TORONTO – Canada’s high-performance computing community on Wednesday called for a infrastructure funding and investment model that would make facilities here internationally competitive and launched a graduate program to train the workforce that will run them.
In a long-range plan called Engines of Discovery: The 21st Century Revolution, a consortium of universities, industry and government called C3.ca said federal and provincial investment will be critical to creating high-performance computing (HPC) networks to rival those in the U.S. and elsewhere. The plan will see the creation of a new national entity, Impact Canada, which will be responsible for providing national leadership for HPC in Canada, the report said. It will co-ordinate facilities, advise Canadian funding agencies, develop and implement HPC awareness and training programs, and promote international collaboration.
One of Impact Canada’s mandates is to foster the development of computational science and engineering (CSE) programs that the report said will “fill the void between conventional science programs. These tend to be discipline specific, while computer science programs are not normally oriented to the application of computational methods for problem solving.”
Hugh Couchman, scientific director of the Shared Hierarchical Academic Research Computing Network (SHARCNet) that links 11 universities and colleges in Ontario, said C3.ca’s plan is intended to highlight the challenges of creating world-class HPC facilities that can keep pace with the needs of researchers and businesses. SHARCNet, which was originally set up on Compaq Alpha systems, is in the process of going through an upgrade to HP servers based on dual-core AMD Opteron chips.
“It’s not just keeping up with the Joneses,” he said, noting how quickly systems that seem state-of-the-art become obsolete. “If you join (one of the lists of the world’s biggest supercomputers) at No. 20, you’ll be on the list for three years if you’re lucky. They just plummet.”
Couchman spoke at the launch, which took place the same day the C3.ca report was submitted in Ottawa, of Canada’s first graduate program in Computational Engineering and Science, which will be offered by McMaster University in Hamilton, Ont., where he is a professor of physics and astronomy.
“These systems have to be tended very carefully,” he said. “It’s very hard to replace people if the resources aren’t there.”
The McMaster program will be an inter-facility school offered in partnership with McMaster’s engineering and science departments, along with some involvement from its business school, according to its director, Tamas Terlaky. Masters, doctoral and post-doctoral level research placements will be available in computational physical sciences, computational optimization, design and control as well as computational biology, Terlaky said.
HPC is commonly associated with systems that run simulations of biomolecular or physical processes that are time-intensive or invisible, even through microscopes. Large computers or clusters of servers run millions of calculations to speed disease research, drug development or other scientific projects. IBM, which developed the Blue Gene/L supercomputer to tackle these kinds of problems, is seeing a demand for similar compute power in a range of enterprise scenarios, said William Pulleybank, vice-president of Big Blue’s Centre for Business Optimization. HPC may be required to perform prize optimization to help retailers determine the best price to sell their products, for example, or track weather fluctuations that would affect what kind of energy demands a power utility could face.
“This is not stuff that would necessary attract the IT person. It goes directly to the line of business,” he said.
Margaret Wright, chair of the Computer Science Department of New York University, recalled HPC work she did at Bell Labs to improve the way wireless networks are set up. Traditionally, she pointed out, teams would install antennae and a tower and then conduct “drive tests” where carriers would draw rough maps that indicated where performance was spotty. Wright’s project ran software to simulate the network, with measurements to scale, which could estimate the same problem areas.
“No one wants to say, ‘OK, we’ve done it on the computer, let’s go,’” Wright said. “But it may mean they conduct only one drive test instead of 20.”
CSE programs will help standardize the way students learn about advanced networks while encouraging multidisciplinary interaction between people who don’t always work together voluntarily, added Barbara Keyfitz, director of the Fields Institute at the University of Toronto where McMaster launched its graduate program.
“This is work that requires a team. It’s not the lone scientist working in the lab or at the blackboard,” she said. “Everyone tends to divide subject materials into disciplines. In the business world, there are turf wars and jurisdictional barriers as a result. The need to protect your own little part of the enterprise is not special to universities.”
A group of technology vendors including Apple, Cray, HP, IBM, NEC, SGI and Sun Microsystems all endorsed the C3.ca’s long-range plan.