Resilience for Global Address Spaces

• funded by: LBNL under the DEGAS project
• funding level: $203,393
• duration: 9/24/2013 - 08/15/2016

The objective of this work is to provide functionality for the BLCR Linux module under a PGAS runtime system (within the DEGAS software stack) to support advanced fault-tolerant capabilities, which are of specific value in the context of large-scale computational science codes running on high-end clusters and, ultimately, exascale facilities. Our proposal is to develop and integrate into DEGAS a set of advanced techniques to reduce the checkpoint/restart (C/R) overhead.

Publications:

• "DINO: Divergent Node Cloning for Sustained Redundancy in HPC" by A. Rezaei, F. Mueller in Cluster, Sep 2015, pages 180-183.
• "Affinity-Aware Checkpoint Restart" by A. Saini, A. Rezaei, F. Mueller, P. Hargrove, E. Roman in Middleware, Dec 2014, pages 121-132.
• "DINO: Divergent Node Cloning for Sustained Redundancy in HPC" by A. Rezaei, F. Mueller in TR 2014-7, Dept. of Computer Science, North Carolina State University, Jun 2014.
• "Sustained Resilience via Live Process Cloning" by Arash Rezaei, Frank Mueller, Workshop on Dependable Parallel, Distributed and Network-Centric Systems, May 2013.

Theses:

• "Fault Resilience for Next Generation HPC Systems" by A. Rezaei, Ph.D. Thesis, North Carolina State University, Mar 2016 (last known position: Samsung, CA)
• "Affinity-Aware Checkpoint Restart" by Anjay Saini, M.S. Thesis, North Carolina State University, May 2014 (last known position: Intel, OR)