Monitoring and Observability
To operate a complex continuum, comprehensive monitoring and analytics are needed. The Monitoring and Observability block ensures that the European Compute and Data Continuum is not only performant and reliable, but also transparent, efficient, and environmentally responsible. For HEP and RA, which operate on multi-decadal timelines and exascale-scale infrastructures, observability and sustainability are not optional features but foundational requirements.
Infrastructure Monitoring
Observability begins with end-to-end monitoring of the health and performance of resources across the continuum. Compute nodes report CPU/GPU utilization, memory and I/O rates while storage arrays report usage and errors. Data center environmental sensors measure temperature, power consumption, and Power Usage Effectiveness (PUE). A centralized telemetry system ingests these metrics in real time. Automated rules detect outliers (e.g. a failed disk, a supercomputer node overheating) so that maintainers can intervene before failure. For example, many HPC sites already use tools like Prometheus and Grafana to visualize cluster health. Over time, trends and predictive analytics will inform capacity planning (e.g. forecasting that a cluster will be full by next year or that network upgrades are needed). On top of raw infrastructure, application-level performance monitoring traces scientific workflows. Each step of a pipeline emits logs and metrics such as job durations, queue wait times, I/O patterns, error counts, etc. Performance monitoring also supports billing and reporting (showing how many core-hours an experiment used). Collectively, these observability data help developers tune code for the continuum and reassure funding agencies that resources are used efficiently.
Operational Intelligence
Raw monitoring data are transformed into actionable intelligence. Analytics and anomaly detection automatically flag unusual behavior, from failing disks to network congestion. Trend analysis and forecasting enable capacity planning, ensuring that infrastructure keeps pace with growing data and compute demands. AI-driven predictive maintenance reduces downtime by identifying hardware likely to fail before it does, a critical feature for 24/7 operations like telescopes or LHC detectors.
Energy Monitoring
Sustainability in the compute and data continuum requires moving beyond simple efficiency indicators toward a more comprehensive, impact-oriented understanding of resource usage. While metrics such as Power Usage Effectiveness (PUE), Carbon Usage Effectiveness (CUE), and water consumption provide useful operational insights, they must be complemented by broader environmental impact assessments, ideally grounded in Life-Cycle Assessment (LCA) methodologies. Fine-grained monitoring of resource usage—covering compute, storage, and data movement—can support more informed decisions, but the objective is not only to optimise efficiency. It is to guide usage toward sufficiency, ensuring that resources are used only where they deliver meaningful scientific value, while avoiding rebound effects associated with unconstrained optimization.
Within this framework, scheduling and execution strategies in the continuum can incorporate environmental awareness as one of several objectives, alongside performance, reliability, and scientific urgency. This includes exploring trade-offs in where and when workloads are executed, taking into account factors such as energy mix, infrastructure characteristics, and overall system impacts. Such capabilities can support more responsible use of distributed resources, improve the robustness and efficiency of data-intensive workflows, and provide transparent accounting of environmental impacts to stakeholders, including the European Commission and member states.