London SW7 2AZ, United Kingdom
Philipp Schlatter (KTH Royal Institute of Technology):
The complex nature of turbulent fluid flows implies that the computational resources needed to accurately model problems of industrial and academic relevance is virtually unbounded. Computational Fluid Dynamics (CFD) is therefore a natural driver for exascale computing both for academic and industrial cases, and has the potential for substantial societal impact, like reduced energy consumption, alternative sources of energy, improved health care, and improved climate models. This proposed minisymposium will be organized by the EU funded Horizon 2020 project ExaFLOW and will feature presentations showcasing their work on addressing key algorithmic challenges in CFD in order to facilitate simulations at exascale, e.g. accurate and scalable solvers, data reduction methods and strategies to ensure fault tolerance and resilience.
In particular, the talks in this minisymposium will highlight the following topics: Adaptive mesh refinement:
The generation of a high-quality mesh is a complex task and is usually considered as a bottleneck for any user of a CFD code. Ideally, the mesh should avoid unnecessary over-resolution in unimportant regions, resolve the critical parts and still require as few computational resources as possible. Adaptive mesh refinement (AMR) techniques have been developed to tackle this issue by, first, identifying poorly resolved regions with effective error estimators based on adjoints, and then, applying a mesh refinement strategy. Both techniques are implemented in Nek5000, an open-source, highly scalable and portable code based on the spectral element method (SEM). Issues of efficient implementation of non-conformal elements, preconditioners and potential stability issues will be discussed, together with an analysis of adjoint-based error estimators in viscous flow. Results for 2D and 3D laminar and turbulent flows are presented.
Mixed CG-HDG schemes: The number of computing cores expected to become available in an exascale environment means that there is huge potential for significantly decreasing the execution times of large-scale CFD simulations. However to capitalise on this, we require methods that have good strong-scaling potential. The strong scaling of high-order methods, whilst typically better than lower-order schemes, is still expected to be an issue for even extremely large simulations. We will show progress in deriving a mixed scheme of continuous Galerkin (CG) and hybridizable discontinuous Galerkin (HDG) methods in order to reduce communication between nodes whilst retaining favorable on-node performance.
Leveraging high-order methods for industrial flows: Another key impact area for high-order methods is their potential to give new insights into complex industrial flows. This is particularly true as we look towards leveraging exascale platforms to perform high resolution LES/DNS experiments, which are yet to be used as part of the industrial design process over existing RANS/DES methods due to the computational requirements. This presentation will highlight the progress being made in applying developments under the ExaFLOW project to investigate key aerodynamic challenges in the Formula 1 motorsport industry.
This session aims at bringing together the CFD community as a whole, from HPC experts to domain scientists, discuss current and future challenges towards exascale fluid dynamics simulations and facilitating international collaboration.
Nonconforming elements in Nek5000: Stability and Implementation - Adam Peplinski and Nicolas Offermans (KTH Royal Institute of Technology), Paul Fischer (University of Illinois), and Philipp Schlatter (KTH Royal Institute of Technology)
Adjoint-based error estimators and adaptive refinement in Nek5000 - Nicolas Offermans and Adam Peplinski (KTH Royal Institute of Technology), Oana Marin (Argonne National Laboratory), and Philipp Schlatter (KTH Royal Institute of Technology)
On a mixed CG-HDG formulation for high-order simulations - Martin Vymazal (Imperial College London), David Moxey (University of Exeter), Chris Cantwell (Imperial College London), Robert M. Kirby (University of Utah), and Spencer Sherwin (Imperial College London)
Comparison of time resolved experimental surveys with simulations using Nektar++ on a Formula One front wing - Julien Hoessler (McLaren Racing Ltd)
A minimally intrusive low-memory approach to resilience for existing transient solvers - Chris Cantwell (Imperial College London) and Allan Nielsen (École polytechnique fédérale de Lausanne)
Powered by iCagenda