The SEA Projects: Three-Year Effort for European Exascale Supercomputing
After three years of research and development, the EuroHPC JU projects DEEP-SEA, IO-SEA, and RED-SEA were recently finalised with great success. All three projects (collectively known as the SEA projects) have developed and enhanced essential technologies for highly efficient, scalable heterogeneous Exascale supercomputers, in particular those based on the Modular Supercomputing Architecture (MSA).
The SEA projects have passed their respective final reviews exceptionally well, receiving excellent feedback from the EuroHPC JU and external experts. Beyond their individual work, they collaborated closely on creating efficient systems that enable the development of highly efficient heterogeneous and MSA systems. The work will now continue in upcoming projects and initiatives driven by the project partners.
Key technologies for future supercomputing
The upcoming Exascale generation, expected to perform at least 1018 floating-point operations per second, will revolutionize scientific applications and research. The first such system in Europe will be JUPITER, soon to be installed at JSC. The need for very high energy efficiency in Exascale supercomputers requires the integration of various, diverse hardware components, including CPUs, GPUs, and other accelerators, as well as communication, memory, and storage technologies. Supporting such heterogeneous systems is precisely the area of focus for the SEA projects.
Collaborating on system software, programming environments, data management and storage as well as interconnects, the SEA projects advanced key technologies required for the future of supercomputing.
DEEP-SEA, led by JSC, focused on integrating an efficient HPC software stack: the result combines “best of breed” programming models, APIs, tools and libraries which are also relevant for non-modular HPC systems. In particular, the innovative optimisation cycles, the support of hierarchical storage systems and the use of modern CI/CD techniques will bring direct benefits for such systems. This also applies to the pioneering work in the area of malleability. The software stack is freely available as open source to all interested parties.
IO-SEA, headed by the Commissariat à l'Energie Atomique et aux Applications (CEA), developed a storage and I/O stack suitable for the MSA that deals with data movements across hierarchical storage systems including novel non-volatile memory, solid state devices, hard disk drives and tape storage. IO-SEA also developed the concept of short lived “ephemeral storage services” that are created on demand and enable applications to achieve high performance and energy efficiency for their I/O operations. The project also introduced extensive use of instrumentation and telemetry across the I/O stack and new I/O APIs that expose the actual semantics of data.
RED-SEA, spearheaded by Eviden, has laid the groundwork for the development and industrialization of the third generation of the European Bull eXascale Interconnect (BXI). The project helped to bring BXI to the Exascale level, with improved scalability and resilience to reach well over 100,000 end-points, while significantly improving the latency and the bandwidth available for each end-point. RED-SEA also addressed real-world network performance with congestion management and adaptive routing mechanisms. Finally, RED-SEA opened up BXI to Ethernet compatibility, by designing a seamless interface between BXI and Ethernet via a new low-latency Gateway solution.
The SEA projects have received funding from the European High-Performance Computing Joint Undertaking (JU) under grant agreements n° 955606, 95811, and, 955776 and support from Belgium, the Czech Republic, France, Germany, Greece, Ireland, Italy, Spain, Sweden and Switzerland.
Contact: e.suarez@fz-juelich.de and h.hoppe@fz-juelich.de