossil fuels, this rising energy demand has serious environ- that aim to increase grid efficiency, while incorporating significant mental ramifications [17]. Additionally, since the growth in soci- amounts of clean renewable energy sources. In today’s grid, DR ety’s energy demand is beginning to outpace its ability to locate, is a rare event that only occurs when actual peak demands exceed extract, and burn fossil fuels, energy prices are on a long-term up- the expected peak. In contrast, smart electric grids incentivize con- ward trend [12]. Even using the “cheap” power available today, the sumers to engage in continuous policy-driven DR to 1) optimize energy-related costs of data centers already represent a significant power consumption for time-of-use pricing and 2) deal with power fraction (~31% [10]) of their total cost of ownership. Thus, design- variations from non-dispatchable renewable energy sources. While ing techniques to reduce the financial and environmental imp