Post-secondary research institutes in Canada’s Atlantic provinces are about to join their counterparts in Ontario, British Columbia and Quebec by establishing a high-performance computing cluster featuring advanced collaboration tools.
The Atlantic Computational Excellence Network, (ACEnet) is made up of seven members including Memorial University of Newfoundland, the University of New Brunswick, Mount Allison University, Dalhousie University, St. Francis Xavier University, St. Mary’s University and the University of Prince Edward Island. The group was given a $9.9-million grant last year from the Canadian Foundation for Innovation and a mix of financial contributions from the provinces and private sector.
Although the first stage of the cluster has been set up with a mix of servers from IBM Canada, Silicon Graphics Inc. and Sun Microsystems, ACENet officially announced Sun as its preferred supplier on Tuesday. However, Graham Mowbray, its executive director, said there would still be some SGI equipment to handle some of the symmetric multi-processing (SMP) work.
Canada already boasts a number of high-performance computing (HPC) clusters such as the Shared Hierarchical Academic Research Computing Network (SHARCNet) and the Western Canada Research Grid (WestGrid), which are used by researchers to perform complex calculations or run simulations in a variety of fields such as computational physics and biology.
Mowbray said ACENet will offer dedicated resources for researchers investigating projects in a variety of areas, including oceanography, aquaculture and offshore oil and gas.
“The obvious question is, ‘Why don’t you just get a big long piece of fibre and connect up with SHARCNet?’” he said. “I think what we’ll find is we’ll tend to build ACENet in a way that reflects the main amount of research that’s being done in Atlantic Canada.”
ACENet will include Sun Solaris-based SunFire x64 servers running AMD Opteron processors, StorEdge arrays and identity management software, along with its Control Station software to manage the grid. Mowbray said most of the cluster’s needs would be satisfied through off-the-shelf tools, making a distinction between ACENet and a lab that might use more experimental or custom technologies.
“My own preference is to run it as a service, where we’re using proven technology wherever we can,” he said. “If you’re an oil and gas researcher or you’re part of a team of chemical researchers, you don’t want to go to ACENet and wonder if it’s up today.”
Peter H. Poole, an associate professor and Canada Research Chair in modeling and computer simulation at St. Francis Xavier University, said ACENet could prove useful in creating simulations of drilling muds, which are pumped into the ground in oil drilling projects but which can be difficult to extract and discard later.
“There are projects out there that we can’t even contemplate starting today, but through something like ACENet we could corner the market,” he said, noting that the region’s academic community is smaller and more widely scattered than other parts of Canada.
“We really need to take advantage of areas where scattered researchers can coalesce around a technology. I’ve already pitched to my colleagues that other regions can make good use of these distributed HPC consortia, but Atlantic Canada needs it.”
Mowbray said the first phase of ACENet, which will probably take a year, includes building everything from computer rooms and running fibre to the clusters and using Sun grid tools to tie SMP machines together. The second phase will be populating the compute nodes with more processors. By 2006-2007, phase three will be dedicated to innovation, he said. The idea is for ACENet to become self-sustaining, though he admitted the partners will have to study clusters such as SHARCNet and learn from their successes. The public sector support ACEnet has received so far is encouraging, he added.
“So much of our economy now is challenged by places like China or the emerging economies. They all have an innovation strategy, some kind of innovation trust that’s mandated for innovation,” Mowbray said.
Though vendors like Sun can provide a set of reference architectures and blueprints about managing large grid environments, forecasting compute needs starts at the customer, said Rob Adley, Sun Canada’s data centre practice manager.
“The reality is, as they start to deploy and understand their own workload and performance of the servers, what happens is they get their own sort of feeling on how things are growing,” he said.
SHARCNet, WestGrid and other HPC consortia have been experimenting with a “lightpath” connection using CANARIE’s CA*Net 4 optical network to establish a more national grid. ACENet is among the potential members to benefit from such a connection. ACENet will also host a high-profile HPC conference next year.