The objective of this work is to complement ORNL's xSim simulator with benchmark generation capabilities. ScalaBenchGen from NCSU will be extended to auto-generate source code suitable for evaluation under xSim. ScalaBenchGen will complement these capabilities with the ability to extract communication benchmark skeletons from actual HPC runs of applications. These skeletons include timings for computational parts and actual MPI communication calls. We propose to combine the ScalaBenchGen and xSim capabilities for sample HPC benchmarks/ applications. Timings for the computational part will be enhanced to allow adaptation with respect to future (exascale) architectures. This co-design exploration supports the research path toward exascale.
Publications: