DEEP – SOFTWARE FOR EXASCALE ARCHITECTURES

Project Information

DEEP-SEA

Grant agreement ID: 955606

DOI

10.3030/955606

Project closed

EC signature date 2 December 2020

Start date 1 April 2021

End date 31 March 2024

Funded under

• INDUSTRIAL LEADERSHIP - Leadership in enabling and industrial technologies - Information and Communication Technologies (ICT)

Total cost

€ 15 069 476,25

EU contribution

€ 7 534 738,15

Coordinated by

FORSCHUNGSZENTRUM JULICH GMBH
Germany

Periodic Reporting for period 2 - DEEP-SEA (DEEP – SOFTWARE FOR EXASCALE ARCHITECTURES)

Reporting period: 2022-10-01 to 2024-03-31

Summary of the context and overall objectives of the project

The DEEP-SEA project (“DEEP – Software for Exascale Architectures”), running from April 2021 to March 2024, developed and validated an integrated software stack for European Exascale systems. Covering low-level drivers, libraries, system management, and programming abstractions, it supports heterogeneous compute, network, and memory configurations.

By doing so, DEEP-SEA promotes efficient and sustainable use of heterogeneous HPC systems, benefiting the society through various applications like climate research, medicine, and energy exploration.

The overall objectives of the project are:
1: Co-design the software- and programming environment of the upcoming European Exascale systems (WP1).
2: Provide tools to map complex applications and non-uniform workflows onto heterogeneous and modular computer architectures (WP2).
3: Enhance the system software, programming paradigms, tools, and runtimes in order to extract the maximum performance from heterogeneous computer platforms and improve performance portability (WP3, WP4, WP5).
4: Improve the use and management of new memory technologies and the placement of data in compute devices with deep and heterogeneous memory hierarchies (WP3, WP4, WP5).
5: Release the DEEP-SEA software stack in production-ready quality to enable its use and exploitation in upcoming European Exascale systems (WP3).

Work performed from the beginning of the project to the end of the period covered by the report and main results achieved so far

The work across WP1-WP7 aligns with the project objectives as detailed below:

- WP1 addressed seven European HPC application fields, encompassing 10 codes, representing diverse multi-disciplinary uses of European HPC systems. Initial analysis of application characteristics and requirements informed the co-design process. DEEP-SEA tools, programming environments, and run-time systems addressed these requirements, with Optimisation Cycles (OCs) guiding specific analysis or optimization goals. Performance limitations were mitigated through improvements in intra- and inter-node APIs for GPUs and CPUs, dynamic load balancing, and high-level programming models. A benchmark suite measured performance and identified potential variations.

- In WP2, existing software tools for performance measurement, modeling, and monitoring were enhanced and extended with new functionalities. Significant impact was achieved by combining enhanced tools into OCs, such as mapping, monitoring, and energy optimization. The mapping OC aided developers in distributing code over heterogeneous Modular Supercomputer Architecture (MSA) systems, while the monitoring OC quantified the impact of different application mappings. Energy optimization tools operated transparently in the background to save energy.

- DEEP-SEA aimed to deliver high TRL software for deployment on upcoming European HPC systems. WP3 documented the software stack comprehensively, ensuring compatibility through continuous integration and deployment (CI/CD). The final software stack was released via EasyBuild configuration and Spack installation files repository, with containerization simplifying software installation and application execution. Defined OCs assisted application developers and users in selecting and handling software components effectively.

- WP4 improved support for different hardware accelerators and implemented mechanisms for dynamic load balancing at the node level, enhancing throughput and energy efficiency. WP5 enhanced composability and interoperability between programming models, focusing on a malleability prototype implementation for dynamic resource allocation changes. Resiliency was improved through enhancements to the Simgrid simulation framework.

- In WP6, DEEP-SEA achieved high visibility for its objectives, approach, and results through internal and external channels, including events, conferences, and social media. Joint communication with other EuroHPC projects was enhanced, and involvement in standardization bodies and training activities facilitated knowledge transfer.

- WP7 involved continuous monitoring by the Project Management Team (PMT) to ensure progress aligned with the project's Description of Action (DoA) and to identify any risks. Quality control measures included internal review processes for deliverables. Financial control involved quarterly financial reports from each partner. Project coordination activities extended beyond DEEP-SEA to cover collaboration within the SEA projects and the ten projects funded in the same EuroHPC call, facilitated by a cooperation agreement signed by all participating institutions.

Progress beyond the state of the art and expected potential impact (including the socio-economic impact and the wider societal implications of the project so far)

DEEP-SEA contributed to the EuroHPC objectives by developing, enhancing and optimising of key system software components, programming environment, tools, and applications, and by making these more mature. The project thus created a software stack for future European Exascale systems with full support for heterogeneous compute and memory system configurations, and for the MSA in particular. Driven by a consortium of industrial and academic institutions that are leaders in the European HPC landscape, DEEP-SEA strengthens Europe’s leadership, competitiveness, and innovation. European IP has been generated and developed in a sustainable way.

The overall DEEP-SEA software architecture was defined with emphasis on meeting application and user requirements and facilitating combination and interoperation of software components. These interactions are formulated through Optimisation Cycles. Existing software components were improved and enhanced with new functions, and various enhanced tools were integrated into OCs which had most significant effects.

The DEEP projects, in general and DEEP-SEA in particular, have developed concepts and technologies and specific strategies to bring holistic and high-risk-high-gain R&D projects to a successful conclusion. An important factor was the provision of hardware platforms and evaluators for early SW development. The first roll-out of the DEEP-SEA software on the DEEP system took place in March 2022, and the final rollout and public release of the software was completed towards the end of the project. The CI/CD infrastructure developed for this purpose is in operation and was an essential factor in achieving the project objectives.

DEEP-SEA project has contributed to the further development of ten important European application codes from seven research fields of high scientific and societal impact. The DEEP-SEA co-design codes are now better prepared to exploit the use of large-scale heterogeneous supercomputers in general, and modular supercomputers in particular. Using the DEEP-SEA SW stack they can better exploit heterogeneous Exascale systems.

The project delivered key elements of the software environment for future Exascale systems and provide vital contributions to get Europe ready for Exascale. DEEP-SEA, together with the IO-SEA and RED-SEA projects, is centrally positioned in the EuroHPC Exascale roadmap: its hardware-agnostic developments will benefit the European Processor Initiative (EPI), as well as the pilot projects EUPEX and EUPilot. Together with all these and further EuroHPC R&D projects, DEEP-SEA is paving the way towards the first Exascale systems in Europe.

deepsea-logo.png