Cluster Challenge: Purdue, ITaP student supercomputing team ready for SC09
November 11, 2009
Purdue students Alex Younts and David King helped build Purdue’s latest supercomputer in July. Call it a summer workout. Younts and King are now part of a Purdue student team running its own supercomputer in a competition at the SC09 Conference in Portland, Ore., which runs through Friday Nov. 20.
The 2009 Cluster Challenge entry from Purdue is like a mini version of the Coates cluster ITaP, Purdue’s central information technology organization, installed over the summer. Coates is the fastest campus supercomputer in the Big Ten and one of the fastest in the world. The student cluster is made up of similar HP computers containing AMD microprocessors. It also boasts similar high-speed networking, called 10 Gigabit Ethernet (10 GbE), from Juniper Networks to link the individual computers into a single, powerhouse machine. One challenge for the team has been tuning the software it must run in the competition to take advantage of the unusually fast connection, says Jeffrey Evans, an associate professor in the Department of Electrical and Computer Engineering Technology who’s working with the team members.
“I think that helped us get some of the pieces together more quickly,” Younts, a junior in computer science from West Lafayette, Ind., says of the experience assembling Coates. Like King, a senior in electrical and computer engineering technology from Lafayette, Ind., Younts is a student worker for ITaP, which operates Coates for Purdue faculty researchers. Hundreds of volunteers and ITaP employees built the new supercomputer in just a morning on July 21.
The refrigerator-sized Cluster Challenge machine, which ITaP helps sponsor, doesn’t have Coates’ more than 10,000 processors. But the student cluster still has more than three times the horsepower of the machine Purdue used for the competition when it entered the first time in 2007, Younts and King note. They think that may be enough to give Purdue an edge at SC09.
The Cluster Challenge, which draws teams of students worldwide, is part of the largest international conference for high-performance computing, networking, storage and analysis. Cluster Challenge teams—limited to undergraduates—have to run a set battery of benchmarking programs and working scientific applications as efficiently as possible. The student supercomputers churn through simulated data provided by the competition organizers, with little likelihood of actually finishing and a goal of simply getting as far as possible in the time allotted. The student team members share the load monitoring and adjusting their machines around the clock over three days as they work to achieve maximum performance.
The 2009 applications cover climate modeling, computational chemistry and data visualization, among other things. The students on the Purdue team spent most of the semester setting up and tweaking their machine, according to Professor Evans and Preston Smith, a senior systems administrator at the Rosen Center for Advance Computing, ITaP’s research computing arm. Smith serves as team leader.
Most of the students participate in Evans’ high-performance computing class at Purdue as part of their involvement in the Cluster Challenge. But they spend up to 20 hours a week on the effort overall and have to juggle other classes, homework and exams to attend SC09 for the competition.
The other team members are Max Hapner, a senior in computer and information technology from South Whitley, Ind.; Alex Miller, a sophomore in earth and atmospheric sciences from Zionsville, Ind.; Michael Niksa, a senior in computer engineering from Cedarburg, Wis.; Charles Timko, a junior in computer science from Highland, Ind.; and Michael Wleklinski, a junior in chemistry and statistics from South Bend, Ind. Hapner and Niksa also worked on the Coates cluster.
Besides Evans’ class, the experience has other educational benefits. For example, Wleklinski joined to learn about the chemistry software NWChem, one of the 2009 Cluster Challenge applications, which he’s interested in using in his own research.
“I'd like to attend graduate school and hopefully get a PhD in either analytical or physical chemistry with a master’s in statistics and then enter industry,” Wleklinski says. “By participating in this I've gained skills in computation that have allowed me to expand my horizons.”
Meanwhile King, who will graduate next spring, plans to take advantage of another Cluster Challenge benefit: the SC09 jobs fair “Take lots of resumes, you may have a job offer before you leave,” Evans advises the students. “That’s how highly Cluster Challenge team members are regarded.”
Miller was attracted by the opportunity to work with WRF a weather modeling and prediction application. “I wanted to get experience in modeling,” he says. “I also thought it would help to get involved with networking outside of my department. This experience will help with my grad school ‘resume.’ In terms of a career, the cluster challenge has helped give me an idea of what is involved in working with prediction systems on a daily basis.”
Hapner wants to become a professional system administrator eventually. He figures a key part of advancement will be knowing at least something about how most things in a data center work. He says the Cluster Challenge was a good opportunity to expand his knowledge.
Niksa is primarily interested in “parallel” computing to get software to take full advantage of multiple processors now featured even in laptops. “It seems as if that's the direction most branches of the discipline will be heading in the future if they're not on that track already,” he says. “I believe working with the parallel technologies required for the Cluster Challenge project put me on the cutting edge of software and computer engineering.”
Timko works for ITaP as a research application programmer and already had an interest in working with supercomputing clusters and parallel programming when Younts encouraged him to join the Cluster Challenge team. He’s been developing a version of the open source Ubuntu Linux operating system for use on clusters. But he thinks he might make a career of developing operating systems for a much smaller kind of computer—smart phones.
Writer: Greg Kline, science and technology writer Information Technology at Purdue (ITAP), 765-494-8167 (office), 765-426-8545 (mobile), firstname.lastname@example.org