Researchers need powerful computers
The international scientific research community is one of the leading users of high performance computing (HPC), and being one of the most diverse scientific research organisations in the world, we've got a whole heap of researchers looking for that bit of extra grunt to help drive their research to new heights.
We refreshed our high performance computing cluster to be three times faster
Our new High Performance Computing cluster, Petrichor, has replaced our Ruby and Pearcey HPC clusters. It has over 400 server nodes with more than 25,000 cores and 235 TB of system memory, which means it has three times the computing capacity of Ruby and Pearcey combined, and can complete computational jobs 2-3 times faster.
Pearcey completed over eight million computational jobs in 2021, using 41 million CPU hours for the more than 340 registered HPC users across CSIRO. With Petrichor's higher speed and capacity, jobs are now being completed even more quickly.
The name 'Petrichor' was selected following a naming competition amongst staff, and describes the smell of earth just before rain. CSIRO researchers Isabel "Joy" Bear and Richard Thomas coined the term after discovering the cause of the heady fragrance.
Faster coastal modelling results
Petrichor supports our researchers across a broad range of areas such as environmental modelling, bioinformatics, fluid dynamics and materials science to solve real issues. Our Information Management and Technology team work closely with our researchers to ensure their work delivers maximum impact and fully harnesses the HPC capabilities on offer.
Since the installation of Petrichor the Sea Level, Waves and Coastal Extremes team have enhanced their coastal modelling capabilities.
"Using an Apache Airflow server to configure ocean wave models and parallelise runs in time, we are now able to perform a 40 year "hindcast" in the southwest WA region in less than a day using Petrichor", said Claire Trenham, experimental scientist in the Oceans and Atmosphere business unit. "Using traditional "hotstart" methods this job would have taken 2-3 weeks, and using the new method but older hardware would have taken a few days."
"This means we are able to do bigger experiments with model physics and parametrisations, as it is easy for us to throw the data away and generate new output very rapidly, or generate ensemble model runs for statistical analyses. In the coming months we plan to apply the same methods to other wave and coastal hydrodynamic models and other geographic regions, as well as performing future coastal climate projections."