Periodic Reporting for period 1 - Plasma-PEPSC (Plasma Exascale-Performance Simulations CoE - Pushing flagship plasma simulations codes to tackle exascale-enabled Grand Challenges via performance optimisation and codesign)
Période du rapport: 2023-01-01 au 2023-12-31
The project's overall vision is to push flagship plasma simulation codes (BIT, GENE, PIConGPU, and Vlasiator) to tackle exascale-enabled grand challenges via performance optimization and co-design.
In more detail, the specific goals of the Plasma-PEPSC project are the following:
1. Objective I: Enable unprecedented plasma simulations and associated extreme-data analytics to address grand challenges in plasma physics that are impossible to solve on current extreme-scale systems and require exascale computing. The four grand challenges Plasma-PEPSC aims at addressing are
a. Controlling plasma-material interfaces
b. Optimising magnetically confined plasmas
c. Designing next-generation plasma accelerators
d. Predicting space plasma dynamics in the Earth’s magnetosphere
2. Objective II: This objective requires maximizing the performance of four European plasma codes (BIT, GENE, PIConGPU, Vlasiator) in terms of parallel performance and efficiency on European exascale and pre-exascale systems, building on algorithmic advances (regarding load balancing, data compression, resilience, etc.) as well as on programming model and library developments (Message Passing, accelerator, and data movement APIs and runtimes).
a. Ambition is to reach good parallel efficiency on the EuroHPC exascale systems for the identified exascale challenges.
3. Objective III: Establish and ensure an integrated HPC software engineering approach for deploying, verifying, and validating extreme-scale kinetic plasma simulations that can serve as a community standard.
4. Objective IV: Establish a continuous and integrated co-design methodology for the four plasma codes to provide/receive direct input to/from the design and development of the EPI Processor from SiPearl, the European accelerator (BSC and FORTH as ones of the most important developers), EUPEX and EuPILOT.
5. Objective V: Enable high-throughput analysis of the large simulation result datasets from the four high-fidelity kinetic plasma simulations with optimized and scalable parallel I/O, data compression techniques for kinetic codes, online data analysis building on efficient data streaming, and integrated AI-based data analysis.
6. Objective VI: Exploit and maximize the usage of the computational power of accelerators available on current and next-generation European supercomputers for the four plasma simulation codes with algorithmic improvements for increasing code transfer bandwidth, extending programming interfaces, and minimizing data movement. Together with optimizing the usage of accelerators, we exploit the on-node memory hierarchy (DRAM, HBM, and persistent memories) by optimizing data locality and decreasing data movement. By doing that, we will also make our plasma codes more energy efficient.
7. Objective VII: Create synergistic collaborations with other CoEs, EuroHPC, and Competence Centers for cross-fertilisation and supporting the significant adoption and full exploitation of the Plasma-PEPSC simulation codes, methodology, approaches and ensuring synergies in preparing training hackathons, workshops, and conferences.
In more detail, the specific goals of the Plasma-PEPSC project are the following:
1. Objective I: Enable unprecedented plasma simulations and associated extreme-data analytics to address grand challenges in plasma physics that are impossible to solve on current extreme-scale systems and require exascale computing. The four grand challenges Plasma-PEPSC aims at addressing are
a. Controlling plasma-material interfaces
b. Optimising magnetically confined plasmas
c. Designing next-generation plasma accelerators
d. Predicting space plasma dynamics in the Earth’s magnetosphere
2. Objective II: This objective requires maximizing the performance of four European plasma codes (BIT, GENE, PIConGPU, Vlasiator) in terms of parallel performance and efficiency on European exascale and pre-exascale systems, building on algorithmic advances (regarding load balancing, data compression, resilience, etc.) as well as on programming model and library developments (Message Passing, accelerator, and data movement APIs and runtimes).
a. Ambition is to reach good parallel efficiency on the EuroHPC exascale systems for the identified exascale challenges.
3. Objective III: Establish and ensure an integrated HPC software engineering approach for deploying, verifying, and validating extreme-scale kinetic plasma simulations that can serve as a community standard.
4. Objective IV: Establish a continuous and integrated co-design methodology for the four plasma codes to provide/receive direct input to/from the design and development of the EPI Processor from SiPearl, the European accelerator (BSC and FORTH as ones of the most important developers), EUPEX and EuPILOT.
5. Objective V: Enable high-throughput analysis of the large simulation result datasets from the four high-fidelity kinetic plasma simulations with optimized and scalable parallel I/O, data compression techniques for kinetic codes, online data analysis building on efficient data streaming, and integrated AI-based data analysis.
6. Objective VI: Exploit and maximize the usage of the computational power of accelerators available on current and next-generation European supercomputers for the four plasma simulation codes with algorithmic improvements for increasing code transfer bandwidth, extending programming interfaces, and minimizing data movement. Together with optimizing the usage of accelerators, we exploit the on-node memory hierarchy (DRAM, HBM, and persistent memories) by optimizing data locality and decreasing data movement. By doing that, we will also make our plasma codes more energy efficient.
7. Objective VII: Create synergistic collaborations with other CoEs, EuroHPC, and Competence Centers for cross-fertilisation and supporting the significant adoption and full exploitation of the Plasma-PEPSC simulation codes, methodology, approaches and ensuring synergies in preparing training hackathons, workshops, and conferences.
In the project's first year, we identified and detailed the scientific grand challenges and associated use cases. In particular, we defined the figure of merits for each code in order to enable the solution of the challenge. These figures will allow us to monitor the progress of each grand challenge during the project. In addition, we established a performance baseline for parallel scaling, I/O performance, and accelerator usage to compare the new developments in the project. The progress towards this objective is detailed in deliverables D1.1–D1.8. We assessed parallel scalability and efficiency on different EuroHPC systems and clusters available in the consortium to provide an initial baseline for current development. We secured access to all the EuroHPC systems. We identified the necessary steps for enabling algorithmic advancements in the application in load balancing, data compression, resilience, I/O, and portable programming systems (MPI, Alpaka …). This work is detailed in D3.1 D4.1 and D5.1. We ensured that each Plasma-PEPSC developer is committed to following modern software software engineering practices, including continuous integration and automated deployment on EuroHPC systems. All the Plasma-PEPSC codes are now available as open-source through different licensing schemes and remote repositories. Currently, the four codes have different maturity levels regarding continuous integration and other software engineering aspects. The progress is detailed below when describing WP1 software engineering tasks. This objective is pursued together with the CASTIEL 2 project. Plasma-PEPSC members are routinely in the calls and contribute to CASTIEL 2 work on providing a framework for continuous integration and deployment.During the first year, we established a direct channel to EPI and a co-design methodology based on evaluating the performance of the Plasma-PEPSC applications on available hardware and software simulators for the Rhea processor and the EPAC accelerators. This effort included a meeting to present the EPI EPAC accelerator to the consortium. The first benchmarking results, together with the plasma simulation code opportunities, such as amenability to vectorization, are presented in D2.1 and D2.2.
Plasma is ubiquitous in nature and its advanced usage in devices impacts different areas of science, technology, society, and economy. In addition, plasma simulations are among the classes of simulations that clearly require exascale performance. We thus expect that the outcomes of our project will be very rapidly widespread in the plasma physics and HPC communities, in line with the ambitions of EuroHPC and with an impact well beyond EuroHPC. Expected impact are:
-Enabling the computational plasma research community to tackle their Grand Challenges on Exascale systems
- Knowledge and expertise on using and deploying extreme-scale plasma simulations for solving grand scientific challenges.
- International expertise and leadership on the use of pre-exascale and exascale systems for plasma simulations.
- Scientific publications from Plasma-PEPSC in these four scientific areas impact the advancement of knowledge in these four fields. We publish peer-reviewed articles in high-impact international conferences and journals, following the guidelines of Open Science. With the work in Plasma-PEPSC, we expect to achieve a high number of citations of our publications, an increase in the number of invited talks, and participation in scientific committees and chairs of academic gatherings, significantly impacting the scientific community.
- Improved collaboration between academia, HPC centers, and industry.
- Demonstration of the usage of pre-exascale and exascale technologies in plasma simulations.
- Promotion of recommendations for the development of the European processor and accelerator
-Enabling the computational plasma research community to tackle their Grand Challenges on Exascale systems
- Knowledge and expertise on using and deploying extreme-scale plasma simulations for solving grand scientific challenges.
- International expertise and leadership on the use of pre-exascale and exascale systems for plasma simulations.
- Scientific publications from Plasma-PEPSC in these four scientific areas impact the advancement of knowledge in these four fields. We publish peer-reviewed articles in high-impact international conferences and journals, following the guidelines of Open Science. With the work in Plasma-PEPSC, we expect to achieve a high number of citations of our publications, an increase in the number of invited talks, and participation in scientific committees and chairs of academic gatherings, significantly impacting the scientific community.
- Improved collaboration between academia, HPC centers, and industry.
- Demonstration of the usage of pre-exascale and exascale technologies in plasma simulations.
- Promotion of recommendations for the development of the European processor and accelerator