Compute Resources
The Compute Resources capability is a foundational layer of the European Compute and Data Continuum, providing seamless access to heterogeneous computational resources distributed across HPC centers, community and commercial clouds, edge infrastructures, and emerging quantum platforms. For data-intensive domains such as HEP and RA, the Compute Continuum transforms a fragmented ecosystem of sites into a federated, performance-optimized environment that can deliver exascale throughput.
HPC and Cloud Computing
As noted previously, HEP experiments have been using HPC resources for more than 15 years, with HPC now making significant contributions to both production and analysis pipelines. To integrate HPC into the continuum, standardized interfaces and APIs must be used so that large-scale batch jobs can be submitted, monitored, and managed across federated HPC centers. This involves bridging local batch schedulers (e.g. Slurm) with global workload managers used by large particle physics experiments, so that simulation and reconstruction tasks can run on HPC without manual effort (as already piloted in HEP). Similar integrations are needed in order to use accelerators such as GPUs transparently, as both CERN and SKAO are exploring heterogeneous computing for AI-accelerated analysis (e.g. GPU enabled queues in SLURM can be made available and accessible via middleware such as DIRAC, Panda and glideinWMS).
At the same time, cloud integration offers a complementary pathway for obtaining additional capacity, although the nature of this elasticity depends strongly on the type of cloud, their configuration, policies and utilization. Public hyperscale clouds aim to provide genuine on-demand, elastic resources due to their large overcapacity and commercial operating model, making them suitable for e.g., bursty or exploratory workloads. In contrast, academic and on-premises research clouds, which typically operate at a smaller scale and are often funded to match expected scientific demand, making it less likely that they will always be able to provide immediate, on-demand provisioning. In these environments, cloud interfaces focus more on operational flexibility than guaranteed elasticity: sites may choose to adjust allocations between tenants or VOs, or reserve capacity for specific communities, but overall utilisation patterns often resemble queue-based (HPC-like) scheduling rather than commercial cloud bursting.
Within these constraints, federated cloud stacks (e.g. the EGI Federated Cloud, or EOSC services) still provide important benefits by exposing a unified mechanism to deploy VMs or containerised environments across participating sites. Suitable scientific workflows can then spin up VMs or Kubernetes pods where needed, for example, cloud bursts for Monte Carlo generation or interactive analysis, and shut them down when done (e.g. CloudComputingElement from DIRAC). This means that once a scientific workflow is set up appropriately, it can automatically start the computing environments it needs, wherever they are required. Container registries and orchestration services ensure portability across cloud and HPC. Several European efforts (EGI Federated Cloud, EOSC EU Node) already demonstrate cloud virtualization for research. For instance, the DataCloud project showed how big-data pipelines could be deployed across federated cloud sites to stay close to data sources (egi.eu). By pooling cloud resources, experiments can handle bursty workloads without long waits for some categories of workloads (e.g. of smaller scale).
Real-Time & Low-Latency Computing
Many HEP and RA instruments produce data at the “edge”, e.g. detector electronics or telescope antenna arrays, that must be reduced in size before transfer to central facilities. Edge computing nodes, located near or on-site at observatories and colliders, run real-time pipelines for filtering, compression, or transient detection. For example, LOFAR and the SKAO radio telescope arrays generate raw streams on the order of 10–100 Tb/s, which is far too much to send directly offsite, so initial calibration and gridding are done at on-site digital signal processors. By processing data at the source, edge nodes reduce network load and latency. Low-latency analytics (using FPGAs, GPUs or other accelerators) enable rapid responses, for instance immediate follow-up of astrophysical transients. Integrating instrument control systems with edge compute also allows health monitoring of detectors and feedback (e.g. beam diagnostics in accelerators). In summary, edge computing extends the continuum by handling the first stages of massive data flows, ensuring that only valuable information traverses the wide-area network to central archives. This data-reduction at the edge comes at the cost of information loss, but is necessary due to the high data rates. It is simply not feasible to store all raw data.
Quantum Computing
While quantum technologies hold long-term promise, their practical impact on HEP and RA remains limited in the near term. Current exploration is largely experimental, focusing on specific proof-of-concept studies in optimization or simulation. Within the compute continuum, quantum integration is best viewed as a forward-looking interoperability layer rather than a core operational component. While it is desirable to keep the door open for incorporating quantum or hybrid quantum–classical workflows in the future, their practical impact on relevant timescales remains uncertain; the aim is therefore to enable potential integration without disrupting established pipelines. The emphasis for now remains on readiness and modular integration, rather than active deployment.
Together, the HPC, cloud, and edge components make the computing infrastructure seamless, scalable and flexible, allowing HEP and RA users to run tightly-coupled simulations, large-scale data analyses, and modern AI/ML workloads across the continuum with minimal friction.