A handful of Canadian companies and universities made it onto a list of the 500 fastest supercomputers in the world.
Canadian high-performance computers on the TOP500 list included WestGrid (a network used by Simon Fraser
University, the Banff Centre, University of British Columbia, University of Calgary, University of Alberta and University of Lethbridge), Environment Canada, University of Toronto, Bell Canada, C.O.R.E. Digital Pictures and University of Montreal. WestGrid was the highest-scoring Canadian supercomputer at 34.
Japan’s Earth Simulator supercomputer, built by NEC, however, retained the No. 1 position with its Linpack benchmark performance of 35.86 teraflops, or trillions of calculations per second.
Dave Turek, vice-president of deep computing at IBM in New York, said the list is important because it shows certain trends in the marketplace, notably that certain technologies in supercomputing are being accepted or displaced.
“”None of us would sit back and say, ‘Aha! The list has told us something about what we should do,'”” he said. “”I think that it does have a more material effect on policy makers at the governmental level in terms of how they view trends and directions in the industry and what they should do to support it, et cetera.””
For example, 1995’s shift from vector architecture to a parallel or cluster computing approach showed recently-formed companies the industry no longer embraced the vector approach, Turek explained.
IBM enjoyed a significant standing in the TOP500, with the vendor boasting 44.8 per cent of supercomputing systems and 50 per cent of installed performance. HP had 28 per cent of the systems and 18.5 per cent of total performance.
Two prototypes of the upcoming IBM BlueGene/L system were among the top 10 ranked supercomputers. Based at the Thomas J. Watson Research Center in New York and Massachusetts, the computing models are a joint effort between Lawrence Livermore National Laboratory and IBM.
According to the ranking, a total of 287 systems use Intel processors, whereas six months ago 189 Intel-based systems were on the list and one year ago only 119. The second most common processor is the IBM Power processor (75 systems), followed by HP’s PA-RISC processors (57) and AMD processors (34). There are 291 systems labeled as clusters, up from 208 in November 2003, making them the most common architecture on the TOP500.
IDC Canada analyst Alan Freedman said clustering and grid technologies, which use several computers, are some of the leading components of supercomputers because they give users resource flexibility and additional processing power not typical of a single machine.
Moreover, there’s definitely more capability and capacity offered by Intel processors, Freedman said. He said the Itanium processor, in particular, has probably become more prominent on the TOP500 list.
“”That’s Intel’s introduction into 64-bit computing. So that’s allowing them to handle workloads that require greater addressing capabilities,”” he said. “”Large database transaction processing, things that used to go towards a 64-bit platform, can now be run on Intel.
Christina Christara, an associate professor of scientific computing at the University of Toronto, said research projects that are netting a big chunk of the funding in Canada involve solving problems in astrophysics, the environment, electromagnetics, higher space and graphics.
In the U.S., Turek said, “”there’s a fair amount of dollars flowing from a lot of different sources that manifest itself in the marketplace in different ways.””
He said the government sponsors basic research in the physical sciences, classic kinds of scientific and engineering research and computer science-related activities.
Turek said industry helps to establish standards and, “”through their own purchasing behaviour,”” stimulate support of technologies that may be deployed in research specific to a particular sector, such as the application of computing to discover oil in the petroleum industry.
The TOP500 list was compiled by Hans Meuer of the University of Mannheim in Germany, Erich Strohmaier and Horst Simon of NERSC/Lawerence Berkeley National Laboratory in California, and Jack Dongarra of the University of Tennessee.