Applying future Exascale HPC methodologies in the energy sector

SN Computer Science, 2020

In this paper, we briefly discuss the challenges in porting hydrodynamic codes to futuristic exascale HPC systems. In particular, we sketch the computational complexities of finite difference (FD) method, pseudo-spectral method, and fast Fourier transform (FFT). The global data communication among the compute cores brings down the efficiency of pseudo-spectral codes and FFT. A FD solver involves relatively lower data communication. However, an incompressible FD flow solver has a pressure Poisson equation, whose computation in multigrid scheme is quite expensive. Hence, a comparative study between the two sets of solvers on exascale system would be valuable. In this paper, we report a comparative performance analysis between a FD code and a spectral code on a relatively smaller grid using 1024 compute cores of Shaheen II; here, the FD code yields comparable accuracy to the spectral code, but it is relatively slower. The above features need to be retested on much larger grids with many more processors.

Alya is the BSC in-house HPC-based multi-physics simulation code. It is designed from scratch to run efficiently in parallel supercomputers, solving coupled problems. The target domain is engineering, with all its particular features: complex geometries and unstructured meshes, coupled multi-physics with exotic coupling schemes and Physical models, ill-posed problems, flexibility needs for rapidly including new models, etc. Since its conception in 2004, Alya has shown scaling behaviour in an increasing number of cores. In this paper, we present its performance up to 100.000 cores in Blue Waters, the NCSA supercomputer. The selected tests are representative of the engineering world, all the problematic features included: incompressible flow in a human respiratory system, low Mach combustion problem in a kiln furnace and coupled electro-mechanical problem in a heart. We show scalability plots for all cases, discussing all the aspects of such kind of simulations, including solvers convergence.

Computational Geosciences

A three dimensional parallel implementation of Multiscale Mixed Methods based on non-overlapping domain decomposition techniques is proposed for multi-core computers and its computational performance is assessed by means of numerical experimentation. As a prototypical method, from which many others can be derived, the Multiscale Robin Coupled Method is chosen and its implementation explained in detail. Numerical results for problems ranging from millions up to more than 2 billion computational cells in highly heterogeneous anisotropic rock formations based on the SPE10 benchmark are shown. The proposed implementation relies on direct solvers for both local problems and the interface coupling system. We find good weak and strong scalalability as compared against a state-of-the-art global fine grid solver based on Algebric Multigrid preconditioning in single and two-phase flow problems.

2011

Multiple concerns over the impact of wide scale changes in land management have motivated comprehensive analyses of environmental sustainability of food and biofuel production. These call for high-resolution land management tools that enable comprehensive analyses of natural resources for decision-making. The agroecosystem simulation models with the most biophysical detail are point models, which often have a user interface that allows users to provide inputs and examine results for agricultural field scale analyses. These are not able to meet the needs of high-resolution regional or national simulations. We describe an efficient computational approach for deployment of the Environmental Policy Integrated Climate (EPIC) model at high-resolution spatial scales using high performance computing (HPC) techniques. We developed an integrated procedure for executing the millions of simulations required for high-resolution, regional studies, and also address building databases for model initialization, model forcing data, and model outputs. We first ported EPIC from Windows to an HPC platform and validated output from both platforms. We then developed methods of packaging simulations for efficient, unattended parallel execution on the HPC cluster. The job queuing system, Portable Batch System (PBS) is employed to control job submission. Simulation outputs are extracted to PostgreSQL database for analysis. In a case study covering four counties in central Wisconsin using HPC-EPIC, we finished over 140 K simulations in a total of 10 h on an HPC cluster using 20 nodes. This is a speedup of 40 times. More nodes could be used to achieve larger speedups. The HPC-EPIC model developed in this study is anticipated to provide information useful for high-resolution land use management and decision making. The framework for high-performance computing can be extended to other traditional, point-based biophysical simulation models.

Reverse Time Migration technique produces underground images using wave propagation. A discretization based on the Discontinuous Galerkin Method unleashes a massively parallel elastodynamics simulation, an interesting feature for current and future architectures. In this work, we propose to combine two recent HPC techniques to achieve a high level of efficiency: the use of runtimes (StarPU and PaRSEC) to easily exploit the hardware capabilitites, and the intergration of accelerators (Intel Xeon Phi). Preliminary results are presented.

Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis, 2021

High-performance Computing (HPC) systems have gone through many changes during the past two decades in their architectural design to satisfy the increasingly large-scale scientific computing demand. Accurate, fast, and scalable performance models and simulation tools are essential for evaluating alternative architecture design decisions for the massive-scale computing systems. This paper recounts some of the influential work in modeling and simulation for HPC systems and applications, identifies some of the major challenges, and outlines future research directions which we believe are critical to the HPC modeling and simulation community.

Journal of Aerospace Sciences and Technologies, 2023

The design of complex engineered systems, such as with aerospace and automotive systems, results in a multi-discipline simulation environment that involve a high number of design subsystems and engineering disciplines as well as interactions between them. The development of novel methods and algorithms for accurately predicting the behavior physics, the increasing levels of high capability and cost effective HPC, and the availability of a broad range of independent software vendor tools are contributing towards the widespread usage of high fidelity simulation models and tools for multi-discipline analysis and optimization of product designs. The increasing usage of CFD and the complexity of associated multi-discipline simulations (analysis and design optimization) are fueling the need for balanced HPC infrastructure involving a combination of high productivity and peak performance computing, shared storage and parallel file systems for IO and data management, and advanced visualization systems, all connected by high speed networks. This paper discusses a hybrid computing architecture that optimally addresses multi-discipline CAE workloads involving high fidelity technical computing applications within a product development workflow.

2022

Rotorcraft technologies pose great scientific and industrial challenges for numerical computing. As available computational resources approach the exascale, finer scales and therefore more accurate simulations of engineering test cases become accessible. However, shifting legacy workflows and optimizing parallel efficiency and scalability of existing software on new hardware is often demanding. This paper reports preliminary results in CFD and structural dynamics simulations using the T106A Low Pressure Turbine (LPT) blade geometry on Leonardo S.p.A.'s davinci-1 high-performance computing (HPC) facility. Time to solution and scalability are assessed for commercial packages Ansys Fluent, STAR-CCM+, and ABAQUS, and the open-source scientific computing framework PyFR. In direct numerical simulations of compressible fluid flow, normalized time to solution values obtained using PyFR are found to be up to 8 times smaller than those obtained using Fluent and STAR-CCM+. The findings extend to the incompressible case. All models offer weak and strong scaling in tests performed on up to 48 compute nodes, each with 4 Nvidia A100 GPUs. In linear elasticity simulations with ABAQUS, both the iterative solver and the direct solver provide speedup in preliminary scaling tests, with the iterative solver outperforming the direct solver in terms of time-to-solution and memory usage. The results provide a first indication of the potential of HPC architectures in scaling engineering applications towards certification by simulation, and the first step for the Company towards the use of cutting-edge HPC toolkits in the field of Rotorcraft technologies.

2012

Resumen Presently, the Solid Earth Sciences started to move towards implementing High Performance Computational (HPC) research facilities. One of the key tenants of HPC is performance, which strongly depends on the interaction between software and hardware. In this paper, they are presented benchmark results from two HPC systems.