Automated Techniques for Enabling Efficient MPI Application Migration
Department of Computer Science, University of Virginia, Charlottesville, VA
12th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid 2012), 2012
@article{sarnowska2012automated,
title={Automated Techniques for Enabling Efficient MPI Application Migration},
author={Sarnowska-Upton, K. and Grimshaw, A.},
year={2012}
}
Applications that use the MPI standard have additional dependencies related to the MPI implementation. When migrating an MPI code to a new computing site, the binary will not run if these dependencies are not resolved by properly configuring the new site. In this work, we present techniques that automatically resolve dependencies before runtime and enable execution without recompilation. Instead of depending on manual user intervention to configure a new site, we present techniques that automatically identify dependencies and automatically resolve them at new sites when possible. Our work aims to enable users to get their codes running more efficiently in new environments. We show the effectiveness of our methods at enabling MPI applications from the NAS Parallel Benchmarks suite to execute across various computing sites without recompilation.
January 20, 2012 by hgpu
