Smart Power

Welcome to the external home page of the SmartPower research group within HP Labs.

Energy and power are rapidly emerging as critical resources across all devices in the computing spectrum including high-end servers, consumer PCs, laptop computers, handheld devices, and in the future, even smaller ``wearable computers.'' While energy dissipation has historically been more of a concern to mobile system designers because of the impact on battery life (and consequently the utility), it has recently gained more importance in other classes of systems (e.g., PCs, servers and data centers) as well, because of its impact on system costs, cooling, reliability, and more recently, compliance with environmental standards.

Our research seeks to identify and evaluate new approaches to systems' and solutions' design to optimize power management. A few of our research results are summarized below. For more details, see our list of publications or contact partha.ranganathan AT hp.com.

Some Representative Projects

• Display power management: We propose the notion of energy-aware user interfaces and study the potential benefits from intra-display adaptivity. Our results show a factor of 2 to 20 reduction in the display power.
• Processor power management: We propose the notion of heterogeneous multi-core architectures and study the power benefits from adaptively choosing different cores for different application points. On average, this new architecture reduces 40% of the processor energy with very little loss in performance.
• Wireless power management: our work evaluates the benefits of adapting lower-level physical layer energy use to higher-level application requirements. For a typical 802.11b implementation, this can reduce up to 90% of the wireless power.
• Server and data center power management: Our work proposed temperature-aware scheduling, hardware-software co-ordination to enforce power budgets, and new blade designs for lower power. Our optimizations are individually successful in reducing system power by 20-50%.
• Energy-driven statistical profiling: We propose and evaluate a novel approach to profile the energy usage of applications on based on energy-driven statistical sampling. Our tool can help help designers reason about the energy impact of software design decisions. 
• Splice: Splice is "an extended knowledge plane" that provides a data center measurement and monitoring framework for low-cost integrated aggregation and analysis of data for both environmental and performance parameters.
• Energy-scaledown:  we provide a taxonomy of energy and power management work. Our energy scale-down framework discusses  optimizations that design and use adaptivity in hardware and software to match energy efficiency of the system with the desired user/workload functionality.