Cluster-Based Parallelization of Simulations on Dynamically Adaptive Grids and Dynamic Resource Management
Martin A. Schreiber · 2014
The first part of this thesis is on the efficient parallelization of simulations using a dynamically adaptive mesh. We developed a new cluster-based parallelization approach, yielding a high scalability on shared- as well as distributed-memory systems. The second part is on varying resource requirements of concurrently executed simulations. We extended two HPC shared-memory parallelization models to optimize the distribution of computing resources based on application-specific information. Our studies show efficiency improvements of up to 45% for typical simulation scenarios.