Saving energy in data center infrastructures

Abstract

At present, data centers consume a considerable percentage of the worldwide produced electrical energy, equivalent to the electrical production of 26 nuclear power plants, and such energy demand is growing at fast pace due to the ever increasing data volumes to be processed, stored and accessed every day in the modern grid and cloud infrastructures. Such energy consumption growth scenario is clearly not sustainable and it is necessary to limit the data center power budget by controlling the absorbed energy while keeping the desired level of service. In this paper, we describe Energy Farm, a data center energy manager that exploits load fluctuations to save as much energy as possible while satisfying quality of service requirements. Energy Farm achieves energy savings by aggregating traffic during low load periods and temporary turning off a subset of computing resources. Energy Farm respects the logical and physical dependencies of the interconnected devices in the data center and performs automatic shut down even in emergency cases such as temperature peaks and power leakages. Results show that high resource utilization efficiency is possible in data center infrastructures and that huge savings in terms of energy (MWh), emissions (tons of CO2) and costs (kε) are achievable.