DiRAC (supercomputing)

DiRAC (supercomputing)
Name	DiRAC
Established	2009
Location	United Kingdom
Type	High Performance Computing Consortium

DiRAC (supercomputing) is a United Kingdom distributed high-performance computing consortium providing specialized supercomputing services for computational research in astronomy, particle physics, cosmology, nuclear physics, and related fields. It connects university research groups, national laboratories, and funding agencies to deliver facility-class clusters optimized for large-scale simulations, data analysis, and model development. DiRAC supports collaborations among researchers affiliated with institutions such as University of Cambridge, University of Oxford, Imperial College London, University of Edinburgh, and national centres including Science and Technology Facilities Council facilities.

Overview

DiRAC operates as a networked set of compute clusters sited across institutions including University of Cambridge, Durham University, University of Leicester, University of Edinburgh, and University of Southampton, offering tailored architectures for parallel computing, memory-intensive workloads, and accelerated computing. Its service portfolio serves research projects funded by bodies such as Science and Technology Facilities Council, Engineering and Physical Sciences Research Council, European Research Council, Royal Society, and collaborates with international partners like CERN, European Southern Observatory, Max Planck Society, Los Alamos National Laboratory, and Brookhaven National Laboratory. DiRAC resources integrate hardware from vendors and consortia including Intel Corporation, NVIDIA, AMD, IBM, and storage providers used in grid and cloud ecosystems akin to OpenStack and XEN deployments.

History and Development

DiRAC began from strategic planning initiatives during the late 2000s involving stakeholders such as Science and Technology Facilities Council and multiple UK universities, launching initial services in 2009 to address compute demands of projects associated with Large Hadron Collider, Planck (spacecraft), and numerical relativity studies linked to researchers from University of Glasgow and University of Birmingham. Subsequent development phases aligned with national research infrastructure roadmaps steered by funding calls from Engineering and Physical Sciences Research Council and programme coordination with UK Research and Innovation and Royal Astronomical Society. Upgrades occurred in cycles responding to advances by firms like Intel Corporation and accelerator adoption led by NVIDIA deployments for workflows used by groups participating in collaborations such as LIGO Scientific Collaboration, LSST Corporation, and Square Kilometre Array precursors. Governance adjustments and centre consolidations have mirrored international initiatives such as PRACE and coordination efforts involving European Grid Infrastructure stakeholders.

Architecture and Facilities

DiRAC comprises heterogeneous clusters configured for distinct workload classes: massive parallelism for cosmological N-body simulations, large shared-memory nodes for stellar evolution and nuclear structure, and GPU-accelerated partitions for machine-learning enhanced analysis. Typical installations include multi-socket compute nodes built with processors from Intel Corporation and AMD, high-bandwidth interconnects supplied by Mellanox (now part of NVIDIA), and parallel file systems provided by vendors like IBM and NetApp. Sites deploy scheduling and middleware stacks leveraging projects and products such as SLURM, PBS Professional, OpenMPI, and container runtimes inspired by Docker and orchestration patterns used by Kubernetes. Facilities interoperate with national data services including British Atmospheric Data Centre-style archives and national laboratories such as Rutherford Appleton Laboratory for data curation, with physical hosting at university data centres following standards influenced by organisations like The Open Group.

Scientific Applications and Workloads

DiRAC supports workloads spanning cosmology, particle physics, nuclear theory, and astrophysics. Examples include large cosmological simulations comparable to projects by teams at University of Durham and Institute for Computational Cosmology, lattice quantum chromodynamics calculations akin to efforts at University of Edinburgh and Fermilab, and stellar hydrodynamics simulations with methodologies related to groups at University of Oxford and Max Planck Institute for Astrophysics. Researchers use community codes and frameworks such as codes similar to GADGET (code), ENZO (software), FLASH (software), and lattice suites resonant with developments at Brookhaven National Laboratory and Thomas Jefferson National Accelerator Facility. Workloads also include data-intensive pipelines for surveys linked to Sloan Digital Sky Survey, transient analysis in collaborations like Zwicky Transient Facility, and machine learning research influenced by publications from Google DeepMind and university groups.

Funding and Governance

DiRAC is funded through a mixture of capital and operational grants from research councils including Science and Technology Facilities Council, Engineering and Physical Sciences Research Council, and institutional contributions from participating universities such as University of Cambridge, Durham University, University of Edinburgh, and University of Leicester. Governance structures involve steering committees comprised of representatives from major stakeholders, with programme management coordinating resource allocation, user access, and project reviews in consultation with advisory panels featuring members from organisations like Royal Society and funding bodies in the United Kingdom. Collaborative agreements align with national strategy instruments under entities such as UK Research and Innovation and engage with international scientific infrastructures such as PRACE and European Research Council-funded consortia.

Performance and Awards

DiRAC systems have been recognised in community benchmarking and peer-reviewed publications for enabling high-impact results in cosmology, nuclear physics, and gravitational-wave modelling; such outcomes have contributed to awards and citations linked to researchers affiliated with LIGO Scientific Collaboration, Planck (spacecraft), and Large Hadron Collider analysis teams. Performance characterisation uses industry and community benchmarks comparable to High Performance Linpack and application-driven metrics used by projects at Los Alamos National Laboratory, with reported capability improvements following hardware refresh cycles from vendors including Intel Corporation and NVIDIA. DiRAC-supported science has been cited in articles in journals produced by organisations such as Nature Publishing Group, American Physical Society, and Institute of Physics, reinforcing the facility’s role in the UK and international computational science ecosystem.

Category:Supercomputing