CSR:Small:A Server-Network Cooperative Approach to Data Center Energy Optimization

George Washington University, Washington DC

Investigators

Abstract

Fueled by the growth in cloud computing which serves critical application needs ranging from banking to national defense, data centers have become a critical part of our nation?s infrastructure . Within data centers, servers and the interconnect networks account for an increasingly significant portion of the overall energy. This is because many server farms are configured to operate at a capacity much higher than necessary, and the data movement between servers is rapidly increasing in volume. This project studies the interplay between the servers and networks, and devises energy-saving strategies through monitoring data center workload patterns. With the changing landscape of data center applications and their communication patterns, this research project offers a much-needed study through adopting a holistic view of energy consumed by both servers and networks. The research outcomes will provide direct benefit to US industry through minimizing data center energy. This project investigates coordinated server-network energy optimization strategies that make judicious use of low-power states available in the processor and network component (micro) architectures while preserving application performance constraints. The solution approach seeks to design workload-specific energy management policies that select an optimal set of low power states and dynamic voltage/frequency settings that are best suited for the workload. To improve power efficiency, power shifting between heterogeneous components will be leveraged such that higher performance can be achieved within a specified power budget. The outcome of this research will include self-adaptive algorithms that will adjust themselves to the changing workload patterns, and techniques that minimize time-to-solution.