KNOXVILLE, Tenn., Dec. 5, 2014—A team of students from the University of Tennessee, Knoxville, sponsored by the National Institute for Computational Sciences captured second place in the Student Cluster Competition at the SC14 supercomputing conference held Nov. 17–19 in New Orleans. The students on this year's UT Knoxville team chose the name Team VIBE (Vols in the Big Easy) in reference to the nickname for the conference's location.
The Student Cluster Competition is a real-time, non-stop, 48-hour challenge in which teams of undergraduates, high school students, or both, assemble a small computer cluster on the exhibit floor and then race to demonstrate the greatest sustained performance across a series of applications while staying under a 3120-watt power limit.
The students dedicate many months of their own time and effort with the guidance of coaches and mentors learning how to architect a system and how to run and optimize various applications on their cluster. In the competition they apply what they've learned about hardware and software to try to win the race to solve scientific problems. The competition organizers and subject matter experts pick the data sets and initial conditions the students are presented with, and the students are graded on both the efficiency of the applications’ performance and interviews by the judges.
This was only the second time that UT Knoxville has participated in the competition, which attracts students from competing universities around the world. According to one of the judges, experience in the competition and the ability to learn from mistakes makes a difference in the results. In rising to win second place out of 12 teams this year, the UT Knoxville team made a huge jump from the prior year's last-place finish, just barely being edged out by the University of Texas at Austin, which was the competition's overall winner for the third year in a row.
|The University of Tennessee, Knoxville|
|Daniel Barry||Freshman||Computer engineering|
|Erik Blokland||Freshman||Computer science|
|Joseph Dorris||Senior||Computer engineering|
|Daniel Graves||Junior||Computer engineering|
|Chris Muzyn||Junior||Computer science|
|Ben Olson||Senior||Computer engineering|
|Dr. Vincent Betro||Computational scientist||National Institute for Computational Sciences|
|Paul Peltz Jr.||High-performance computing systems administrator||National Institute for Computational Sciences|
According to Dr. Vince Betro, one of the UT Knoxville team coaches, the Student Cluster Competition is really two separate competitions in one that require very different hardware, and the UT Knoxville team "proved that its balanced system was the best for going between both tasks."
Paul Peltz, the UT Knoxville team's other coach, expressed appreciation for corporate support and described his and Betro's role. "Our partnerships with Cray, Intel, NVIDIA, and Mellanox allowed the team to build and tune a highly efficient cluster," he said. "This was no easy task, though, as most of the students had little experience in building a cluster and running applications on the cluster. Dr. Betro and I taught the students the fundamentals of high-performance computing (HPC) and gave them the tools and knowledge to succeed."
The National Tsing Hua University in Taiwan won the first phase of the competition, for highest High Performance Linpack (HPL) score with a 10.08-teraflops (TF) performance. "UT Knoxville achieved a 9.808 TF run, and the next-best team accomplished 8.7 TF, so our machine was well positioned for that competition by using lots of accelerators," Betro said. "In the second phase of the competition, the teams are scored on their application runs, as well as interviews, through a complex scoring rubric that judges the correctness of the results and how quickly those results were calculated. The students' performance in this phase went equally well (74.76 points) with a second-place finish behind the University of Texas at Austin (76.49 points), which had a CPU-only cluster. If these two contests were summed, UT Knoxville could have won overall, with the most balanced and powerful machine in the competition, and an efficiency of about 3.25 gigaflops per watt, which would have ranked them fifth on the June 2014 Green500 of the world's most energy-efficient computers."
Betro added that the close finish left the UT Knoxville team excited to renew the UT–UT rivalry. "We’re really looking forward to doing this again next year on their home field in Austin [the site for the SC15 supercomputing conference]," he said. "We [Betro and Peltz] were very proud of our team's results, both in the HPL run and the general application runs. We also were extremely pleased with the way they conducted themselves. They were very genial. Everybody at the competition whom I know said they were in the top couple of teams as far as personality and skills."
The Student Cluster Competition is a win–win for the students and for the HPC industry because it assists in training the professionals of the future and in acquainting students and members of the vendor community with one another.
[Note: The following video is subtitles/CC enabled.]