Located at Perth's Curtin University, The Pawsey Supercomputing Centre is growing to become Australia's pre-eminent supercomputing centre and has frequently been represented on the Top500 supercomputer list. This new system "will provide a dedicated system for astronomers to process in excess of 30 PB — equal to 399 years of high definition video — of MWA telescope data using Pawsey infrastructure". It is not clear at what rate this data will be delivered to the centre.
The $2 million computer cluster will support extended compute requirements as the precursors to the SKA grow.
As with most modern supercomputers, GPUs are included to handle much of the "grunt" work, providing capabilities to power AI, computational work, machine learning workflows and data analytics.
Until now, processing of data collected by both the MWA and ASKAP telescopes has been done on Galaxy, Pawsey's real-time supercomputing system dedicated to radio astronomy. However, the data processing needs of both instruments has been growing: MWA has doubled the number of antennas available, and ASKAP will soon be ready to undertake full surveys of the sky.
To meet this growing demand, the new MWA cluster has been procured ahead of the main supercomputing system, as part of a $70 million Pawsey capital refresh project funded by the Australian Government.
According to Mark Stickells, Pawsey executive director, "the upgrade will allow Pawsey to deliver a service that is tailored to the Australian scientific landscape, and to keep pace with global advances in supercomputing technology. Procurement of the new MWA cluster was the result of a thorough consultation process with key stakeholders and will provide the best system possible to respond to the specific needs of MWA telescope users".
Stickells continued, "The new MWA cluster at Pawsey will feature 156 of the latest generation of Intel CPUs and 78 cutting-edge GPUs with more high-bandwidth memory, internal high-speed storage and more memory per node."
About the importance of this process and its results, Professor Melanie Johnston-Hollitt, MWA director, said "As the MWA director, I am delighted to see the conclusion of the procurement process, it was an outstanding example of collaboration between Pawsey and MWA and I am glad we had the opportunity to provide input into Australia's HPC future.
"As a researcher, I am excited that this new infrastructure will give us the chance to accelerate our workflows, leading to faster scientific discoveries and for providing the opportunity to continue to use the MWA as a scientific, technical, and operational testbed for the future Square Kilometre Array," she said. In conjunction with this upgrade, Pawsey is recruiting for a "Supercomputing Applications Specialist."
The Pawsey Supercomputing Centre's 546 TeraFlops MWA cluster will comprise 78 nodes, each with two Intel Xeon Gold 6230 processors operating at 2.1 GHz and providing 40 compute cores in total, a single NVIDIA V100 with 32 GB of high-bandwidth memory, 960 GB of local NVMe storage and 384 GB of main memory. All this will be contained in just two racks, saving on floor space as well as power and cooling requirements. This performance is not sufficient to earn a place in the Top500 supercomputer list, where the current minimum entry requirement is PetaFlop performance.
Pawsey officials noted that HPE was chosen not only because they successfully meet MWA's technical requirements, but also their ability to leverage resources around the world to provide the highest level of support for the lifetime of the system, in addition to their local support. HPE has 70 machines in the Top500 list, half of which were constructed in co-operation with Cray.
Commissioning of the new MWA cluster system is expected to be finalised by Q2 2020.