ParTec

Research Projects:
Driving progress, shaping the future

ParTec participates in a variety of German- and EU-funded research and innovation projects in Quantum Computing, HPC, and Artificial Intelligence. These projects focus on developing technologies that enable efficient use of leading-edge, heterogeneous systems up to and including Exascale levels of performance.

ParTec’s R&D activity is focused on further enhancing the dMSA, and, using and extending ParTec’s ParaStation Modulo Software Suite to new environments to support novel use cases.

ParTec contributes to the development of new software architectures, to the development of runtime systems, middleware components, and management tools for production-ready, heterogeneous systems. In doing so, it puts the focus on improvements and extensions of the scheduling and resource management subsystems for the dynamic Modular Supercomputing Architecture (dMSA) as well as programming models, libraries, and management tools.

In addition, ParTec also performs project management functions in some of the projects.

Partners

At present ParTec is working with a large number of research partners in the active research and innovation projects described below. The map of Europe indicates the location of these partners, with orange showing cities with just one partner, and blue indicating the presence of multiple partners at the same location.

High-Performance Computing (HPC) Research Projects

(December 2011 – March 2021)

The Dynamical Exascale Entry Platform (DEEP) projects, Dynamical Exascale Entry Platform (DEEP), Dynamical Exascale Entry Platform – Extended Reach (DEEP-ER), Dynamical Exascale Entry Platform – Extreme Scale Technologies (DEEP-EST),  developed a novel, flexible, and energy-efficient Exascale-enabling supercomputing platform and ported/optimised a set of key European HPC simulation codes. The original DEEP System realised a Cluster-Booster Architecture that addressed the limitations imposed by Amdahl‘s Law by further developing and deploying ParTec’s ParaStation Modulo SW suite. DEEP-ER extended this Cluster-Booster architecture by a highly scalable I/O system and implemented an efficient mechanism to recover application tasks that fail due to hardware errors. The DEEP-EST project created the first incarnation of the dynamic Modular Supercomputing Architecture (dMSA) and demonstrated its benefits using ParaStation Modulo. ParTec was a key contributor to all three projects focusing on SW engineering for ParaStation Modulo, creating the runtime and middleware of the DEEP Cluster-Booster system and the DEEP-EST dMSA, and adding modules for multi-level, application-based fault tolerance.

Funding

DEEP projects have received funding from the European Union’s Framework 7 and Horizon 2020 research and innovation programmes under grant agreement No. 287530 (DEEP), No. 610476 (DEEP-ER), and No. 754304 (DEEP-EST).

(April 2021 – March 2024)

DEEP-SEA delivered an integrated programming environment for future European Exascale systems, adapting all levels of the software stack – including low-level drivers, computation and communication libraries, resource management, and programming abstractions with associated runtime systems and tools. This stack enables dynamic resource allocation, application malleability, programming model composability, and included tools for an efficient mapping of applications to the dMSA. The result is a SW environment that enables applications to run on the best suited hardware in a scalable and energy efficient manner. The DEEP-SEA software elements were designed and validated in a collaborative co-design approach with key EU HPC and AI applications. ParTec provided central elements of the DEEP-SEA stack by extending ParaStation Modulo with respect to malleability, the scheduling of dynamic workflows, and the execution of containerised HPC workloads. Additionally, ParTec significantly contributed to a semi-automated tuning of the runtime, and addressed interoperability and composability aspects of ParaStation MPI with other programming paradigms.

The project activities have received funding from the European High Performance Computing Joint Undertaking (EuroHPC JU) under grant agreement No 955606 (DEEP-SEA), No. 955811 (IO-SEA), No. 955776 (RED-SEA), No. 101033975 (EUPEX) and No. 101036168 (EPI SGA2). The JU receives support from the European Union’s Horizon 2020 or Horizon Europe research and innovation programme and from national funding agencies, in the case of ParTec from the German ministry of education and research (BMBF).

IO Software for Exascale Architecture (April 2021 – March 2024)

IO-SEA developed a novel data management and storage platform for Exascale computing based on hierarchical storage management (HSM) and on-demand, adaptive provisioning of storage services. The platform made effective use of multiple storage tiers, including non-volatile memory express (NVMe) and non-volatile random-access memory (NVRAM), all the way down to tape-based technologies. The use of ephemeral data nodes and data accessors enabled users to flexibly allocate I/O resources as part of complex application workflows and to use them via well-known data access paradigms. ParTec enhanced its ParaStation MPI Resource Management to support ephemeral data services and contributed a comprehensive solution for fault detection and preventative maintenance by enhancing the ParaStation HealthChecker.  The project targeted dMSA systems and uses the DEEP-SEA development system.

The project activities have received funding from the European High Performance Computing Joint Undertaking (EuroHPC JU) under grant agreement No 955606 (DEEP-SEA), No. 955811 (IO-SEA), No. 955776 (RED-SEA), No. 101033975 (EUPEX) and No. 101036168 (EPI SGA2). The JU receives support from the European Union’s Horizon 2020 or Horizon Europe research and innovation programme and from national funding agencies, in the case of ParTec from the German ministry of education and research (BMBF).

Network Solution for Exascale Architectures (April 2021 – March 2024)

The internal high-speed network enables parallel applications to scale efficiently to the Exascale level and beyond and thus is a critical architectural element of massively parallel supercomputers. The RED-SEA project leverages key European technology to prototype a European interconnect with accompanying SW stack that is scalable to Exascale levels and supports heterogeneous accelerators plus compute units as well as data-centric and AI-related applications in a highly efficient way. In the RED-SEA project, ParTec enhanced, extended and optimised ParaStation MPI for the Bull eXascale Interconnect (BXI).

The project activities have received funding from the European High Performance Computing Joint Undertaking (EuroHPC JU) under grant agreement No 955606 (DEEP-SEA), No. 955811 (IO-SEA), No. 955776 (RED-SEA), No. 101033975 (EUPEX) and No. 101036168 (EPI SGA2). The JU receives support from the European Union’s Horizon 2020 or Horizon Europe research and innovation programme and from national funding agencies, in the case of ParTec from the German ministry of education and research (BMBF).

European Pilot for Exascale (started January 2022)

The EUPEX consortium designs, builds, and validates the first native EU technology platform for HPC, leveraging a wide spectrum of European assets end-to-end from system architecture based on the dMSA, via processors, system software, development tools to the applications. The EUPEX prototype is designed to be open, scalable, and flexible, including the modular OpenSequana-compliant platform and the corresponding HPC software ecosystem for systems being built by following the paradigms of the dynamic Modular Supercomputing Architecture (dMSA). The prototype will be installed at CEA and operated by this institution. EUPEX is also an important vehicle for preparing the European HPC, AI, and Big Data processing communities for the upcoming European Exascale systems and technologies. With the ParaStation Modulo Software Suite, ParTec provides the dMSA-enabling system SW for the management and operation of large-scale, modular platforms, and develops and optimises it to fully leverage the potential of European SW and SW technologies.

The project activities have received funding from the European High Performance Computing Joint Undertaking (EuroHPC JU) under grant agreement No 955606 (DEEP-SEA), No. 955811 (IO-SEA), No. 955776 (RED-SEA), No. 101033975 (EUPEX) and No. 101036168 (EPI SGA2). The JU receives support from the European Union’s Horizon 2020 or Horizon Europe research and innovation programme and from national funding agencies, in the case of ParTec from the German ministry of education and research (BMBF).

European Processor Initiative specific Grant Agreement 2 (Started January 2022)

The SGA2 is the second step towards the implementation of the European Processor Initiative. The project has devised a roadmap for future European low power processors and the accompanying SW ecosystem. It develops a line of European low-power processor and accelerator technologies to be used in highly efficient HPC servers up to the Exascale level. By doing this, the project enhances Europe’s digital sovereignty via fostering the creation of a European ecosystem of people, products and full-stack solutions, from the processor and accelerator technologies to the applications and software ecosystem. In SGA2, ParTec enhances and optimises its ParaStation Modulo Software Suite for supporting the SiPearl RHEA CPU prototype and the corresponding platform planned for 2025 availability.

The project activities have received funding from the European High Performance Computing Joint Undertaking (EuroHPC JU) under grant agreement No 955606 (DEEP-SEA), No. 955811 (IO-SEA), No. 955776 (RED-SEA), No. 101033975 (EUPEX) and No. 101036168 (EPI SGA2). The JU receives support from the European Union’s Horizon 2020 or Horizon Europe research and innovation programme and from national funding agencies, in the case of ParTec from the German ministry of education and research (BMBF).

(Started December 2022)

DaREXA-F develops innovative methods for reducing the data volume which is exchanged between compute nodes of HPC systems or stored on parallel file systems.

Following a co-design approach, the GENE Plasma turbulence simulation code will be extended to use variable precision arithmetic, data compression and altogether new data formats, and the improvements of these approaches regarding data volumes, execution time and energy use will be validated.

A successful DaREXA-F project can significantly advance the European ITER experiment and fusion research in general. ParTec contributes its expertise in communication libraries and create experimental data reduction extensions to the ParaStation MPI library.

DaREXA-F

ParTec’s activities in the DaREXA-F, EECliPs, IFCES-2, MExMeMo and QSOLID projects receive funding from the German ministry of education and research (BMBF).

(Started September 2022)

The EECliPs project explores the potential of significantly reducing the energy use of climate simulations via the use of advanced hardware systems and software stacks. After systematically exploring the power and energy consumption behavior of the ICON climate and weather model on a wide range of heterogenous systems, the project will define and install a small proof-of-concept system with the most promising, leading-edge heterogeneous architectures. This system will serve to measure and optimize the ICON code both for minimum execution time and minimum energy used to compute a solution, and to assess how the latest heterogeneous (and therefore specialized) hardware can be best integrated with the DKRZ data center. ParTec advises the project in the selection of leading-edge heterogeneous hardware components and will support the installation and operation of the proof-of-concept system.

EECliPs

 

ParTec’s activities in the DaREXA-F, EECliPs, IFCES-2, MExMeMo and QSOLID projects receive funding from the German ministry of education and research (BMBF).

(Started October 2022)

IFCES-2 explores innovative methods for optimizing the parallel execution of the ICON earth system model on heterogeneous HPC systems, with the objective of efficient scaling to the Exascale level.

This includes raising the general degree of parallelism in the model components and workflow, improving the communication between the model components, and using dynamic load balancing techniques. ParTec contributes to experiments with functional parallelization and will adapt and optimize its ParaStation MPI library to support code coupling and dynamic load balance measures.

IFCES-2

ParTec’s activities in the DaREXA-F, EECliPs, IFCES-2, MExMeMo and QSOLID projects receive funding from the German ministry of education and research (BMBF).

(Started September 2022)

MExMeMo develops a novel multi-scale model for the manufacturing of soft materials. These pose major challenges for HPC, since several physical processes are coupled at very different scales and the associated simulation codes place very different demands on the hardware. On conventional, even GPU-accelerated systems, such a coupled simulation will run with low efficiency.

The MExMeMo code will fully leverage the MSA, support a wide range of spatial and temporal scales, and map them to the best suitable compute resources on a dynamic MSA system. The objective is to show good scaling up to the Exascale level.

ParTec contributes its unmatched experience with MSA systems and will adapt/extend its MPI implementation and system management suite to support the extraordinary requirements of the MExMeMo code. Additionally, ParTec leads the design and the development of a coordinator enabling the efficient scheduling of such tightly coupled simulations on MSA systems.

ParTec’s activities in the DaREXA-F, EECliPs, IFCES-2, MExMeMo and QSOLID projects receive funding from the German ministry of education and research (BMBF).

(Started October 2022)

Destination Earth develops a highly accurate “digital twin” model of the Earth to monitor and predict environmental change and human impact.

It will help policy-makers to monitor and simulate the Earth’s system developments and human interventions, anticipate environmental disasters and resultant socio-economic crises to save lives and avoid large economic downturns.

Furthermore, the digital twin will enable the development and testing of scenarios for ever more sustainable development. HPC systems co-funded by the EuroHPC JU will be used to run the Destination Earth software.

A consortium of three leading European research institutions European Centre for Medium Range Weather Forecasts, European Organisation for the Exploitation of Meteorological Satellites, and European Space Agency) coordinates this major European Commission initiative. ParTec experts are contributing as sub-contractors to the European Technology Platform for High-Performance Computing to a technology roadmap which will inform the Destination Earth implementation phases.

 

Destination Earth is a European Union funded initiative and is implemented by ECMWF, ESA, and EUMETSAT.

Quantum Computing (QC) Research Projects

High Performance Computer and Quantum Simulator hybrid (Started December 2021)

HPCQS is preparing European research, industry, and society for using and operating federations of Quantum Computers and simulators. HPCQS develops the programming platform for a Quantum simulator based on the Eviden Qaptiva, and the deep, low-latency integration of Quantum Computers into HPC systems based on the dMSA. A twin pilot system, developed as a prototype by the French company Pasqal, is being implemented and integrated at CEA/TGCC (France) and FZJ/JSC (Germany), the hosts of the first two European Exascale systems. The HPCQS technology is co-designed with selected exemplary use cases from chemistry, physics, optimisation, and machine learning suitable for quantum-hybrid HPC simulations. ParTec integrates the developed integration SW with its ParaStation Modulo SW suite and achieve seamless QC/HPC systems integration.

The project activities have received funding from the European High Performance Computing Joint Undertaking (EuroHPC JU) under grant agreement No 955606 (DEEP-SEA), No. 955811 (IO-SEA), No. 955776 (RED-SEA), No. 101033975 (EUPEX) and No. 101036168 (EPI SGA2). The JU receives support from the European Union’s Horizon 2020 or Horizon Europe research and innovation programme and from national funding agencies, in the case of ParTec from the German ministry of education and research (BMBF).

Quantum Computer in the Solid State (Started January 2022)

25 leading German companies and research institutions have joined forces in the collaborative project QSolid. Together, the project partners develop a comprehensive ecosystem for a demonstrator based on superconducting qubits. The project focusses on qubits of very high quality, i.e., with a low error rate. The Quantum Computer will be integrated  into JSC’s supercomputing infrastructure at an early stage and will contain several next-generation superconducting quantum processors, including a “moonshot” system with computing power exceeding that of conventional supercomputers, an application-specific system designed to perform quantum calculations for industry, and a benchmarking platform that prioritises the development of digital twins and industry standards. The first demonstrator will go into operation end of 2024 and enable testing applications and industry standard benchmarks. In QSolid, ParTec contributes to the HPC integration of the Quantum Computer, extending the process management and health checking components of the ParaStation Modulo SW suite to guarantee efficient management, allocation, and utilisation of quantum resources in dMSA systems.

ParTec’s activities in the DaREXA-F, EECliPs, IFCES-2, MExMeMo and QSOLID projects receive funding from the German ministry of education and research (BMBF).

Artificial Intelligence (AI) Research Projects

Center of Excellence “Research on AI- and Simulation-Based Engineering at Exascale” (January 2021 – June 2024)

CoE RAISE enabled complex applications in industry and academia that combine Artificial Intelligence and HPC technologies at the Exascale level. It supported the development of entirely new multi-physics and/or multi-scale big data workflows that efficiently run on todays and future HPC architectures.

In addition, RAISE integrated existing AI SW technology and optimized it for efficient use of European HPC systems and scaling up to Exascale.

ParTec supported the coordinator Jülich Supercomputing Centre in all project management tasks, advised the project on IP and legal organization topics, and,  contributed its software expertise for exploiting heterogeneous supercomputers built with dMSA.

CoE RAISE and the previous DEEP projects have received funding from the European Union’s Framework 7 and Horizon 2020 research and innovation programmes under grant agreement No. 951733 (CoE RAISE), No. 287530 (DEEP), No. 610476 (DEEP-ER), and No. 754304 (DEEP-EST).