MareNostrum

MareNostrum
Name	MareNostrum
Caption	Supercomputer at the Barcelona Supercomputing Center
Manufacturer	IBM
Introduced	2004
Location	Barcelona, Spain
Operators	Barcelona Supercomputing Center
Os	Linux
Peak performance	Varied by generation (TFLOPS to PFLOPS)
Memory	Varied by generation
Storage	Varied by generation

MareNostrum is a family of high-performance computing installations hosted at the Barcelona Supercomputing Center in Barcelona, Catalonia, Spain. Conceived as a national flagship for computational science, the system has served as a focal point for collaborations with European Union research initiatives, Spanish National Research Council, and international partners such as IBM and Intel. Its deployments have intersected with projects funded by the European Commission, coordinated via frameworks like the Framework Programme (EU) and Horizon 2020.

History

MareNostrum originated as a project within the Centro Nacional de Supercomputación hosted by the Polytechnic University of Catalonia and later consolidated under the Barcelona Supercomputing Center after organizational restructuring. Early procurement involved vendors such as IBM and equipment drawn from global supply chains linked to firms including Intel Corporation and Cray Inc.; subsequent generations responded to shifting procurement policies from the Spanish government and directives from the Ministry of Science and Innovation (Spain). The machine’s evolution paralleled European initiatives including PRACE and collaborations with institutions like CERN, EMBL, and the European Space Agency. Periodic upgrades reflected advances from the TOP500 ranking era through exascale planning influenced by projects such as EuroHPC Joint Undertaking.

Technical Specifications

Each MareNostrum generation combined multicore processors, interconnect fabrics, and storage subsystems sourced from vendors like IBM, Intel, NVIDIA, and HPE. Early generations used processors such as the IBM PowerPC 970 and later moved to Intel Xeon families and accelerator technologies like NVIDIA Tesla GPUs. Networking relied on high-speed fabrics including InfiniBand and proprietary interconnects developed with partners such as Mellanox Technologies. Storage architectures integrated parallel filesystems like Lustre and large-scale disk arrays with backup and archival tiers interoperable with research data infrastructures such as EUDAT and PANGAEA.

Architecture and Performance

MareNostrum architectures employed massively parallel designs combining thousands of compute cores arranged in clusters, linked by low-latency fabrics and hierarchical storage. System topologies incorporated dense compute nodes, GPU-accelerated partitions, and specialized I/O nodes to serve workloads from groups like Barcelona Supercomputing Center users, Spanish National Bioinformatics Institute researchers, and EU consortia. Performance metrics varied by generation, with peak FLOPS measured against benchmarks established by LINPACK and rankings such as TOP500 and Green500. Cooling and power infrastructure integrated technologies used in data centers operated by organizations including Red Eléctrica de España and regional utilities, while environmental assessments referenced standards from entities like European Committee for Standardization.

Applications and Research Contributions

MareNostrum supported multidisciplinary research across climate science with groups like IPCC contributors, computational biology collaborating with European Molecular Biology Laboratory, and astrophysics teams connected to European Southern Observatory initiatives. Notable scientific domains included numerical weather prediction partnering with MeteoSwiss-style institutes, genomics aligning with Genome Project consortia, and materials science studies linked to European Materials Modelling Council. The platform enabled simulations relevant to Horizon 2020 projects, facilitated workflows in data-intensive disciplines using tools developed by communities such as Apache Hadoop contributors, and supported software stacks from scientific software projects like OpenFOAM and GROMACS.

Operational Management and Funding

Operational management combined academic governance at the Barcelona Supercomputing Center with funding from national bodies such as the Spanish Ministry of Science and Innovation and European funding instruments including Horizon 2020 and European Regional Development Fund. Access policies aligned with transnational infrastructures like PRACE and allocation committees including representatives from institutions such as the Spanish National Research Council. Maintenance contracts involved commercial partners such as IBM, HPE, and Intel, while energy procurement and sustainability planning referenced regional authorities including the Government of Catalonia and utilities such as Naturgy.

Notable Deployments and Upgrades

Key deployments included the original 2004 installation, subsequent refreshes that incorporated Intel Xeon Phi accelerators and NVIDIA GPU partitions, and installations timed to align with tranche funding from the European Commission. Upgrades frequently coincided with participation in continental projects such as PRACE and the EuroHPC Joint Undertaking, enabling research campaigns in partnership with CEA laboratories and the Max Planck Society. Demonstrator uses comprised climate ensembles serving European Centre for Medium-Range Weather Forecasts-aligned studies, bioinformatics pipelines used by European Bioinformatics Institute projects, and computational chemistry calculations associated with Max Planck Institute for Chemistry collaborations.

Category:Supercomputers Category:Barcelona Supercomputing Center Category:High-performance computing facilities