Molecular-Dynamics-Simulations are computationally highly demanding but can produce very deep insights in Chemistry or particle physics. In order to perform simulations more efficiently, a vast number of strategies and optimizations are available. Because of this large selection, the simulation must automatically be adjusted during run-time. This process of testing and selecting options is called auto-tuning. AutoPas is a C++ particle simulation library, which provides this functionality for simulations implemented by the user. It is possible to parallelize this tuning process by sharing the workload with other processes. However, this was previously only possible if the simulated scenarios were homogeneous. This thesis extends this parallelization scheme in order to use it for inhomogeneous scenarios. The communication is based on MPI. The new implementation is compared with the old one and also with not using parallel tuning. Eventually, significant performance improvements can be achieved, but only at the cost of more manual configuration. This could be solved in the future by also automating the selection of the new configuration parameters.

Keywords: MPI, auto-tuning, AutoPas