From cTuning.org
BOF: Collaboratively mining rich information to prepare the Exascale challenges
Contents |
Place
Organizers
- Grigori Fursin, INRIA, France
- Marie-Christine Sawley, Intel, France
Abstract
In this BOF we want to discuss novel ways to overcome the complexity of Exascale computing systems through community-driven auto-tuners and optimization repositories. Modern computing systems have reached unprecedented levels of complexity due to a large number of available design and optimization choices. Various auto-tuning and machine learning techniques have been proposed recently to effectively tackle this complexity. However, current architectures, applications and tools are not yet well adapted to such automatic tuning. Furthermore, data mining techniques require collection of a massive amount of data that is not trivial to process. Therefore, we are developing collaborative repositories and techniques that can transparently distribute characterization and optimization of applications and architectures among multiple users. We believe that such collaborative approaches will be vital for developing efficient Exascale computing systems and software. With this BOF, we would like to share our experience with some existing frameworks and how to improve them.
Presentation
This presentation summarizes research and development of the Application Characterization and Optimization group created and led by Grigori Fursin at Intel/CEA Exascale Lab in France during his industrial leave from INRIA between 2010.03 and 2011.09 based on his cTuning concept and research experience.
We would like to thank Jean Christophe Beyler for presenting this work since Grigori Fursin could not attend this event due to family reasons.