HPC Requirements of High-Fidelity Flow Simulations for Aerodynamic Applications
Axel Probst, Tobias Knopp, Cornelia Grabe, and Jens Jägersküpper
This paper relates the computational demand of turbulence-resolving ﬂow simulations for aircraft aerodynamics to the parallel scalability of the DLR ﬂow solvers TAU and THETA, as well as the new CODA solver optimized for many-core HPC systems. Based on existing lower-ﬁdelity simulations, the computational requirements for wall-resolved LES are ﬁrst estimated for single aircraft components at windtunnel scale. It is shown that such simulations at reduced Reynolds numbers would be realizable within days to weeks with the current methods, if the largest available HPC clusters with more than 100,000 cores were used. However, an extrapolation to the HPC requirements of LES for a full 3D wing of an aircraft at ﬂight Reynolds numbers highlights the urgent need for larger HPC resources and adapted parallel code designs, as well as more efficient numerical algorithms and physical models.
Data-adapted Parallel Merge Sort
Johannes Holke, Alexander Rüttgers, Margrit Klitz, and Achim Basermann
In the aerospace sciences we produce huge amounts of data. This data must be arranged in a meaningful order, so that we can analyze or visualize it. In this paper we focus on data that is distributed among computer processes and then needs to be sorted by a single root process for further analysis. We assume that the memory on the root process is too small to hold all sorted data at once, so that we have to perform the sorting and processing of data chunk-wise. We prove the efficiency of our approach in weak scaling tests, where we achieve a near constant bandwidth. Additionally, we obtain a considerable speed up compared to the standard parallel external sort. We also demonstrate the usefulness of our algorithm in a real-life aviation application.
In Situ Visualization of Performance-Related Data in Parallel CFD Applications
Rigel F. C. Alves and Andreas Knüpfer
This paper aims at investigating the feasibility of using ParaView as visualization software for the analysis and optimization of parallel CFD codes’ performance. The currently available software tools for reading proﬁling data do not match the generated measurements to the simulation’s original mesh and somehow aggregate them (rather than showing them on a time-step basis). A plugin for the open-source performance tool Score-P has been developed, which intercept an arbitrary number of manually selected code regions (mostly functions) and send their respective measurements – amount of executions and cumulative time spent – to ParaView (through its in situ library, Catalyst), as if they were any other ﬂow-related variable. Results show that (i) the impact of mesh partition algorithms on code performance and (ii) the load imbalances (and their eventual relationship to mesh size / simulation physics) become easier to investigate.