CAREER: Energy Prediction and Conservation for Server Clusters

Rutgers University New Brunswick, New Brunswick NJ

Investigators

Abstract

Previous power and energy research has focused almost exculsive on battery-operated devices. In contrast, this research is focused is focused on power and energy prediction and conservation for clusters of servers. The main goals of the research are: (1) to develop a methodology and a tool that combine analytical modeling, experimentation, and optimization algorithms to predict the power/energy conservation techniques server clusters; (2) to develop power/energy conservation techniques that can use the prediction infrastructure to optimize consumption for both curren and future server cluster without excessive performance degradingl and (3) to design, implement, and evaluate cluster systems that exploit these techniques. This research will allow server cluster to be engineered to a specific and low consumption. Furthermore, it allow conservation techniques to make fundamental tradeoff decisions about power, energy, and performance. These developments will have important implications, given that the computational model upon which the Internet is oragnized relies on large server clusters.