Research Computing

Elastic, on-demand computing rented from remote data centres. Rather than moving petabytes of survey data to a researcher's own machine, cloud computing lets the analysis run next to the data, a shift that underpins the science platforms being built for the next generation of surveys.

Packaging software together with its exact dependencies into a portable container, so that an analysis runs identically on a laptop, a supercomputer, or the cloud. Containers have become a cornerstone of reproducible research, freezing the computational environment in which a result was produced.

Spreading a computation across many loosely-coupled machines so that datasets far too large for one computer can be processed in parallel. Frameworks for distributed data processing let astronomers cross-match billion-row catalogues and reduce survey images across clusters of ordinary servers.

Graphics processing units, originally built for rendering, excel at the massively parallel arithmetic that powers modern simulation and machine learning. Astronomy uses them to accelerate everything from N-body dynamics and radiative transfer to the training of the large neural networks now applied to survey data.

The supercomputers on which astronomers run the largest cosmological simulations and process survey data at scale. By dividing a problem across thousands of tightly-coupled processors, high-performance computing makes it possible to model the growth of cosmic structure or the collision of neutron stars from first principles.

Integrated, cloud-based environments that bring the analysis tools to the data. Platforms such as the Rubin Science Platform give researchers notebooks, catalogue databases, and image access alongside a survey's archive, so that discovery no longer requires downloading the data at all.