Large scale simulation of multiphase field problems
Multiphase field simulations often require long computation times as well as large amounts of memory to store a multitude of phase fields with different properties. The application of distributed memory parallelism on multiphase field solvers allows simulations of significantly larger scale and accuracy as the computational power and total amount of usable memory is increased. Therefore OpenPhase is expanded by a framework for MPI parallelization.
The concept used here is a distribution of the computational domain into blocks, that are assigned to processors and communicate overlapping data at certain synchronization points. As the primary portion of computational work is located at the interface between phase fields the costs of calculations can differ locally. To achieve the desired efficiency a graph partitioning method, that minimizes the amount of communication between processors, is currently used to provide dynamic load-balancing.
In the future parallel-in-time methods will be implemented and evaluated.