Energy-Efficient Computing: Datacenters, Mobile Devices, and Mobile Clouds

摘要

Energy consumption is a key design driver for electronic systems ranging from warehouse-size datacenters to battery-powered mobile devices to mobile clouds. It is well known that energy efficiency is best achieved by an application-specific mix of power-efficient hardware and runtime energy governance. Power efficient hardware requires low power devices, cell libraries, circuits, and architectures whereas effective energy governance needs significant hardware and software support e.g., to achieve dynamic power/performance scaling, power gating, core consolidation, and computation offloading. In my talk I will discuss three example problems to illustrate the range of low power solutions that can be employed and the kind of power savings which are achievable. These problems are: (i) Power-efficient resource management and job scheduling in a geo-distributed cloud infrastructure, (ii) Design of low-power application processors exploiting the temperature effect inversion of deeply scaled devices, and (iii) Energy-efficient computation offloading for deep neural networks in a mobile cloud computing environment. I will conclude my talk with a list of best power-efficient design practices.