Abstract: Xolotl is a cluster dynamics simulator used to predict gas bubble evolution in solids. It is currently being used to simulate bubble formation in the plasma-facing surface within fusion reactors and the nuclear fuel used in fission reactors. After observing performance problems in coupled-code simulations of fusion reactors, we used Xolotl's built-in performance data collection infrastructure and an external profiling tool to identify inefficiencies when writing Xolotl's two types of checkpoint files. We changed the code to use true parallel writes via the HDF5 data management library, resulting in a code that is approximately 57x faster when writing the program's main checkpoint file at the scale used in the coupled-code simulations, and that exhibits less performance variability due to external system activity. We also identified and addressed a memory usage problem that reduced Xolotl peak memory usage by approximately 88% per compute node.
Best Poster Finalist (BP): no
Poster summary: PDF
Reproducibility Description Appendix: PDF
Back to Poster Archive Listing