Skip to main content

Programming Models and I/O with DEEP-SEA and IO-SEA

Stefano Markidis, CST, KTH

The High-Performance Computing (HPC) group at the KTH Royal Institute of Technology (KTH) does ground-breaking research on programming models and computer architectures to help build the road to exascale supercomputing. Late last year, the group (which is led by Stefano Markidis) successfully completed three high-profile exascale-related projects: EPiGRAM-HS, SAGE2, and VESTEC. All three of those projects were funded by the European Commission. This year the group started two new projects, called DEEP-SEA (Dynamical Exascale Entry Platform) and IO-SEA (Storage I/O and Data Management for Exascale Architectures), which are funded by EuroHPC and the Swedish Research Council (VR).

The three earlier projects paved the way for the work currently being carried out in the new projects. The EPiGRAM-HS project focused on extending and developing the Message Passing Interface (MPI) and GPI-2 programming systems on heterogeneous systems with graphics processing units (GPUs) and Field Programmable Gate Arrays (FPGAs). [GPI-2 is the second generation of the Global address space Programming Interface (GPI) model that was developed to help overcome limitations with using MPI for increasingly large systems.] SAGE2 investigated delivering HPC object storage for exascale systems that would provide efficient I/O on large-scale systems. VESTEC focused on interactive supercomputing enabling in-situ visualisation and data analytics for decision makers.

The two new projects continue European efforts towards developing and implementing a European software stack, including programming models and efficient I/O. DEEP-SEA, which is led by the Jülich Supercomputing Centre in Germany, will provide computation and communication libraries, resource management, and programming abstractions with run-time systems and tools. DEEP-SEA targets the Modular Supercomputing Architecture (MSA) where different components, like standard CPUs, and accelerators, such as GPUs, constitute a complex grid of heterogeneous technologies. IO-SEA, which is led by the French Alternative Energies and Atomic Energy Commission (CEA ), will deliver an innovative data management and storage framework based on hierarchical storage management and on-demand provisioning of storage services. The KTH HPC group is contributing to these two projects with the development of programming systems (such as MPI and HDF5, which is the fifth version of the Hierarchical Data Format), DaCe (a parallel programming framework that maps code to high-performance CPU, GPU, and FPGA programs) and the demonstration of applications and use cases.