Kiwi scientists need two different kinds of supercomputers, but their big-computing needs aren’t being fully met, says a report on New Zealand’s science infrastructure just released by the Research Infrastructure Advisory Group (RIAG).
The RIAG report says high-performance computing is a priority for many scientific research sectors and some economic sectors need it, too. The report was prepared for the Ministry of Research, Science and Technology, and emphasises the need to share large, expensive scientific equipment.
The report commends KAREN (Kiwi Advanced Research & Education Network) as one way to get access to big-computing power. To government’s credit, KAREN was evolved outside the usual science funding model, which expects each research project to justify its own costs, says the report. More such sharing is needed to avoid duplication and the scattering of equipment across the country, it says.
“The current model of ‘a mass spectrometer here, an NMR [nuclear magnetic resonance facility] there, a screening facility and a [gene] sequencing facility somewhere else’… makes little practical sense in a country of the size of New Zealand,” says RIAG.
“There is a compelling case to build national centres of high-level competence where at least some of these capabilities are clustered and an effective access regime instituted.”
Yet, in costing out the high-performance computing facilities needed, RIAG has evaluated the needs of different sectors separately. It says that, in particular, the physical sciences sector — “including materials, nanotechnology and energy” — needs high-performance computers of “several architectures, distributed across the country”.
It’s not quite so necessary that big computers be centralised (compared to other scientific equipment) because KAREN’s availability makes it is somewhat irrelevant where computing power is actually located, says RIAG chairman Professor Jim Metson.
However, there is still an economic case for having bigger blocs of computer power than we have currently, he says. “This is an area where New Zealand has not done things well. People tend to buy their own machine for their own work.”
This is partly a reflection of the current science funding model, which prefers capital and operational expenditure to be fully accounted for within each particular project.
Criticism of this funding model is a persistent theme of the RIAG report.
We need at least two fundamentally different high-powered computer architectures, says Metson. We need “high-capability” computers, like the IBM Blue Gene computer at Canterbury University, which offers massive computing power in one box. And we also need “high-capacity” machines, built as clusters, which are able to do many sub-tasks in parallel.
Another high-priority need is archiving of research results, but archives’ location can be flexible, again because of KAREN’s availability, says Metson.