Design and modeling of non-blocking checkpoint system
Kento Sato, Todd Gamblin, Adam T. Moody, Bronis R. de Supinski, Kathryn Mohror, Naoya Maruyama · IEEE International Conference on High Performance Computing, Data, and Analytics · 2012
In purpose to minimize the overhead of checkpointing, we propose a non-blocking checkpointing system, based on writing checkpoints to the parallel file system in the background. Our preliminary evaluation shows that our non-blocking checkpointing system is highly efficient. It incurs less than 1% overhead for CPU-bound applications.