Highlights from the High-Performance Computing Seminar MetaCentrum 2025

This year’s seminar highlighted the Czech Republic's national distributed computing environment, MetaCentrum, which provides HPC, cloud services, and specialized tools (including AI support) for scientific research. The seminar featured presentations on new infrastructure developments, user-centric services, and technological trends. The event underlined MetaCentrum’s central role in Czech science and its growing integration into the European Open Science Cloud (EOSC).

The event was organized by the MetaCentrum project of the CESNET association and the CERIT-SC Center at the Institute of Computer Science of Masaryk University.

Building the Future of Czech Research Infrastructure

In the opening lecture, Miroslav Ruda presented the current status and development plans of MetaCentrum. Among the most significant updates are hardware upgrades, including GPU servers (NVIDIA H100) and DGX systems (with NVIDIA B200/B300) optimized for artificial intelligence. The storage infrastructure has been expanded with BeeGFS and Ceph systems, providing capacity even for data-intensive projects.

Equally important as the hardware are the services running on top of it. Through single sign-on, users have access to interactive environments such as Jupyter, RStudio, and Alphafold, as well as commercial tools like MATLAB and Ansys. Platforms such as Galaxy, Kubernetes, and OnDemand are also available, enabling easier execution of reproducible workflows, orchestration of containerized tasks, and efficient scaling of experiments.

Currently, MetaCentrum supports over 3,500 active users, manages more than 53,000 CPU cores, over 500 GPUs, and 20 PB of storage. Thanks to compliance with European standards and connections to international communities (e.g., EOSC, Galaxy, and ELIXIR), Czech researchers can operate at a global level while also having access to national computing resources and support.

Preserving Scientific Data: The National Repository Platform

The challenge of long-term data preservation was addressed by David Antoš, who introduced the National Repository Platform (NRP). Unlike short-term storage systems, the NRP is designed as a permanent, certifiable home for research datasets. By adhering to FAIR principles—Findable, Accessible, Interoperable, and Reusable—the platform guarantees that scientific data remains accessible and transparent for decades.
Through services like the National Metadata Catalogue (nma.eosc.cz), researchers can search across repositories, while specialized interfaces allow them to upload, annotate, and share their datasets. Pilot repositories are already live, and further domain-specific repositories will be launched in the coming year. This initiative addresses a critical gap in Czech science, ensuring valuable data does not disappear into “dark storage” but instead contributes to open, reusable knowledge.

From Genomics to Ecology: Research Enabled by HPC

User presentations illustrate the practical benefits of MetaCentrum services for modern research attracted great interest.

Tomáš Pavlík from Masaryk University described genomic research focused on assembling complex genomes and annotating DNA sequences to advance personalized medicine. Projects from his group have consumed hundreds of CPU-years and tens of terabytes of storage—resources that would not be so readily accessible without the national HPC infrastructure. Pavlík also pointed out the challenges faced by newcomers when using the infrastructure, especially students without Linux experience, and emphasized the need for interactive tutorials and targeted training for young researchers.

Jan Filip from Charles University presented an exceptional ecological project supported by the GAČR Junior Star grant. His team uses AI models trained on MetaCentrum GPU nodes to analyze billions of video recordings from long-term field monitoring. The goal is to detect and quantify interactions between insects and plants—including nocturnal pollination, which traditional methods have overlooked. Thanks to parallel computing and containerized workflows, the project has already processed hundreds of thousands of videos, drastically shortening processing time that would otherwise take decades, and produced usable ecological data within just a few months. The team thus gained unprecedented insights into global biodiversity.

Aleš Křenek and Adrian Rošinec introduced MDDash, a virtual research environment for molecular dynamics simulations. As the number of simulations grows, so does the need for specialized repositories that enable researchers to search and reuse stored data. Tools such as MetaDump for metadata extraction and BioID for automatic protein identification bring order to what was previously regarded as “dark data.” The establishment of a national MD repository integrated into the EOSC will provide high-capacity storage and ensure that Czech contributions are visible and reusable in global bioinformatics.

Afternoon Focus: Security, Platforms, and Cloud

The seminar’s second half shifted to technological trends and user--oriented services.

Dan Kouřil emphasized that MetaCentrum’s openness does not come at the expense of security. Using a real example from September 2025, he demonstrated how a compromised account was detected within minutes and swiftly resolved through coordinated action. This case illustrates well the balance between flexibility and security—crucial for an infrastructure used by several thousand researchers.

Aleš Křenek dispelled the myth that the Galaxy platform is limited to genomics or education. Today, it supports workflows in areas such as natural language processing, climatology, and astronomy. Thanks to the integration of usegalaxy.eu with usegalaxy.cz within MetaCentrum, Czech researchers gain higher computing quotas, access to GPUs, and direct connections to national data repositories. Its open architecture and emphasis on reproducibility make Galaxy an indispensable tool of modern science.

Anežka Melounová demonstrated how accessible and flexible the Jupyter environment in MetaCentrum is—from teaching to demanding HPC and machine learning projects. Users can choose Jupyter notebooks via the OnDemand interface or in Kubernetes, with access to modern tools such as AI assistants, GPU acceleration, and integrated storage.

Lukáš Hejtmánek followed with two presentations focused on the practical use of Kubernetes and the expansion of AI-related services. In the first part, he introduced the latest developments in Kubernetes. The Kubernetes infrastructure now includes three instances: a general public one, a secure one for sensitive data, and a new one with NRP support. Various web services also run in Kubernetes, such as Foldify, CryoSPARC, and JupyterHub.

In his second talk, Hejtmánek presented the latest AI activities. The new AI Chat platform (chat.ai.e-infra.cz) offers an environment similar to ChatGPT, with locally running models optimized for logical reasoning, Czech language, and programming, including features for search, image generation, and API integration. It is complemented by the AI Web Builder (DeepSite) for automatic website creation, a documentation chatbot, and AI assistance in Jupyter Notebook or RStudio.

Klára Moravcová presented the MetaCentrum Cloud as a flexible self-service environment that enables researchers to launch virtual machines, configure networks, or deploy Kubernetes clusters directly from the OnDemand interface. The cloud covers everything from personal sandboxes to large team projects, extending HPC capabilities with a user-friendly platform. Among the upcoming features are FastStack instances with next-generation networking and NVMe storage, promising even higher performance.

Listening to the Community: Survey Insights

The seminar concluded with Jiří Vorel, who presented the mission of MetaCentrum and, using feedback from the user survey, outlined the main principles of working with MetaCentrum. The 2024/25 satisfaction survey collected nearly 300 responses. While researchers appreciate the scope and benefits of MetaCentrum, they also pointed out shortcomings in documentation, the complexity of accessing GPUs, and challenges for beginners. In response, MetaCentrum is revising its tutorials, providing clearer information on GPU usage, and launching an Ambassador Program to offer local user support.

HPC Seminar's web page (including presentations and videorecording)