Surveillance dynamique des communications MPI au cours de l’exécution : guide d’utilisation scientifique et documentation technique
George Bosilca, Clément Foyer, Emmanuel Jeannot, Guillaume Mercier, Guillaume Papauré · HAL (Le Centre pour la Communication Scientifique Directe) · 2017
Understanding application communication patterns became increasingly relevant as the complexity and diversity of the underlying hardware along with elaborate network topologies are making the implementation of portable and efficient algorithms more challenging. Equipped with the knowledge of the communication patterns, external tools can predict and improve the performance of applications either by modifying the process placement or by changing the communication infrastructure parameters to refine the match between the application requirementsand the message passing library capabilities. This report presents the design and evaluation of a communication monitoring infrastructure developed in the Open MPI software stack and able to expose a dynamically configurable level of detail about the application communication patterns, accompanied by a user documentation and a technical report about the implementation details.