janez volmajer - Fotolia

Australian university reaches for the stars after HPC boost

The University of Sydney has upgraded its supercomputing infrastructure to answer big questions on cosmology and keep pace with growing research needs

At the University of Sydney in Australia, Welsh astrophysicist Geraint Lewis has been poring over synthetic universes simulated on supercomputers to better understand how the cosmos has evolved since the Big Bang.

Such calculations demand huge computing power, not only to crunch massive datasets, but also to run sophisticated artificial intelligence (AI) algorithms in deep learning applications.

Although the university’s existing Artemis 2 supercomputing system was last upgraded in 2016, it lacks dedicated graphics processing units (GPUs) that are more suitable for powering AI and machine learning workloads.

In June 2018, the institution acquired Artemis 3, a multi-purpose Dell-EMC PowerEdge high-performance computing (HPC) cluster with Intel Xeon processors and dedicated Nvidia Tesla V100 GPUs, to upgrade its supercomputing infrastructure.

Besides supporting projects in established fields such as geophysics and cosmology, and in the rapidly growing areas of genomics and proteomics, the $2.3m Artemis 3 will also be used for further research in areas such as economics, logistics and medical imaging.

“The addition of advanced deep-learning capabilities to our Artemis supercomputer is a mission-critical dimension of our research infrastructure,” said Lewis, who is also deputy director of the Sydney Informatics Hub at the university.

“With a greater number of research problems being data-driven or more accessible because there is data, our researchers will be able to investigate questions that were previously unanswerable.”

These include big questions on dark matter – the dominant mass in the universe that follows the pull of gravity.

“Normal matter, made from atoms, can undergo many more physical processes, including shocking, collapsing, star formation and feedback,” said Lewis. “Including this material increases the computational aspects of the problem, and huge computational resources are necessary.”

Lewis’s most recent work has examined other cosmological models that introduce new physics into the “dark sector”, where dark matter and dark energy can interact and decay.

Read more about HPC in APAC

“The influence of this new physics can be subtle, modifying the pattern of galaxies scattered through space,” he said. “Using our synthetic universes, we can search for the observational signatures of this non-standard physics, and guide the searches of the next generation of telescopes.”

Lewis said all of those calculations are computationally hungry, and upgrading Artemis – which delivers 125 teraflops of deep learning performance – would enable his team to “keep expanding the physics we consider, and improve the resolution so we can probe finer and finer structure within our synthetic universes”.

He added: “The ultimate goal is to provide the observational tests that will allow us to know whether our universe is truly as simple as it appears to be, or if there is dark physics that we need to find.”

Jeremy Hammond, director for strategic ventures at the University of Sydney, said Artemis 3 will also make it easier for the university to build a system suitable for all levels of academia, from undergraduate students to professors.

“By exposing our undergraduate and postgraduate students to supercomputing, we provide them with the skills to support and shape the knowledge-based economy essential for Australia’s technological future,” he said.

Apart from the University of Sydney, Australia’s Commonwealth Scientific and Research Organisation (CSIRO) has upgraded its HPC infrastructure with Dell-EMC systems to speed up supercomputing workloads. In some cases, it has managed to slash processing time from five hours to two hours.

On the software side, Australian researchers such as CSIRO’s Denis Bauer have developed machine learning libraries to analyse sheer volumes of genomic data in real time using big data processing techniques.

Read more on Clustering for high availability and HPC

Data Center
Data Management