The execution of state-of-the-art simulation software on supercomputers is increasingly affected by unpredictable execution times of work. Further, it becomes vulnerable to system faults. In this thesis, I develop reactive techniques for task-based simulation applications to tackle these challenges. The presented techniques do not only mitigate load imbalances effectively and thus improve performance significantly, but they also render simulation applications more resilient against system faults.
«
The execution of state-of-the-art simulation software on supercomputers is increasingly affected by unpredictable execution times of work. Further, it becomes vulnerable to system faults. In this thesis, I develop reactive techniques for task-based simulation applications to tackle these challenges. The presented techniques do not only mitigate load imbalances effectively and thus improve performance significantly, but they also render simulation applications more resilient against system faults...
»