Excalibur-SLE Exascale HPC for System Level Engineering

Workshop III: Towards Exascale Simulation of Integrated Engineering Systems

The imminent deployment of exascale computing will usher in a new era of engineering system prediction, design and optimisation. It would enable a whole-system approach in a seamless fashion that significantly cuts the development cycle and cost. Before such a full potential can be realised, a fundamental rethink is needed about how we construct physical models and application software codes for the new computer architectures and software libraries. Sophisticated engineering systems typically involve multi-physical processes, such as flow, heat transfer, chemical reactions, electromagnetics, acoustics, and fluid-structure interaction, often in complex geometries. Spatial scales can span 9-12 orders of magnitude and the range of temporal scales can be even wider. In most cases, computer codes have been developed and optimised for specific applications with a limited range of time and length scales. The corresponding models and numerical methods pose different challenges on exascale computers. How to integrate existing methods and/or develop completely new methods to make full use of exascale computing is a grand challenge for mathematicians, physicists, computer scientist as well as engineers. In this workshop, leading experts in the field will showcase recent efforts in the development of modelling methodologies, computer algorithms, software interfaces, and software/hardware co-design in preparation for the arrival of exascale computing. Cutting-edge simulations of engineering systems at extreme time and length scales will be demonstrated. Finally, existing gaps and remaining issues will be identified to stimulate further discussions and actions in the field.

21-22 January 2021

Session 1: Chair – David Emerson (DL, UK)

Session 2: Chair – Peter Coveney

Session 3: Chair – Timo Betcke

Session 4: Chair – Jianping Meng

Benjamin Dudson (University of York, UK)

Coupling codes at Exascale for the ExCALIBUR UKAEA NEPTUNE Nuclear Fusion Project

Andrew Davis (UKAEA)

Multiphysics in support of Fusion Component Design

Peter Coveney (UCL, UK)

Applications and challenges for systems engineering on the road to the exascale

In this talk I will highlight some of the work that we are doing within the Centre for Computational Science at UCL that is enabling practical evaluation of several of the current globally leading multi-petascale HPC facilities in preparation for the advent of exascale machines. This work includes simulations at the bleeding edge for applications in cardiovascular blood flow, materials design and drug discovery. Uncertainty quantification is an essential element in the development of actionable engineering solutions using computers. I will illustrate this by discussing an application of the software tools we have developed to assess the uncertainty in an epidemiological model of the COVID-19 pandemic used to advise UK government ahead of the March 2020 lockdown.

Peter Coveney holds a Chair in Physical Chemistry and is Director of the Centre for Computational Science. He has been at the forefront of HPC-based scientific research for the past twenty years and has led the development of several codes, including the HemeLB and LB3D lattice-Boltzmann packages, which scale to hundreds of thousands of cores on several of the largest supercomputers available. He leads many large projects (MAPPER, ComPat, VECMA FETHPC and CompBioMed1 and 2 phases of the CoE) in which HPC is a central component, including the EPSRC RealityGrid project and subsequent Platform Grant, and the EU HPC Centre of Excellence in Computational Biomedicine.

Yvan Fournier (EDF)

Continuous evolution of an industrial CFD code for future environments

As one of the largest electricity production utilities, EDF has made heavy use computational fluid dynamics (CFD) simulations for many years, both to study improvements to various production means (from nuclear to wind), and as a part of safety studies around pressurized water reactors. For most of these applications, EDF has been continuously developed and maintained the code_saturne CFD solver since 1997, consolidating experience with older codes. To encourage and facilitate collaboration, it has been released under a free software license since 2007.

As with most industrial codes of this type, design must balance ease of use, robustness, performance, an a wide application and validity domain, where a tool’s life cycle is often longer than thar of many underlying computing technologies it can use.

Over the years, this has been handled by many incremental code base evolutions, maintaining features and compatibiliy across several versions, while rewriting and improving various sub-systems, in a progressive upgrade cycle.

To keep the user workflow simple and strive for maximum portability and perfoemance, we have striven over time to ensure as many pre and post-processing operations as possible are integrated as steps in the main distributed parallel run, including several in-situ post-processing options, and parallel coupling features. This has been done using a combination of internal library functions based on the code’s main data structures, and optionally leveraging external libraries, especially for mesh partitioning and in-situ post-processing.

We will present several of the latest evolutions related to high performance computing aspects of code_saturne, both related to in-situ pre and post-processing, and to the computational models and kernels that have allowed us to adapt for current compute architectures and prepare for the future, while keeping a consistent long-term approach.

Yvan Fournier obtained a “diplôme d’ingénieur” (equivalent to a Master) in aeronautics in 1994 from École Centrale Paris. He has worked as a researcher at EDF R&D since 1998, and is currently a principal research engineer, working on various HPC, pre and post-processing, and software engineering aspects of EDF’s CFD in-house codes, mainly code_saturne (code-saturne.org). His current interests include in-situ mesh improvement and post-processing, distributed algorithms, software engineering, and high performance computing. Past interests also include CFD modeling of cooling towers and PWR fuel assemblies.

Prof Haohuan Fu (Tsinghua University & National Supercomputing Center in Wuxi)

Optimizing CESM-HR on Sunway TaihuLight and An Unprecedented Set of Multi-Century Simulation

CESM is one of the very first and most complex scientific codes that gets migrated onto Sunway TaihuLight. Being a community code involving hundreds of different dynamic, physics, and chemistry processes, CESM brings severe challenges for the many-core architecture and the parrallel scale of Sunway TaihuLight. This talk summarizes our continuous effort on enabling efficient run of CESM on Sunway, starting from refactoring of CAM in 2015, redesigning of CAM in 2016 and 2017, and a collaborative effort starting in 2018 to enable highly efficient simulations of the high-resolution (25 km atmosphere and 10 km ocean) Community Earth System Model (CESM-HR) on Sunway Taihu-Light. The refactoring and optimizing efforts have improved the simulation speed of CESM-HR from 1 SYPD (simulation years per day) to 5 SYPD (with output disabled). Using CESM-HR, We manage to provide an unprecedented set of high-resolution climate simulations, consisting of a 500-year pre-industrial control simulation and a 250-year historical and future climate simulation from 1850 to 2100. Overall, high-resolution simulations show significant improvements in representing global mean temperature changes, seasonal cycle of sea-surface temperature and mixed layer depth, extreme events and in relationships between extreme events and climate modes.

Haohuan Fu is a professor in the Ministry of Education Key Laboratory for Earth System Modeling, and Department of Earth System Science in Tsinghua University, where he leads the research group of High Performance Geo-Computing (HPGC). He is also the deputy director of the National Supercomputing Center in Wuxi, leading the research and development division. Fu has a PhD in computing from Imperial College London. His research work focuses on providing both the most efficient simulation platforms and the most intelligent data management and analysis platforms for geoscience applications, leading to two consecutive winning of the ACM Gordon Bell Prizes (nonhydrostatic atmospheric dynamic solver in 2016, and nonlinear earthquake simulation in 2017).

Serge Guillas (UCL, UK)

Reduced Order Modelling

Uncertainty quantification of complex coupled models is done via building surrogates (known as statistical emulators) that mimic the input-output relationship, in order to alleviate the computational burden of running ensembles. However, the complexities of models are often large so building emulators can be challenging. In this talk, we discuss strategies to reduce input and output dimensions with little loss in information, exploiting input sensitivities and output variability over time or space. We also propose a bespoke approach for systems of sub-models where emulators can be linked to achieve large gains in computational efficiency and approximation accuracy.

S. Guillas is Professor of Statistics at UCL and the Met Office Joint Chair in Data Sciences for Weather and Climate. He obtained his PhD (Paris 6 Pierre-et-Marie-Curie, France) in 2001. He was postdoc at the University of Chicago, USA over 2002-2004, Assistant Professor (Georgia Institute of Technology, USA) over 2004-2007, and joined UCL in 2007. He was vice-Chair of the SIAM activity group on Uncertainty Quantification (UQ) over 2015-2016. He is currently Chair of the Working group on Uncertainties in the COST action “Accelerating Global science In Tsunami HAzard and Risk analysis” (AGITHAR, 25 countries). He leads several projects on UQ, mostly for multi-scale and multi-physics simulations on HPC. He founded in 2019 the UQ interest group at the Alan Turing Institute.

Prof Shantenu Jha (Rutgers University, USA)

ZettaScale Computing on Exascale Platforms

We outline the vision of “Learning Everywhere,” which captures the impact of learning methods coupled to traditional HPC methods. We present several examples of “effective performance” improvements for traditional HPC simulations that learning (MLforHPC) provides. We discuss how we are applying the “Learning Everywhere” paradigm to advance therapeutics for COVID19. We will showcase the challenges and performance of scalable integrated HPC & AI software infrastructure developed to support long-running computational campaigns as part of the DOE’s Medical Therapeutics project under the umbrella of the National Virtual Biotechnology Laboratory.

Shantenu Jha is the Chair of Computation & Data Driven Discovery Department at Brookhaven National Laboratory, and Professor of Computer Engineering at Rutgers University. His research interests are at the intersection of high-performance distributed computing and computational & data science. Shantenu leads the the RADICAL-Cybertools project which are a suite of middleware building blocks used to support large-scale science and engineeringapplications. He was appointed a Rutgers Chancellor’s Scholar (2015) and was the recipient of the inaugural Chancellor’s Excellence in Research (2016) for his cyberinfrastructure contributions to computational science. He is a recipient of the NSF CAREER Award (2013), the Gordon Bell Award (2020) and several other prizes at SC’xy and ISC’xy,well as the winner of IEEE SCALE 2018. More details can be found at: http://radical.rutgers.edu/shantenu

Tzanio Kolev (Lawrence Livermore National Laboratory)

Efficient Finite Element Discretizations for Exascale Applications

Exascale architectures require rethinking of the numerical algorithms used in many large-scale applications. These architectures favor algorithms that expose fine-grain parallelism and maximize the ratio of floating point operations to energy intensive data movement. One of the few viable approaches to achieve high efficiency in the area of PDE discretizations on unstructured grids is to use matrix-free/partially-assembled high-order finite element methods to increase the accuracy and/or lower the computational time due to reduced data motion.

In this talk we report on recent work in the Center for Efficient Exascale Discretizations (CEED) [1], a co-design center in the US Exascale Computing Project that is focused on the development of next-generation discretization software and algorithms to enable a wide range of finite element applications to run efficiently on future hardware. CEED is a research partnership involving 30+ computational scientists from two US national labs and five universities, including members of the Nek5000, MFEM [2], MAGMA, OCCA and PETSc projects.

Topics of discussion will include recent progress in CEED packages and applications, new miniapps, benchmarks and libraries developed in the project, efficient GPU algorithms, and our efforts in unstructured adaptive mesh refinement, matrix-free linear solvers and high-order data analysis and visualization.

[1] Center for Efficient Exascale Discretizations, https://ceed.exascaleproject.org

[2] MFEM finite element library, https://mfem.org

Tzanio Kolev is a computational mathematician at the Center for Applied Scientific Computing in Lawrence Livermore National Laboratory (LLNL), where he works on finite element discretizations and solvers for problems in compressible shock hydrodynamics, multi-material arbitrary Lagrangian Eulerian methods, radiation hydrodynamics, and computational electromagnetics. Tzanio is the director of the Center for Efficient Exascale Discretizations (CEED) in US Exascale Computing Project and leads the high-order finite element discretization research and development efforts in the MFEM and BLAST projects at LLNL. Tzanio’s research interests include the development and analysis of advanced finite element discretization methods, massively parallel preconditioners, discretization-enhanced algebraic multigrid algorithms, and the design and implementation of large-scale scientific software.

Stephen Longshaw (STFC, DL, SC; UK)

General High-Performance Code Coupling

In this talk I will describe the open-source Multiscale Universal Interface (MUI) code coupling library. A joint development between UKRI STFC, Lawrence-Berkeley National Laboratory, IBM Research UK and Brown University, MUI is a C++ header-only library that provides a simple point-based data transfer paradigm, alongside an extensible collection of spatial and temporal sampling algorithms. It is designed entirely around the MPI MPMD (Multiple Program Multiple Data) concept and offers great potential for use in an HPC environment. As example I will also discuss a recent effort using MUI to couple a Molecular Dynamics and Direct Simulation Monte Carlo solver to enable solution of microscopic problems involving gas and large simulation domain sizes.

Stephen is a Principal Computational Scientist in the Computational Engineering Group based at UKRI STFC’s Daresbury Laboratory. He is a computer scientist that specialises in computational modelling (especially computational fluid dynamics (CFD) using particle methods such as smoothed particle hydrodynamics) and scientific code coupling for both multi-scale and multi-physics problems. He also has a keen interest in understanding and utilising novel and emerging computing hardware and HPC and, in particular, how these two interests combine to form the new exascale landscape.

Dr Daniel Lycett-Brown (Dassault Systèmes)

Multiscale simulation of multiphase fluid dynamics for engineering applications

The computational simulation of fluid dynamics is required across a wide range of engineering applications. Here we focus on multiphase fluid dynamics using the lattice Boltzmann method, with applications ranging from the micron scale (porous media, ink-jet printing, lab-on-a-chip etc.) to the meters scale (car wading, offshore wind turbines, ship dynamics etc.). Different physics dominates at different scales which has led to the development of different modelling approaches for different applications. However many practical applications bridge a wide range of scales and this modelling approach breaks down. We propose solutions to this problem, for which exascale computing plays a significant role. Highly efficient and scalable codes that can take full advantage of modern CPU supercomputers and GPU clusters are allowing simulations with higher resolutions than previously possible. Combined with technologies for adaptive mesh refinement and effective load balancing across MPI processes this will allow a larger range of scales to be simulated with acceptable turnaround times. However in some circumstances subgrid scale phenomena will still require modelling. Development of models for these subgrid phenomena and coupling between models at different scales is required. Exascale computing can again play a role; large numbers of high-resolution simulations at the microscale can be used to inform these subgrid models (for example droplet collision maps can be built up to inform spray dynamic models). The combination of new models, coupled across time and length scales, and the power of exascale computing will be revolutionary in the simulation tools needed to design the engineering applications of tomorrow.

Dr Christopher Maynard (University of Reading and Met Office, UK)

Realising Richardson’s dream: Developing modelling capability at Exascale for the Met Office

The Met Office is developing a new weather and climate modelling system to exploit Exoscale computing. The atmospheric model is named LFRic after Lewis Fry Richardson, who first attempted numerical weather prediction a century ago. The LFRic model is being co-designed with a Domain Specific Language (DSL) and new scientific methods, such as communication avoiding algorithms. The DSL allows for single source science code with parallel code for different programming models. Some Initial performance results are shown.

Chris Maynard has been an Associate Professor of Computer Science at the University of Reading since January 2018 and he retains his position as an Expert Scientific Software Engineer at the Met Office which he has held since 2012. Prior to joining the Met Office he worked at the Edinburgh Parallel Computing Centre (EPCC) where he developed and optimised scientific software for supercomputers in diverse fields such as Particle Physics, Computational Fluid Dynamics (CFD), Magnetic materials, acoustic properties of materials, Group Theory, Numerical Analysis and Quantum Computing. He obtained his PhD in 1998 from the University of Edinburgh in theoretical particle physics. Since joining the Met Office he has optimised the solver for the UM dynamical core called ENDGame. He is leading the development of a domain specific language for the new Met Office Model, Gung Ho/LFRic to enable performance on diverse exascale computing architectures. His research interests include methods for High Performance Computing, parallel programming models, languages and algorithms and domain specific languages.

Dr Ernesto Monaco (ESS Engineering Software, Austria)

Numerical simulation of industrial relevant applications by the ALSIM™ Platform

The talk will give an overview about the most recent developments of the ALSIM™ platform, which represents the unique ESS solution to tackle several relevant industrial applications, with special emphasis on topics related to car manufacturing: examples will include Paint-shop (E-Coating, Oven), as well as washing, pre-treatment and deformation of metal sheets. ALSIM employs a portfolio of several solvers (LBM, FDM, SPH, DEM) in combination to specifically tailored GUIs to deliver accuracy, performance and easiness to use even for engineers with limited CFD expertise.

Dr. Ernesto Monaco got his Master Degree in Aerospace Engineering at University of Naples in 2004. In 2009 he obtained a Ph.D. at Southampton University about Lattice Boltzmann simulations of complex multiphase flows, which has been his most relevant research interest since then. From 2008 to 2013 and 2016 to Spring 2018 he was Post-Doctoral researcher, respectively at TU-Clausthal and Erlangen University (Germany). After working as software developer at Fluidyna (now part Altair Group) he is leader of the LBM group at ESS (Engineering Software Steyr) since July 2018.

Pierre Sagaut (Aix-Marseille Université)

Unsteady simulations of complex turbulent flows using Lattice-Boltzmann Methods

The presentation deals with recent developments in the design of Lattice-Boltzmann methods (LBM) for the simulation of turbulent flows in complex geometries, with a broad range of applications ranging from urban physics, micrometeorology to aerospace engineering. Two key issues are addressed, more precisely 1) accounting for realistic thermodynamics (from perfect gas with compressibility effects to humid air with phase changes) while optimizing the numerical efficiency of the method and 2) stabilizing the method while preserving the low-dispersion/low-dissipation feature of the original LBM method. The latter point will be addressed within the Scale Resolving Approach for turbulent flow simulation, and the optimization of the stabilizing procedure for Implicit LES, explicit LES and hybrid RANS-LES methods will be discussed. Since LBM is based on the use of Cartesian grids, the development of efficient immersed boundary techniques along with the coupling with turbulent wall models is a key issue that will be illustrated. During the talk, the emphasis will be put on the Hybrid Recursive Regularized LBM approach developed at the M2P2 laboratory that is implemented in the ProLB software.

Professor Pierre Sagaut is Director of M2P2 Laboratory at Aix-Marseille Université, associated with CNRS and Ecole Centrale Marseille. He also holds the Chaire Airbus-Renault-Safran on Lattice-Boltzmann Methods for CFD. He has a PhD in Fluid mechanics (1995) from Université Pierre et Marie Curie, and has worked for various organisations including the French National Aerospace Lab. Since 1997, he has successfully supervised 68 PhD students and 15 postdoctoral fellows. He has published extensively in journals, conferences and books on topics such as computational fluid dynamics, aerodynamics, aeroacoustics, uncertainty quantification, aerospace engineering, urban physics, turbulence dynamics and modelling, numerical methods. Prof Sagaut has received numerous honours and awards including 2002: John Green prize (International Council for Aeronautical Sciences) in 2002, Prize of Association of European Research Establishments in Aerospace (shared with G. Desquesnes, E. Manoha, M. Terracol) in 2007 and Grand Prix de l’Académie des Sciences/ EADS in « Science & Engineering » in 2010. He has served as an editor/editor-in-chief for five international journals.

Sauro Succi (Istituto Italiano di Tecnologia & CNR & University of Rome, Italy: Harvard University, USA)

Simulating Dropland: Computational Design of Exotic Soft Materials with an Eye to Exascale Computing

We present recent variants of the Lattice Boltzmann method for the mesoscale modelling of exotic soft materials, such as microfluidic crystals, dense and mul-ticore emulsions and recent Petascale simulations of deep-sea glassy sponges, with an eye on prospective exascale simulations

Dr Sauro Succi holds a degree in Nuclear Engineering from the University of Bologna and a PhD in Plasma Physics from the EPFL, Lausanne, Switzerland. He has held a research staff position at the IBM European Center for Scien-tific and Engineering Computing, Rome. Till 2018 he served as a Director of Research at the Istituto Applicazioni Calcolo of the Italian National Research Council in Rome and he is also a Research Associate of the Physics Department of Harvard University and a regular Visiting Professor at the Institute of Ap-plied Computational Science at the School of Engineering and Applied Sciences of Harvard University. Since 2018 he is a Senior Research Executive and Prin-cipal Investigator at the Center for Life Nanosciences of the Italian Institute of Technology at La Sapienza, Rome.

He has published extensively on a broad range of topics in computational physics, soft matter and the interface between physics and biology.

He is the author of the highly cited monograph ”The lattice Boltzmann equation for fluid dynamics and beyond”, (Oxford Univ. Press, 2001) and ”The Lattice Boltzmann Equation for Complex States of Flowing Matter” (OUP, 2018).

Dr Succi is an elected Fellow of the American Physical Society (1998), a a member of the European Physical Society and an elected member of the Academia Europaea (2015). He has received the Humboldt Prize in physics (2002), the Killam Award of the the University of Calgary (2005) and the Ra-man Chair of the Indian Academy of Sciences (2011). In 2017, he has been awarded a European Research Council Advanced Grant on the computational design of mesoscale porous materials. He is also the recipient of the 2017 APS Aneesur Rahman Prize for Computational Physics for seminal contributions to the development and application of the Lattice Boltzmann method and the 2019 Bernie J. Alder CECAM prize for exceptional contributions to the microscopic simulation of matter.