Exploiting Hierarchical Exascale Hardware using a PGAS Approach
Karl Fürlinger · 2015
Hardware architectures that enable Exascale-level performance are expected to break some long-held programmability assumptions on the node level and will come with a plethora of additional challenges that make the productive development of efficient applications difficult. One critical issue is data locality, which will become even more important than it is today. A shift towards data-centric programming models will be required to exploit the full potential of these machines. We present an overview of our work in progress on DASH, a data-structure oriented PGAS library implemented in C++, with which we attempt to address some of the challenges posed by upcoming hardware architectures by focusing on flexible data layout and by supporting a hierarchical locality model.