CSIRO and the Australian National University (ANU) have raised the possibility of merging the governance of Australia’s tier-one research supercomputers, in part to solve endemic funding problems.
In submissions to a federal inquiry that aims to create a ten-year infrastructure roadmap for research, the two institutions laid out their visions for where they expected researchers to get computing power from over the next decade.
Australia’s high-performance computers (HPCs) for research include NCI and the Pawsey Supercomputing Centre.
But funding of these facilities has become a long-term problem.
“The current position in this respect is dire,” CSIRO said.
“The NCI has just received what is effectively emergency support to ensure its continuation beyond its current four year life.
“The Pawsey Board has agreed that the Pawsey computer requires funding for a capital replacement in 2017/18 to remain viable.
Both CSIRO and ANU have now raised the prospect of harmonising NCI and Pawsey under a single governance structure, which could make it simpler to attract funding and keep them among the world’s best computers.
“The definition of Peak (Tier 1) HPC is traditionally defined as a compute capability that is in the top 200 globally,” CSIRO said.
“Given the pace of development, this means that Australia needs an ongoing mechanism to ensure it maintains a peak compute capability for the nation.
“There may be benefit in moving to a single unitary capability for Peak (Tier 1) HPC.
”However, this would require a recasting of the operational funding model for facilities. This is because the current operational funding is different for each computer. And furthermore, each computer has a different mix of peer review and directed access, as a result of the funding model for operations.
“CSIRO is happy to provide logistical support for a new national governance approach.”
ANU believed there should be a goal “of providing one integrated national e-infrastructure environment (comprising a supercomputer, a high performance private cloud for data-intensive computation, fast large scale storage, technical and domain expertise etc).”
It recommended lumping this under ‘advanced computing services’ and – like CSIRO – said that present funding models would need to be flipped to make it feasible.
While they believed in keeping a central Tier One HPC resource, CSIRO – at least – believed any research authority worth its salt should also have its own compute power available, and that this should not be reliant on federal funding.
“Below Tier 1, CSIRO maintains its own Tier 2 capability which is a powerful compute environment, as do many Australian research institutions,” it said.
“Given this, CSIRO does not see an argument for continued Commonwealth investment and believes all research institutions should be able to maintain a relevant scale Tier 2 compute capability as part of their basic infrastructure.”
CSIRO said costs could be reduced for internal compute by relying more on the cloud, though it noted the cloud wasn’t always cheaper, had some security challenges, and generally “is not at a level that can currently replace HPC, and hence for the foreseeable period cloud cannot replace HPC.”
“Given these points, CSIRO believe cloud compute is an area to be considered for Tier 2 compute and not Tier 1 at this time,” it said.
“CSIRO also sees the potential of quantum computing as a capability to make a significant step forward in HPC capability.
“Although not yet scalable, the development of this capability has great potential benefit.”