Case Study
CERN and GAIA, via STFC: data infrastructure for particle physics and a billion stars
Stakes
Two of the most demanding scientific programmes in the world depend on the data behind them. At CERN, the Large Hadron Collider generates raw collision data on the order of a petabyte per second across its four main experiments at peak luminosity; hardware and software triggers discard around 99.99% of it in real time, leaving roughly 1 PB/day written to permanent storage and about 90 PB/year added to a tape archive now totalling around 1 exabyte. The European Space Agency's GAIA spacecraft is mapping the position, motion and properties of more than a billion stars. Both programmes need data infrastructure that simply cannot fail. A pipeline outage is not a service ticket; it is lost science. Capacity has to be planned for missions that run a decade or more.
Constraints
- Mission lifecycles measured in decades, not quarters
- A heterogeneous compute estate spanning very different scientific workloads
- Secure network and communications infrastructure for research use
- Operating within UK government scientific computing standards
- Capacity planning under uncertainty: scientific demand grows in ways business demand does not
Approach
Treat infrastructure as a long-lifecycle asset
Scientific computing rewards infrastructure designed for ten- and twenty-year horizons. We approached compute, storage and network as long-lifecycle assets, with clear refresh cadences, capacity buffers and operational documentation built to survive multiple generations of staff.
Engineer for known and unknown demand
Some workloads had predictable shapes. Others would only emerge as scientists found new questions to ask. We sized for both: deterministic provisioning for the known, headroom and elasticity for the unknown.
Operate to research-grade reliability
Mission-critical scientific compute does not get a maintenance window during a collider run. We designed for in-place operations, tolerance of partial failure and rapid recovery, and every change was reviewed against the mission calendar rather than an IT change calendar.
Document for mission longevity
Configuration documentation, operational procedures and capacity baselines were treated as deliverables, not by-products. The next engineer through the door, in five or ten years, needed to inherit a runnable system.
Deliverables
- Designed and operated the mission-critical compute and storage estate at the Royal Observatory Edinburgh
- Maintained secure network and communications infrastructure for research use
- Capacity planning and lifecycle management for scientific compute and storage
- Operational procedures and configuration documentation built for long-lifecycle handover
- Disaster recovery and business continuity validation across scientific workloads
Outcome
Reliable data pipelines fed two of the most demanding scientific programmes on Earth: CERN's particle physics analysis and ESA's GAIA stellar census. The infrastructure ran to the cadence the science required, not the cadence enterprise IT defaults to. The work shaped how Cipherer approaches every mission-critical engagement since: design for the longest reasonable lifetime, document for the next engineer, and treat reliability as a scientific instrument rather than a service-level agreement.