ABSTRACT
The high-performance computing landscape is evolving rapidly on the way toward exascale computing. Changes mandated by alternative methods for either observing or controlling power distribution will be confined to the Application-Specific Energy Template (ASET). Incorporating power optimizations into applications would require developers to be additionally concerned with possibly arcane architectural features of each potential execution platform. This poses a severe limitation to power portability, and hence the ability of application software to utilize the available power-saving features across a range of platforms. In a power-constrained system, an ASET with global information can also assist when a system is power constrained, that is, when the full system cannot draw a given amount of power at a specific time. While architectural approaches to energy optimization have been and continue to be explored, further energy savings can potentially be made available by incorporating application information into intelligent runtime decision-making. Performance optimization within power budget constraints.
