Scaling massive models
For the past two years, the Australian Bureau of Meteorology (BoM) has been working closely with NCI to test and refine their next-generation national weather prediction models.
“We’re trialling our next-generation models on Raijin in preparation for them becoming operational in a year or two,” explains Dr Michael Naughton from BoM.
“Every day at NCI we run Australian Community Climate and Earth System Simulator (ACCESS) numerical weather prediction suites, which support research and evaluation of prototype systems, with a special focus on severe and high-impact weather events in the Australian region.”
As part of a collaboration with Fujitsu, the NCI code optimisation team have been working with BoM, Fujitsu, CSIRO and the ARC Centre of Excellence for Climate System Science to increase the efficiency of the ACCESS component models. The faster they run, the larger the model that can be supported, and the greater the detail and accuracy provided.
Already collaboration on asynchronous input/output processing has realised a time saving of up to 20% using 2,000 cores.
“The collaborators have worked very effectively together to realise these improvements. The expertise of the NCI team has been essential to solving a number of critical software problems,” says Dr Naughton.
NCI’s experts have also been collaborating with the ARC Centre of Excellence for Climate System Science to improve the scalability of the US Geophysical Fluid Dynamics Laboratory’s Modular Ocean Model, which is critical for BoM’s weather prediction models.
The collaboration has already yielded a substantial increase in the scalability of the model, from 2,000 to 20,000 cores, allowing use of more than a third of Raijin’s total cores within a single run.
“Our spectacular Modular Ocean Model work, led by ANU, has benefitted from fantastic support from the NCI team,” says Professor Andrew Pitman, Director, ARC Centre of Excellence for Climate System Science.