Kurzfassung:
Future exascale computers will offer unprecedented performance gains, but their increased complexity introduces new obstacles. System faults will likely affect parallel simulations on a regular basis, so applications should be able to react accordingly. In this thesis, we show how to make a solver for high-dimensional PDEs aware of different types of faults, using primarily the properties of the algorithm. We argue that this numerics-based approach to fault tolerance will be key at exascale.