From Tailored Systems to Scalable Machines, Multiple Architectures to Meet Competing Demands in a Data-Intensive World
Neil Stringfellow, Pawsey Supercomputing Centre, Australia
From the selection of architectures to the operation on the full systems, identifying real science cases is central in prioritising focus for effort and expenditure in Australia. Researcher engagement in the selection process for a petascale systems has involved close collaboration and this continues as we look at the demands for scalable usable systems, and from these engagements involving a general purpose community as well as science focus areas, a set of complementary research instruments have been acquired tuned to specific purposes.
With many researchers having been able to take advantage of community codes already able to scale to tens of thousands of processors on other machines internationally, this has freed up effort to be applied in performance and scaling enhancements in unfamiliar science areas. Science priority efforts at both national and state level require innovation in working with poorly scaling science codes, but to ensure continued investment in supercomputing for the whole science community, it is vital to address the weaknesses in these priority areas. An examination of some of these fields of research, the corresponding uptake efforts and the associated challenges will be presented.
The existing and potential use of supercomputers in Australia is showing an increasing focus on data intensive computing in addition to the existing simulation sciences. As a major data-processing project, a significant effort in Australian Supercomputing for the coming years will be engaging with international partners in the planning for the Square Kilometre Array (SKA) radio telescope, which will require exascale data-processing capabilities. Current operational requirements for the precursor telescopes demand access to both CPUs and GPUs, and the existing operational setup and the grand scaling challenges for the future in the SKA project will be discussed.