While technology is delivering increasingly sophisticated and powerful chip designs, it is also imposing alarmingly high energy requirements on the chips. One way to address this problem is to manage the energy dynamically using adaptive Low-Power Techniques (LPTs) in the processor. The first part of this thesis presents the design and evaluation of the first energy-management framework that deals with multiple generic LPTs and that tackles both energy efficiency and temperature control in a unified manner. We call this general approach Dynamic Energy Efficiency and Temperature Management (DEETM). The goal of the framework is two-fold: maximize energy savings without extending application execution time beyond a given tolerable limit, and guarantee that the temperature remains below a given limit while minimizing any resulting slowdown. The framework successfully meets these goals. For example, it delivers a 40% energy reduction with only a 10% slowdown.; The second part of the thesis discusses how to further improve the algorithm for energy efficiency. We observe that applications change their demands on the hardware as they execute. This suggests that certain hardware adaptations can be made to save energy at little performance cost. In the context of a general purpose system with multiple LPTs, we address the twin problems of when to adapt, and what LPT to use to adapt. We demonstrate that rather than adapting the processor at time intervals, it is better to do it at the grain of subroutines: the reaction of a subroutine to system adaptation is usually highly predictable, and the best adaptation for a subroutine can be easily remembered and reused later. Using this insight, we design architectural support for an adaptive processor to decide what specific LPTs to activate and when to activate them. Targeting different environments, we propose a framework of different schemes to exploit an adaptive processor, where these decisions are made off- or on-line. Overall, the schemes perform better than an approach based on fixed-time intervals similar to the original DEETM: in a system with three LPTs, energy savings increase by 40–63% with less performance degradation.
展开▼
