Numerical simulations of coupled computational fluid dynamics (CFD) and computational structural mechanics (CSM) simulations are an essential part of aircraft design development. Larger high-performance computing (HPC) systems enable more resolved and, in some cases, even scale resolving simulations providing much more meaningful results. The efficient usage of these HPC systems requires a good parallelization of the simulation software and an optimized partitioning of the mesh data, typically using graph partitioning libraries such as Zoltan or ParMETIS. However, for large meshes or core counts, current partitioning software may take too long to partition and distribute the data. The resulting domain decomposition is often not optimized for the system architecture either. In some cases, it is not even possible to compute a mesh partitioning. One way to improve on this problem is employing hierarchical partitioning, i.e., partitioning the data on multiple hierarchy levels with respect to the HPC system architecture. Such an approach uses less resources on each hierarchy level and considers the inter node communication time. This enables numerical flow simulations utilizing more resources efficiently and, hence, also more precise simulations than previously possible. We will present our implementation of the hierarchical partitioning approach and its effect on the runtimes of CFD simulations within DLR’s FlowSimulator framework when compared to using ParMETIS’ graph partitioning method. For this demonstration, we will use meshes of more than a billion cells and more than 65,000 cores.