Exascale systems will be extraordinarily complex, dynamic architectures with massive numbers of potentially heterogeneous components and deeper memory hierarchies, requiring ever more complex software to use them.
The goal of the PRIMA-X project is to develop an exascale-enabled performance measurement system for applications, intended to serve as a reference implementation for performance tools in future exascale software stacks, including but not limited to TAU and Scalasca. Main challenges to be addressed include extreme scalability, dynamism, and heterogeneity. Because exascale architectures and targeted technologies will be in flux during the period of the PRIMA-X project, an important part of our research will be on designing a measurement model that is sustainable across exascale system prototypes and their evolution to real platforms.
The project will leverage results of the projects PRIMA and SILC, namely the performance-measurement infrastructure Score-P, extending its design towards deployment at exascale.
- University of Oregon
Performance Research Lab
- Technische Universität Darmstadt
Laboratory for Parallel Programming
- Forschungszentrum Jülich
Jülich Supercomputing Centre