The Multicore Challenge: Petascale DNS of a Spatially-Developing Supersonic Turbulent Boundary Layer Up to High Reynolds Numbers Using DGSEM

Abstract

With increasing computational power on modern supercomputing systems , direct numerical simulations (DNS) have been gaining in importance for the investigation of wall-bounded turbulent flows. The applied numerical method, however, has to enable an efficient usage of high performance computing systems to satisfy the involved computational costs. In this context, discontinuous Galerkin (DG) methods have become a promising candidate to conduct DNS in an efficient way as they offer an excellent scaling combined with arbitrary high spatial accuracy in complex geometries. On the other hand, the DG method has been also suffering from being considered as inefficient and slow within the computational turbulence community, which doubted the suitability of the DG method when applied to turbulent flows. In this work, we performed a DNS of a compressible spatially-developing supersonic flat plate turbulent boundary layer up to Re  D 3878 using the discontinuous Galerkin spectral element method (DGSEM). To our knowledge, the present simulation is currently the biggest computation within the DG community and enabled us to generate a reliable high-fidelity database for further complex studies. The usage of the DGSEM approach allowed an efficient exploitation of the whole computational power available on the HLRS Cray XC40 supercomputer and to run the simulation with a near-perfect scaling up to 93,840 processors. The obtained results demonstrate the strong potential of the DGSEM at conducting sustainable and efficient DNS of high Reynolds number wall-bounded turbulent flows.