Energy-Efficient Adaptive Computing for Data Centers

 

Temperature and power are two of the main challenges in computing. Powser is a major constraint in scaling mobile computing and hot spot temperature is a major constraint on scaling server systems depolyed in cloud computing. Unfortunately performance, power and temperature are conflicting goals, and thus, they need to be jointly co-optimized and managed during the design and runtime operation of processors. Such management will get more critical in future computational machines, which will rely on many-core processors for their number crunching. Many-core machines will have numerous degrees of freedom that can control their physical behavior, and it will not be possible to scale and leverage their full potential unless these degrees of freedom are carefully managed.

At SCALE lab, we work a number of projects to devise novel physical management techniques during runtime. In the process of working on our techniques, we follow rigorous scientific methods that lead to fundamental new understandings of the physical limits of processor management system. For our experimental verifications, we use state-of-the-art equipment, including power meters, and infra-cameras that are capable of monitoring the temporal and spatial temperatures of silicon die to a resolution of 5 microns. Check the videos displayed below which are captured using our infrared camera.

Current People Involved:

- Sherief Reda
- Xin Zhan

Related Publications