VANCOUVER — The University of Northern British Columbia in Prince George, B.C., is showing you don’t have to be a large institution to be a big player in high performance computing.
Attendees at BCNET’s fifth Annual Advanced Networks Conference
(subtitled this year “The Power of High Performance Computing”) heard Tuesday how smaller institutions can leverage their smaller size and nimbleness to build a world class, multidisciplinary HPC facility.
In addition to its small size, with just some 3,500 students, UNBC is also a relatively new institution, opening its doors in 1994. However Peter Jackson, an associate professor of atmospheric science, environmental science and engineering programs at UNBC, said their newness gives them some distinct advantages. They have one of the youngest faculties in Canada, and everyone knows everyone. It makes cross-faculty co-operation a lot easier.
“It’s much easier to centralize when you’re small because you don’t really have any choice,” said Jackson. “It would be tougher for a larger institution because there’s so much noise going on, and everyone is doing their own thing.”
Centralization was key when it came to building an HPC facility at UNBC, making it easier to attract funding. The initial funding application included 26 different investigators from across campus.
With a $1.4 million grant from the Canada Foundation for Innovation and other sources UNBC built their HPC facility in 1999, a 28-processor SGI 3400 shared memory machine with 14GB of RAM along with hardware to support graphics, simulation and visualization.
Centralization has also made service and support easier. While hydro use and air conditioning are always an issue with HPC facilities, having it all in one location rather than dispersed at locations across campus, as is the case at some larger universities, has helped.
Jackson said the facility has been constantly busy with researchers from a range of departments running projects, from physics and the environmental sciences to geographers and even political science.
The facility also helped UNBC attract two Canada Research Chairs in climate prediction and hydrometeorology and helped fuel a 106.4 per cent change in sponsored research income, earning UNBC the ranking of second best small research university in Canada.
“It’s running all the time, we consistently see near 100 per cent utilization,” said Jackson. “It’s inherently general purpose, people from across all faculties are able to use it.”
Jackson’s own research is using HPC to model wind movement as a way of forecasting the future spread of the mountain pine beetle, a scourge that is devastating B.C.’s forests and is threatening to spread to Alberta. It is hoped the model could help predict future movement, so steps could be taken to prevent it.
UNBC’s HPC facility has been so successful they are now planning a $2.2-million upgrade — which they hope to have in place by September — to a 128+ processor 64-bit Linux cluster with a 32-processorAltix 350 shared memory machine.
The new lab will also have a portable visualization capability, linking back to the facility so images can be projected in the field. For example, projecting a model for locals in a remote village to show them the impacts of pollution on their environment.
“It’s not just researchers in a lab, it’s taking the visualization out to the stakeholders in the field,” said Jackson.
In an earlier keynote address, Eng Lim Goh outlined the challenges in reaching the next generation of HPC. As senior vice president and CTO of SGI, Goh is heading-up Project Ultraviolet, with the goal of designing and building the company’s next generation science-driven computer architecture.
“Our goal is to achieve pentaflop computing by 2010,” said Goh.
Most applications don’t scale well, so as they move from 1,000 to 10,000 processors Goh said they’re working to make sure every CPU cycle is fully utilized; and when a processor isn’t being utilized it is put to sleep to save power.
There are other challenges to be overcome but Goh said he sees great potential, and HPC is already making a major impact in industry.
Automotive companies are using it replace clay models, cutting the 36-month new-car design process down to 18 months, with a goal of getting it down to 12. Other applications include medical imaging, allowing surgeons to practice a surgery first, and flight simulation. NASA has a virtual airport tower it uses to train air traffic controllers, simulating crashes and other incidents they can’t otherwise practice for.