EECS Publication
Unified Model for Assessing Checkpointing Protocols at Extreme-Scale
George Bosilca, Aurelien Bouteiller, Elisabeth Brunet, Franck Cappello, Jack Dongarra, Amina Guermouche, Thomas Herault, Yves Robert, Frederic Vivien, and Dounia Zaidouni
In this article, we present a unified model for several well-known checkpoint/restart protocols. The proposed model is generic enough to encompass both extremes of the checkpoint/restart space: on one side the coordinated checkpoint, and on the other extreme, a variety of uncoordinated checkpoint strategies (with message logging). We identify a set of parameters that are crucial to instantiate and compare the expected efficiency of the fault tolerant protocols, for a given application/platform pair. We then propose a detailed analysis of several scenarios, including some of the most powerful currently available HPC platforms, as well as anticipated Exascale designs. This comparison outlines the comparative behaviors of checkpoint strategies at scale, thereby providing insight that is hardly accessible to direct experimentation.
Published 2012-06-05 04:00:00 as ut-cs-12-697 (ID:9)