Periodic Reporting for period 1 - exaFOAM (Exploitation of Exascale Systems for Open-Source Computational Fluid Dynamics by Mainstream Industry)
Okres sprawozdawczy: 2021-04-01 do 2022-09-30
The exaFOAM project aims to exploit new architectures and hybrid approaches is undertaken through the development and validation of algorithmic improvements, across the entire CFD process chain (preprocessing, simulation, I/O, post-processing). Effectiveness is demonstrated via a suite of HPC Grand Challenge and Industrial Application Challenge across all sectors of industry (transportation, power generation, disaster prevention, health and safety) to increase energy efficiency comfort and health/safety.
Stated AIMS:
- Demonstrate one or more orders-of-magnitude performance improvement on industry-based grand challenges
- Release technology advances realised during this project via opensource OpenFOAMvYYXX as respecting the terms of GPLv3
- Exploit performance gains among European Industry partners vested in exaFOAM as supporters/stakeholders in coordination with the OpenFOAM Governance Structure
Our collective Mission, Goals and Objectives are unchanged.
They are further consolidated through engagement with HPC-entities worldwide including AWS and Huawei (expect to be ratified as new Stakeholders shortly) - such are the direct means for technology end-users to gain access to the performance gains realised during this project.
Society IMPACT:
We restate that the Supporters and Stakeholders represent original equipment manufacturers whose design improvements realised by performance gains in this project benefits society directly in the form of energy-efficiencies, pollution-reduction and improved health/safety.
WP1 — Management and Coordination: Addressing the challenges w.r.t. beneficiaries’ changes in management, resourcing and workplans, in particular challenges due to Covid in the period April-2021 to June 2022.
Regular meetings monitor the progress of the exaFOAM consortium members and report the progress to the Stakeholders
- Weekly touchpoint held every Wednesday @4.00CET via TEAMS virtual meetings
- Monthly Management Board meetings actions with minutes, held on the last Wednesday of every month via TEAMS virtual meetings
- Six-monthly General Assembly (GA) meetings, initially held virtually via TEAMS virtual meetings during Covid restrictions, and physically at Month-18.
- GA closely followed by TEAMS virtual meetings with global stakeholders to update them on progress
WP2 — Validation and Assessment: Grand Challenges, Industrial Benchmarks and Micro-benchmarks are all identified, and together with the Stakeholders, undergoing continuous monitoring, revision and updating to be current and progressive. These benchmarks are systematically made available to the end-user Stakeholders in order to quantify the performance gains realised during this project.
WP3 — Code Refactoring: Rearchitecting of software in order to realise HPC technology advanced. This mainly comprises the processing of sparse matrices (LUD) using matric compression (CSR) assessed on several state-of-the-art architectures based around Intel-Xeon, AMD-EPYC and Nvidia-GPU chip-sets. Performance gains of “percentage” and “factors” are realised which collectively increment towards the Objective of one-or-more orders of magnitude performance improvement. “Order-of-magnitude” performance improvements are addressed in the next workpackage.
WP4 — Evolution: Realisation of highly parallel solver gains based on coupled-solver techniques, I/O efficiencies, load-balancing especially for rotating-machinery and multi-physics applications, and external linear algebra solvers. The potential for orders-of-magnitude performance gains are clearly stated in respect of GPU or hybrid CPU-GPU architectures.
WP5 — Co-design, Profiling and Performance analysis: Profiling tools are in place and used to assess the benchmarks delivered in WP2, in respect of
- Parallel efficiency: The extent to which all resources in the system are kept active doing useful work
- Load Balance: The efficiency loss due to the global distribution of work among processes, in particular due to one process doing more computation than the others.
- Communication efficiency: The efficiency loss due to the communication of data.
- Transfer efficiency: Efficiency loss related to the non-instantaneous nature of communication mechanisms.
- Serialization efficiency: Inefficiencies caused by circular dependences or non-uniform imbalances.
- Computation scalability: How the time spent computing scales with respect to the reference case.
- Instructions scalability: How the number of instructions scales with respect to the reference case.
- IPC scalability: How the IPC scales with respect to the reference case.
- Frequency scalability: How the frequency scales with respect to the reference case.
WP6 — Integration: The several technologies under consideration in the previous work-packages are in pre-release format. The implementation of the final versions of these various technologies, validations and regression tests are in preparation to be release in the second half of this project.
WP7 — Dissemination, Impact and Exploitation: Partner exchange platform, external website for public visibility and dissemination programme are all in place, together with metrics of publications and presentations to conferences and workshops.
From the start of the project, the HPC performance of general-purpose Computational Fluid Dynamics algorithms used in mainstream is notably worse than idealised algorithms due to inherent bandwidth needs, three-dimensional and time-date handling with non-sparse matrix dependencies, spatial domain decomposition requirements, I/O challenges. The HPC-picture is constantly evolving, most notable in the development of GPU and hybrid CPU-GPU systems.
BEYOND THE STATE OF THE ART:
So far in this project the follow significant results have been achieved:
WP2 — Validation and Assessment: Benchmarks identified which are directly relevant to HPC profiling, repeatable and sharable to the Stakeholders
WP3 — Code Refactoring: Incremental gains which may collectively sum to order-of-magnitude performance gains
WP4 — Evolution: Significant potential for order-of-magnitude performance gains by adapting the algorithms to new architectures (GPU and hybrid GPU/CPU)
WP5 — Co-design, Profiling and Performance analysis: Profiling tools in place to quantify parallel efficiency, load-balance, communication efficiency, transfer efficiency, serialization efficiency, computation scalability, instructions scalability, IPC scalability and frequency scalability
WP7 — Dissemination, Impact and Exploitation: Partner exchange platform, external website for public visibility and dissemination programme are all in place
ONGOING OBJECTIVES:
We expect all three stated objectives to be achieved by the end of the project, namely; demonstration of order-of-magnitude performance improvements for general purpose CFD applications, visibility in the public domain via Quality Assured opensource releases and end-user engagement via dissemination to the key Stakeholders and European user community.
IMPACT:
European original equipment manufacturers (OEMs) will use the technologies realised in this project to accelerate the time-to-market and improve engineering design towards energy efficiency, comfort and health/safety.