Investigation of Parallel-In-Time Algorithms for Fluid Flow Problems of Varying Complexity : From Burger's Equation to an Operational Ocean Circulation Model

In this thesis an application of the time-parallel algorithm Parareal to the ocean-circulation model FESOM2 is described, and which challenges have to be expected during the development process. The concept of parallelization in time allows for runtime reduction in the numerical approximation of initial value problems by decomposing the temporal domain into sub-intervals, on which the problem can be solved concurrently. If the problem at hand is governed by partial differential equations, as it is in FESOM2, then Parareal offers its greatest potential for further runtime reduction, when the speedup by classical domain decomposition techniques is saturated. In general, climate models are executed at this point of saturation in space-parallelism and still, they require notable computational effort in order to evolve the problem over decades up to centuries of simulation time. With the capacities of modern high performance clusters not being exhausted, an usecase for parallel-in-time methods like Parareal to further reduce walltimes is more than welcome. Parareal, being an black-box algorithm that is non-intrusive to model routines, therefore appears to be an ideal choice to be utilized for problems of high complexity, like in climate research. The Parareal algorithm has not been applied so far to sophisticated climate models, which are employed for state-of-the-art climate research. The dissertation aims to answer the question whether Parareal can achieve the anticipated runtime reduction of FESOM2 simulations.

Rights

Use and reproduction:


CC BY-NC-SA 4.0

Please note that individual components of the publication may be subject to other licensing or copyright conditions.

Cite

Citation style:
Could not load citation form.