DiRAC deploys supercomputers at four universities: Cambridge, Durham, Edinburgh and Leicester; the Project Office is at University College London. Technical teams at the four sites provide expert support to DiRAC users and work with industry partners on innovation projects.
The Data Accelerator in DiRAC’s Data Intensive Service uses flash storage to accelerate data-intensive workflows. Inspired and co-funded by DiRAC, and developed in collaboration with Dell-EMC, on deployment it was the world’s fastest storage system (IO500, June 2019).
The HPE/Arm/Suse “Catalyst UK” programme deployed an Arm-based HPC cluster within the DiRAC Data Intensive Service to drive the adoption of Arm processors in super-computing. To date, the system has supported research, user training and testing of system software.
The Memory Intensive system at Durham pioneered the use of a solid-state memory “burst-buffer”, dramatically reducing the time taken to save output data and increasing the effective computing power by around 13%. It is being used as a reference system by Dell-EMC.
Partnership between DiRAC and the Intel Pathfinding and Architecture Group has influenced the shape of future Intel HPC systems, including exascale hardware, novel network topologies and a patented new data format, bfloat16, for machine learning applications.
The DiRAC project office has supported nine STFC-funded Innovation Placements, which embed DiRAC students and post-doctoral researchers in external private and public sector partners to carry out projects in areas from computing hardware to NHS clinical care.
Since its inception in 2012 DiRAC has trained more than 100 students annually. Almost a quarter of these have graduate destinations in technical fields in the private sector, supporting companies in harnessing the power of HPC and contributing more than £50m to the UK Exchequer.
HPC is essential for studying complex biological structures such as viruses, allowing vaccines and drugs to be identified quickly. DiRAC supports consortia working on Covid-19: HEC BioSim, Scottish COVID-19 Response Consortium, JUNE modelling team, Folding@home, HPC consortium. The development of the JUNE agent-based model by a team of 10 PhD students, postdocs and academic staff at Durham with two collaborators at UCL was supported by access to computing resources and staff effort from the DiRAC Memory Intensive service at Durham
Exposure to solar storms would prove fatal to astronauts. DiRAC provides cutting-edge supercomputing systems which are enabling UK-led breakthroughs in solar weather forecasting, preparing for the safest approach to boundary-pushing space travel and exploration. DiRAC simulations are used to interpret data from major international satellite missions, such as Planck and Gaia. There is a growing synergy between the methods used for analysis of the outputs of large-scale simulations and the processing of satellite data streams.
A DiRAC-supported Innovation Placement student, based at Guy’s & St Thomas’ Trust with the Palliative Care Team, used machine learning to develop an Advanced Care Planning model. It identified those at risk of serious decline, greatly improving the care of frail patients. A Neural Net was also developed to help administrative staff to identify palliative care patients requiring extra appointment reminders, reducing ‘do not appear’ rates, improving patient outcomes and saving the NHS money.
DiRAC researchers support the application of artificial intelligence in new sectors including Government decision-making, planning and transportation. The Department of Work and Pensions are exploring the application of machine learning tools developed by DiRAC scientists, and a DiRAC innovation placement with Transport for London worked on a project aimed at keeping the Underground moving.