There is a strong interest in predicting and managing power usage for HPC data centers for many reasons (e.g., more efficient cooling, avoiding high power usage during peak demand periods). This requires a cross-disciplinary approach that addresses different abstraction layers, from the infrastructure, to hardware, to system software and up to applications. The advances at the hardware layer need to be followed by evolution on systems software and middleware in order to provide efficient results. Various techniques have achieved a balance of application performance and power or energy consumption using technologies like Dynamic Voltage and Frequency Scaling (DVFS). There is also research on using resource management and scheduling tools to monitor energy and power consumption and adjust it through coordinated actions. Yet another area of research is investigating run-time operating systems for managing system energy and power. These research efforts are making their way into production environments with implementations at a few leading-edge supercomputing centers.
Lessons learned and best practices are being gained in energy aware resource management and job scheduling software that relies on power monitoring hardware. The Energy Efficient HPC Working Group (EE HPC WG) has created this project to focus on power and energy management with resource management and scheduling tools. The initial deliverable of the team will be a whitepaper that surveys those sites already deployed and/or are exploring the use of this software for HPC power and energy management.