This year, several improvements have been made to optimize the space and infrastructure. A new area has been set up in the room, new racks with their corresponding cabling have been installed, and smart PDUs equipped with temperature and humidity sensors have been incorporated for better environmental control. Additionally, a fire suppression system has been integrated to enhance the safety of the facilities.
A major Operating System upgrade has been successfully completed, including the migration of approximately 400 servers to AlmaLinux 9. Additionally, a monitoring system compatible with EL9 has been implemented, and comprehensive support has been provided to various projects throughout the migration process. Puppet modules, which automate operating system management and were not compatible with EL9, have been modified or replaced. Furthermore, the installation system has been adapted to ensure a seamless and efficient transition.
After more than a year of development, Shepherd, our in-house Hadoop distribution, has been deployed to the production cluster. A key achievement this year was the successful implementation of a backfilling mechanism for our Hadoop cluster, enabling the execution of jobs originating from HTCondor. This integration effectively increased the pool of available computational resources and broadened the potential applications of this powerful infrastructure. Importantly, this expansion was achieved without compromising the quality of service for our core services, Cosmohub and SciPIC. At the end of the year, the cluster was further expanded with 12 additional nodes. In its final configuration, the PIC Big Data Service is comprised of 4 head nodes and 32 storage/processing nodes, with an aggregate capacity of 768 CPUs, 16 TiB of Ram, 60 TiB of NVMe and 4 PiB of disk storage.
In 2024 the Rucio jupyterlab extension was adopted as an official Rucio Project. PIC assumed the leadership of the component development and contributed to the implementation of different features like building a test environment, the migration to jupyterlab 4 or DID filtering by metadata attributes. The PIC Rucio client was completely reformatted and the new version is now used for the file discovery and transfer triggering for MAGIC.
The environment supporting the jupyter notebooks service was updated to the latest versions available. The deployment strategy was modified, encapsulating PIC’s customizations into a python module to promote better maintainability and evolution of the service. Significant effort was dedicated to exploring and facilitating the integration of Dask within our JupyterHub environment. We tested and documented this integration, ensuring a seamless experience for users seeking to accelerate their interactive analysis of large datasets. Furthermore, we actively supported several research groups interested in leveraging Dask’s parallel computing capabilities.
By the end of 2024, the resources deployed by PIC for LHC computing amounted to approximately 121 kHS06 (equivalent to around 9,000 CPU cores), ~18.3 PB of disk storage, and ~39 PB of tape storage. A defining feature of Tier-1 centres, beyond their substantial storage and computing capacity, is the ability to deliver these resources through highly reliable services. As such, critical services at a Tier-1 site operate on a 24/7 basis. In 2024, PIC ranked at the top of the WLCG’s stability and reliability metrics.
In addition to its role in providing computing resources to the WLCG, the PIC team has been actively involved in R&D across various LHC experiments. These efforts are crucial for evolving the infrastructure to keep pace with the increasing scale and complexity of the LHC scientific programme and to prepare for the upcoming HL-LHC phase. The team has made significant contributions to conferences and publications in computing, particularly in areas such as the ongoing utilization of HPC resources at the Barcelona Supercomputing Center, the commissioning and use of the new MareNostrum 5 resources, the development and testing of new services for Analysis Facilities, and the assessment of the potential benefits of integrating data caches into the WLCG.
Sustainability has also been a key focus, with ongoing studies aimed at aligning PIC’s compute farm usage with external factors such as green energy availability and lower electricity price periods, ultimately reducing the overall CO₂ footprint. These efforts include the development of machine learning and AI models for predictive power scaling. If successful, this approach could enable the intelligent identification of optimal time windows for draining compute nodes, allowing the system to adapt dynamically and efficiently to changing external conditions.
As the HL-LHC era (starting in 2030) approaches, WLCG computing centres are expected to support wide-area network (WAN) throughputs of several tens of terabits per second. This will require major upgrades to the WAN infrastructure at key sites, including the Spanish LHC Tier-1 at PIC. Preliminary assessments suggest that PIC will need network upgrades in 2026 and 2029, targeting bandwidths of 300 Gbps and 600 Gbps, respectively. These requirements have already been communicated to both CSUC and RedIRIS to ensure that research activities continue to benefit from robust and scalable connectivity.
To assess readiness for HL-LHC demands, a series of large-scale tests (Data Challenges) have been planned. In 2024, Tier-1 sites were evaluated at 25% of HL-LHC requirements (DC24), and PIC successfully passed all of the tests, demonstrating its preparedness in terms of network, infrastructure, and services.
In 2024 PIC continued to fulfill its role as Spanish Science Data Center (SDC-ES) within the Science Ground Segment (SGS) of the Euclid mission. The data processing includes two types of campaigns, the continuous On-The-Flight (OTF) processing that handles approximately 20 observations per day and the Regression Reprocessing (RR) which prepares data sets for major releases. In both cases SDC-ES contributes approximately 5% of the SGS computing budget to the processing campaigns. The images show the coverage of the observations and the related data processing in terms of the VIS pipeline in the SGS (red), and the contribution of SDC-ES (blue). By plotting the footprint on top of the galactic background one can appreciate the survey strategy which is focussing on areas around the galactic poles. Our involvement with ESA’s Euclid mission in 2024 extended beyond our role as the Spanish Data Center (SDC-ES). We also supported the local Euclid community, in particular, an effort for the integration in the Science Ground Segment infrastructure of a source injection code that can be very useful to validate the MER pipeline, and some activities within OU-SIM for pointing selection, catalog validation and pipeline orchestration.
In 2024, several simulation releases produced at PIC were delivered, providing fully controlled inputs for testing and characterising the Euclid science pipeline. The Euclid Flagship mock galaxy catalogue is featured in one of the Euclid reference publications: “Euclid. V. The Flagship galaxy mock catalogue: a comprehensive simulation for the Euclid mission.” The article has been accepted by Astronomy & Astrophysics and will be published in the Euclid Special Issue.
In September 2024, the Physics of the Accelerating Universe Survey (PAUS) released a groundbreaking cosmic distance catalogue, providing unprecedented precision in measuring distances to 1.8 million galaxies across 50 square degrees of sky. This extensive dataset, collected over 200 nights between 2015 and 2019 using the PAUCam on the William Herschel Telescope in La Palma, offers invaluable insights into the formation of cosmic structures and the influence of dark matter and dark energy. The Port d’Informació Científica (PIC) serves as the main data center for PAUS, hosting and maintaining all survey data. In collaboration with the PAUS team, PIC has ensured public access to the complete set of data products through platforms like CosmoHub, facilitating broader scientific exploration and discovery.
PIC is using Rucio, a highly scalable data management tool developed originally by ATLAS in the LHC, to manage the MAGIC data distribution. By the end of 2024, the refactoring of the PIC Rucio client for managing MAGIC data transfers was completed. The new version simplifies the code and is designed to be reusable for other data transfer workflows at PIC, including those for the Cherenkov Telescope Array Observatory (CTAO).
The CTAO-LST data center at PIC reached a total of 6.5 PB of data stored, becoming the third experiment at PIC in terms of data volume, just after ATLAS and CMS. The regular and stable operations of the LST1 telescope produced and transferred 1.1 PB during 2024. Additional simulated datasets, supporting joint analyses between MAGIC and LST, were transferred and distributed to the CNAF (Italy) and CSCS (Switzerland) data centers. These transfers enabled the first official off-site production and analysis activities during the autumn period.
PIC also played a significant role in the development of CTAO’s computing infrastructure, particularly in the context of the Data Processing and Preservation System (DPPS) work package. Contributions included testing deployment procedures, infrastructure validation, and the integration of the on-site data center with a remote site. A roadmap for further development was defined in the final quarter of the year.
Furthermore, PIC contributed to the CTAO Science Data Challenge portal, focusing on testing federated authentication mechanisms for both web and storage access, and integrating the CosmoHub platform to support interactive data analysis workflows within the portal.
The goal of the InCAEM project is to design, install, commission and define the exploitation strategy of an infrastructure for correlative analysis of advanced materials for energy applications. The project is the Catalan branch of the Advanced Materials coordinated project within the Planes Complementarios, funded by the by the European Union – NextGenerationEU in the context of the “Recovery, Transformation and Resilience Plan” (PRTR) and the Regional Government of Catalonia.
During 2024 PIC and ALBA worked together in setting up and testing a dedicated network path, comparing data transfer mechanisms like FTS and Globus, and planning for data challenges to validate the infrastructure. The scientific requirements gathering process continued, including discussion with peer scientific and technical personnel at sites like DESY, Jülich or DIAMOND) to exchange knowledge and define hardware specifications for CPU/GPU clusters, disk storage, and network equipment.
In terms of equipment for the offline data analysis platform, the majority of the hardware will be deployed in 2025 to prepare for the start of service. However, some initial hardware was purchased in 2024, and a detailed plan for the remaining hardware was created.
Our team also works on methods for data analysis and data pipelines. This includes work on denoising various types of data like EELS spectra and PEEM images using machine learning techniques. The collaborative work with ICN2 and ALBA resulted in a publication “Artificial Intelligence End-to-End Workflow for Transmission Electron Microscopy: From Data Analysis Automation to Materials Knowledge Unveiling” that was submitted to the arxiv, and a second one “Applied Artificial Intelligence in Materials Science and Material Design” that was published in Advanced Intelligent Systems.
The AI group at PIC applies machine learning and data analysis techniques across a range of scientific domains. Among our externally collaborative efforts, we have made significant contributions to the InCAEM and ICFO projects. InCAEM, part of the Planes Complementarios initiative and conducted in partnership with ICN2, ICMAB and ALBA, focuses on developing computing infrastructure at PIC to support advanced materials research. As part of this collaboration, we have worked closely with ALBA’s methodology group on AI-driven analysis workflows. Together with ALBA and Eurecat, we contributed to a poster presentation showcasing the use of autoencoders for signal denoising in the analysis of magnesium steel.
In collaboration with ICFO and the SNL lab, we have been studying neural activation using calcium imaging data. Our contribution focused on essential pre-processing tasks, including image registration and background subtraction. Unlike astronomical data pipelines, which are predominantly automated, this workflow initially relied on manual annotation of regions of interest. To address this, we adopted the open-source CAIMAN package, enabling automated preprocessing and integrated neuron detection, background subtraction, and signal deblending.
The group currently supervises two PhD students working on AI applications in astronomy: one focused on redshift estimation and another on image processing, including denoising of astronomical images. A master’s student also successfully completed and defended a thesis on AGN classification, in collaboration with researchers at ICE. In addition, we supervised several undergraduate thesis (TFG) projects and hosted an intern who explored distributed evaluation strategies for deep learning models.
On the dissemination front, postdoctoral researcher Laura Cabayol delivered a contributed plenary talk at the Euclid consortium meeting in Rome, presenting a novel technique for distance estimation using domain adaptation. Hanyue Guo presented results on photometric redshift estimation using the Flagship simulation at the OU-PZ meeting in Munich. The group also co-authored multiple publications within the PAUS collaboration.
The group is also involved in the Quantum Spain initiative. Antoni Alou, the project technician, has begun a PhD jointly supervised with J. Calsamiglia at GIQ, UAB. His research focuses on applying AI methods to optimize control pulses for qubit manipulation, with simulations of quantum systems carried out at PIC.