Massively parallel and fault-tolerant multigrid solvers on peta-scale systems

In this thesis, different aspects are considered that are necessary to make simulations on future exa-scale computing systems feasible. In this context, efficient and fault-tolerant multigrid solvers are studied on a peta-scale system to explore the current frontiers of solving large linear systems with more than ten trillion unknowns and are optimized with respect to high-performance computing. For these solvers, strategies are developed that recover in case of a process failure the lost data in the solver without overhead. In addition, these solvers serve for studying geophysical problems using real-world data.