This Innovation Study is a pursuit of high-fidelity simulation of turbulent fluid flow stands to increase exascale computing capabilities as a result. In this study, the convergence of adequate computational resources and the need for numerical methods that balance accuracy and computational efficiency emerge as solutions to computational challenges. High-order numerical techniques, such as spectral element methods, show potential for this research due to their ability to produce detailed grids that are truly accurate and to solve problems faster. These methods have properties that favour large-scale parallelism and intensive computational operations, allowing efficient use of modern computer hardware.
The effectiveness of current spectral element codes depends on the distribution of the computational workload across the processing elements. While exhibiting reasonable scalability under optimal conditions, achieving peak performance at exascale requires either a monumental escalation in problem size, which may become impractical, or the introduction of algorithmic innovations capable of preserving parallel scaling even in scenarios with few elements per core. This introduces a host of new challenges, including increased heterogeneity, complex memory hierarchies, and the assimilation of novel memory and storage technologies, underscoring the urgency for adapted software engineering methodologies and innovative numerical techniques to effectively exploit the capabilities of these systems.
STRAUSS will address these challenges by focusing on enabling the use of accurate and efficient computational fluid dynamics simulations. The novel algorithms developed in this study will exploit breakthrough advances in hardware architectures, novel programming paradigms and efficient numerical methods. The algorithms will be prototyped in the framework of Neko, a high-fidelity flow solver based on spectral element methods and supported by the EuroHPC Centres of Excellence, CEEC and EXCELLERAT P2.