For more than ten years, the DEEP projects have been at the forefront of research into the next generation of supercomputers. Today, Exascale computing power has become a reality: the US system FRONTIER has entered the Top 500 list at number 1 in June 2022 with 1.1 Exaflop/s on 9472 compute nodes with almost 40 thousand GPGPU accelerators. The first European Exascale computer JUPITER (Joint Undertaking Pioneer for Innovative and Transformative Exascale Research) will be hosted at the Jülich Supercomputing Centre (JSC) from 2023 on. The system will be acquired by the European supercomputing initiative EuroHPC JU. The overall costs for the system amount to 500 million euros. Of this total, 250 million euros is being provided by EuroHPC JU and a further 250 million euros in equal parts by the German Federal Ministry of Education and Research (BMBF) and the Ministry of Culture and Science of the State of North Rhine-Westphalia (MKW NRW).
At the Exascale Day we had the opportunity to talk to Prof. Dr. Estela Suarez, project leader of the DEEP-SEA project. Dr. Suarez joined the JSC and later the DEEP projects in 2010. She is an expert in heterogeneous HPC system architectures and holds a chair in High Performance Computing at the University of Bonn.
Estela, the installation of JUPITER at JSC is expected for 2023. How much of the DEEP projects is reflected in JUPITER?
The DEEP projects have developed the underlying concept for JUPITER, the Modular Supercomputing Architecture or MSA. In the DEEP projects, we first built hardware prototypes comprised by several, partially specialized modules. To demonstrate that the MSA concept works, we then developed various layers for a software stack that allows efficiently using this kind of supercomputer modules. Either each individually, like on classical supercomputers, or multiple of them working together, to solve more complex problems.
What distinguishes the MSA from other approaches?
The MSA is a particular way of combining different types of processors (CPUs, GPUs, other accelerators): they are grouped in separate, interconnected compute modules. This gives maximum flexibility in the selection of hardware resources to be used by a particular application, matching the specific requirements of its code. This way, MSA can support very diverse applications, offering to each the best suited set of hardware resources, and ensures that highly efficient use of the overall machine.
Will the work on Exascale be done with the installation of JUPITER? What would still need to be done?
JUPITER will bring huge computational power and, as a production machine, aims at maximum availability, stability, and security. However, as with any system of this size and complexity, extracting maximum performance requires very advanced features in the software stack and a deep understanding of the system’s capabilities by the users. Within the DEEP-SEA project we are enhancing tools and software packages to make them more dynamic and at the same time make the life of our users easier.
As part of the system software, we are also developing flexible schedulers, resource managers and programming environments to dynamically reserve and release resources during the lifetime of an application. In other words, we are making the applications and HPC systems more “malleable” and helping them to achieve maximum performance and efficiency. On the user-side, applications would be able to flexibly react to changing resources, and could also initiate such changes.
How do future supercomputers benefit from other European Exascale projects like RED-SEA and IO-SEA?
IO-SEA and RED-SEA work on topics which are largely orthogonal to DEEP-SEA: high-performance, intelligent I/O software stacks and storage systems, and next-generation interconnect fabrics with their software stacks. These results, together with those in DEEP-SEA and other EuroHPC projects, will ultimately become part of the software stack running on future European supercomputers, improving their operation and utilisation.
What do you expect from JUPITER? How will society in Europe and beyond benefit from the system, considering the huge computing power and the significant power requirements?
JUPITER is above all a scientific instrument. In the same way as larger telescopes allow to see deeper and further into the Universe, European researchers will use JUPITER to study larger and more complex problems that cannot be tackled today. In doing so, these researchers will achieve new scientific results in areas as important as medicine, climate and environmental research, or materials science, to name just a few.
Society will profit from the scientific results created by JUPITER in many different ways. Examples are: new therapies, drugs and vaccines developed based on simulations of how specific treatments affect the human body and its basic building blocks (e.g. proteins); new policies and mitigation strategies based on much better understanding of climate change and its consequences; advanced materials that make better and more energy-efficient products.
It is true that systems like JUPITER consume significant amounts of electrical power. Therefore, JSC undertakes to make JUPITER’s operation as environmentally friendly as possible. It will run with green energy, use free cooling and plans for waste heat usage are being made. On the software side, the research in the DEEP-SEA, IO-SEA and RED-SEA projects will contribute to make the JUPITER system as energy-efficient as possible. At the same time, applications will progress and strive to generate results with less use of energy. Compared to conventional methods of producing scientific results (like experiments), supercomputing has been shown to be more energy-efficient, and JUPITER as a large system will be more efficient than a set of smaller systems with the same aggregated compute power.