Researchers and graduate students from two Québéc-based post-secondary schools will run a wide range of physics, chemistry and data mining applications on an $8.8-million supercomputer that’s set to top the Canadian record in the high-performance segment.
The University of Sherbrooke, in partnership
with the University of Montreal, has had one of a set of two server clusters running serial calculations since October of last year. A second cluster, to be used for parallel computing calculations, is in the testing phase and is expected to be available sometime this summer, officials said Wednesday. The two-cluster system, code-named Mammouth, are based on 872 Dell PowerEdge 750 servers with Intel Pentium 4 3.2 GHz processors and 576 Dell PowerEdge SC1425 servers with 64-bit Intel Xeon 3.6 GHz processors. Mammouth will offer 6.3TB of memory and more than 200TB of overall disk storage. The clusters will run Red Hat Linux.
Supercomputing is benchmarked internationally by a group of universities who publish the results on a Web site called Top500.org. In the most recent list released last November, the top-ranking Canadian site was 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. An IBM-based BladeCenter cluster using Intel Xeon chips, WestGrid placed 54 on the list, with a capability of 3,755 gigaflops per second.
Dell said Mammouth will have a theoretical peak performance of 8.3 trillion operations per second, or teraflops. The total computing power from the Dell HPCCs at the University is 13.9 teraflops, Dell said. According to the Top100 list, IBM’s Blue Gene/L retains the top spot with 70.72 teraflops. A more recent list will be released at a supercomputing conference in Germany next month.
David Sénéchal, a professor of physics at the University of Sherbrooke, said it took four years to secure a grant from the Canada Foundation for Innovation and Ministère de l’Éducation du Québec to fund Mammouth. The supercomputer may be used by more than 100 people to study high-temperature superconductivity, nanoelectronics, pharmaceutical development and weather and climate forecasting, he said. Once a year, researchers will be invited to outline how many CPU hours and processes they will need for their work, which will be reviewed by a committee of four to five staff members of both universities.
“It may become very popular,” he said. “In four years, the capacity of computers grow tremendously, and if you ask the researchers, they’ll always ask for more power.”
The supercomputing market has traditionally been dominated by IBM and HP, but Dell won the Mammouth RFP because it demonstrated the most financially competitive hardware, said Alain Veilleux, Manager of Centre de Calcul Scientifique at the University of Sherbrooke.
“People are worried that it’s taxpayers’ money, will it be it useful . . . but in our field, for high-performance computing, we got the most bang for the buck we could have,” he said. “If we took clone PCs and stacked them, we would have paid more.”
Debora Jenson, vice-president of the advanced systems group for Dell Canada, said the company had originally won the bid for only the Pentium cluster in Mammouth. It was its ability to trump competitor’s efforts to set up the second cluster that gave it a bigger piece of the project, she said.
“The effort to set up a computer room with all those servers, and rack and stack them, is a very big task,” she said. “You think about all the packaging in that kind of an installation, it has to be orchestrated.”
Sénéchal said the original strategy was to stay away from a single company in choosing its suppliers.
“The tradition in Canada is to have the preferred provider. If you start with this it’s very hard to get a competitive deal,” he said.
Dell’s services team, which helped set up the clusters, will provide ongoing hardware maintenance and support, Jenson said. The university will also dedicate about six technical staff to the supercomputer, including one person to write program software that will help researchers send a batch of calculations to a database, said Veilleux.
“The people make the machine, in a sense,” said Sénéchal. “You also need people to advice researchers on computing details, such as optimizing code.”