The International Conference for High Performance Computing, Networking, Storage and Analysis
Adaptive Power Efficiency: Runtime System Approach with Hardware Support.
Student: Ehsan Totoni (University of Illinois)
Supervisor: Laxmikant V. Kale (University of Illinois at Urbana-Champaign)
Abstract: Power efficiency is an important challenge for the HPC community and demands major innovations. We use extensive application-centric analysis of different architectures to design automatic adaptive runtime system (RTS) techniques that save significant power. These techniques exploit common application patterns and only need minor hardware support. The application's pattern is recognized using formal language theory to predict its future and adapt the hardware appropriately.
I will discuss why some system components such as caches and network links consume extensive power disproportionately for common HPC applications, and how a large fraction of power consumed in caches and networks can be saved using our approach automatically. In these cases, the hardware support the RTS needs is the ability to turn off ways of set-associative caches and network links.
I will also give an overview of an RTS approach for handling process variation.