Controlling the Memory Subscription of Distributed Applications with a Task-Based Runtime System

Marc Sergent (1, 2, 3) , David Goudin (4) , Samuel Thibault (1) , Olivier Aumage (1, 2)

1 STORM - STatic Optimizations, Runtime Methods
2 LaBRI - Laboratoire Bordelais de Recherche en Informatique
3 UB - Université de Bordeaux
4 CESTA - Centre d'études scientifiques et techniques d'Aquitaine

This talk discuss the need for cooperation between a task-based distributed application code and a run-time system engine to control the memory subscription levels throughout the execution. We show that the task paradigm allows to control the memory footprint of the application by throttling the task submission flow rate, striking a compromise between the performance benefits of anticipative task submission and the resulting memory consumption. We illustrate the benefits of our contribution on a compressed dense linear algebra distributed application.

Mots clés: memory control; task-based run-time systems; compressed linear algebra; distributed computing

Domaines: Informatique [cs]; Calcul parallèle, distribué et partagé [cs.DC]

Marc Sergent, David Goudin, Samuel Thibault, Olivier Aumage. Controlling the Memory Subscription of Distributed Applications with a Task-Based Runtime System. SIAM Conference on Parallel Processing for Scientific Computing (SIAM PP 2016), Apr 2016, Paris, France. pp.318 - 327. ⟨hal-01380126⟩