Towards Discovering Data Center Genome Using Sensor Nets

The IT industry is the fastest growing sector in US energy consumption. Improving data center energy efficiency is a pressing issue with significant economic and environmental consequences. Heat distribution is a key operational parameter that affects data center cooling and energy consumption. However, typical data centers lack effective finegrained sensing systems to monitor heat distribution at a large scale. In this paper, we motivate the use of sensor networks as a dense instrumentation technology to understand and control cooling in data centers. We present Microsoft Research Genomote sensors designed for data center monitoring, and the RACNet for reliable data acquisition. We describe lessons learned from early pilot deployments, and discuss architectural and technical challenges in developing data center sensor networks.