Sensing Data Centers for Energy Efficiency

Data centers are large energy consumers today and their consumption is expected to increase further, driven by the growth in cloud services. The large costs and the environmental impact of this consumption have motivated data center operators to optimize data center operations. We argue that one of the underlying reasons for the low energy utilization is the lack of visibility into a data center’s highly dynamic operating conditions. Wireless sensor networks promise to remove this veil of uncertainty by delivering large volumes of data collected at high spatial and temporal fidelities. The paper summarizes data center operations in order to describe the parameters that a data center sensing network would need to collect and motivate the challenges that such a network would face. We present technical approaches for the problems of data collection and management and close with an overview of Data Center Genome, an end-to-end data center sensing system.