Automated Characterization of Parallel Application Communication Patterns

Overview

A concise description of an application’s communication pattern is useful to support activities such as hardware/software co-design, application debugging, and performance optimization.  Traditional methods of recognizing communication patterns require a high degree of expertise.  We have developed an automated approach for recognizing and parameterizing communication patterns in MPI applications.  We use AChaxP, an MPI profiling tool to collect information about an application’s communication topology and message volume, and then our AChax tool uses a search-based analysis to compare the application’s observed communication pattern against a library of common communication patterns to identify the combination that best matches the observed behavior.

Participants

  • Philip C. Roth, Oak Ridge National Laboratory

Sponsor

DOE Office of Science Advanced Scientific Computing Research office through the SUPER Institute of the Scientific Discovery through Advanced Computing (SciDAC) program

Publications

P. C. Roth, J.S. Meredith, J.S. Vetter, “Automated Characterization of Parallel Application Communication Patterns,” Proceedings of the 24th International Symposium on High-Performance Parallel and Distributed Computing (HPDC’15), Portland, Oregon, USA, 2015, pp.73-84.