Adjusting global data dynamically to available resources remains a frequent issue in
HPC applications. This task is particularly challenging with legacy MPI code. Developers
require an in-depth understanding of the code base and must write applicationspecific
code to enable dynamic adaptation. To tackle this challenge, the authors of
LAIK, a library for fault-tolerant distribution of global data for parallel applications,
have worked on developing an efficient solution. LAIK streamlines the process by
offering a solitary function to adjust the number of participating processes. In this
work, we ported the High Performance Conjugate Gradient Benchmark to LAIK to
enable dynamic global data adaptation. Furthermore, we used the abstract concepts of
LAIK for interprocess communication. This was accomplished incrementally. There
were many factors to consider, but ultimately achievable. Our final version yielded
comparable results to the original application. To complete this work, we evaluated
computing performance, memory consumption and the effectiveness of dynamically
adapting resources.
«
Adjusting global data dynamically to available resources remains a frequent issue in
HPC applications. This task is particularly challenging with legacy MPI code. Developers
require an in-depth understanding of the code base and must write applicationspecific
code to enable dynamic adaptation. To tackle this challenge, the authors of
LAIK, a library for fault-tolerant distribution of global data for parallel applications,
have worked on developing an efficient solution. LAIK streamlines the...
»