SC12 It’s November, which can mean only one thing for computer sports enthusiasts: it’s time for another Student Cluster Competition. The seventh edition of this annual event begins in about two weeks at the SC12 conference in beautiful Salt Lake City, Utah. The competition pits teams of university undergrads against each other in a marathon battle to prove that they can design, build, and run the fastest (and most efficient) cluster.
There are two different tracks in the competition this year. The first is the traditional "build your own hardware" big iron track in which students design, build, and optimise their own clusters. The second track is a new pilot competition: teams are issued the same hardware (a six-node, Atom-based LittleFE cluster in a box) and challenged to wring the most performance out of it. This article discusses the big iron track; I’ll take a look at the LittleFe competition in an upcoming story.
For the big iron competition, the rules are pretty simple. Each team is composed of six undergraduate students, typically from a single university. The teams design and build their own clusters, getting equipment from one or more sponsors. They can use any hardware or software that’s available on the market and that they can wheedle out of their sponsors. The only limit on their systems is that they can’t draw more than a total of 26 amps of juice.
While most of the kids are computer science or engineering majors, many of them haven’t had much exposure to HPC, system design, or cluster building, so they have a lot to learn over the nine months or so before the competition takes place. During the design/planning/learning stage, the teams can tap their faculty advisors or others for advice. But once the competition begins, they’re on their own. They’re also out in the open – each team is in a small booth on the exhibit floor of the show.
The scientific applications will also be new to most of the students. The successful teams tend to have a deep understanding of these applications and how to tune/manage them for maximum throughput. They’ve also run the apps enough so that when they finally see the data sets on Monday evening, they’ll have a good idea of how to best schedule app runs in order to achieve higher scores. (I’ll cover the apps in a separate article.)
The schedule for the teams is jam-packed. After arriving in Salt Lake City, they’ll spend the weekend before SC12 (10 and 11 November) setting up and testing their gear. The competition begins on Monday, when the teams run the HPCC benchmark. Their HPCC results will be factored into their overall score, but they’ll also have a chance to run LINPACK independently in order to compete for the ‘Highest LINPACK’ award. On Monday evening, coinciding with the traditional opening-of-the-exhibit-floor cocktail party, teams will get the data sets for the four scientific applications that they’ll be running over the next 46 hours.
HPCC benchmark scores, results from the scientific applications, the teams’ public displays, and interviews with judges are weighted to make up each team’s final score. The winning team gets an amazing prize package that consists of… well, some certificates for sure… probably a plaque… and some team pictures. While every team wants to win, and believes they will win, the opportunity to compete is truly the big prize. These kids have fun and learn a lot at the same time. They also end up getting a lot of attention at the show from recruiters who are looking for highly motivated and knowledgeable interns and full-time hires.
In upcoming articles I’ll lay out the applications in detail and introduce the teams who will be competing for the Cluster Crown. (There isn’t a real crown - pity.) I will also be putting up an online betting mechanism that will allow you to bet an imaginary $1,000 on any team or teams you desire. To me, it’s not a real sport unless you can bet on it, and I can always be counted on to add that seamy and sordid touch to anything I’m associated with. It’s a gift, I guess. ®