You are here

Representing Northeastern at Supercomputing 2016

November 30, 2016

About Me

Maddy Leger is a third year student pursuing a BS in Computer Engineering with a minor in Business Administration. She has completed one co-op at Microsoft in Cambridge, MA and is starting her second at Amazon Robotics in North Reading, MA in January. Maddy started doing undergraduate research with Dr. David Kaeli as a freshman through the ALERT and Gordon-CenSSIS Scholars Program, and has continued doing research with his group NUCAR since. She is the Northeastern IEEE Chapter President and also works with the Center for STEM Education at NU doing K-12 outreach.

Representing Northeastern at Supercomputing 2016

Since May, a group of 7 undergraduate students at Northeastern have been preparing for the Student Cluster Competition (SCC) that is held at the annual Supercomputing Conference. I was the team advisor, and the other six students were sophomores or middlers that all do research in Dr. David Kaeli’s research group NUCAR. The SCC brings together teams of students from top Computer Engineering or Computer Science programs around the world to run a set of physics, math and biology applications as quickly as possible in 56 hours. The teams build their own supercomputers (we call them clusters), ship them to the conference, set them up in a day and run the applications non-stop from Monday morning until Wednesday night. This year, the conference was in Salt Lake City, Utah, and the Northeastern team was ready to win. We had made a powerful cluster with a new software platform from AMD that was going to do really well in the competition, and we had worked hard for months learning the applications and optimizing them for our cluster.

I would love to write a story about how we came to the competition and had a flawless setup, ran the applications and finished first out of the fourteen teams there. What happened instead was far from that, but showed a lot about the talent Northeastern students have, from our research and co-op experience to our ability to handle pressure. When we arrived in Salt Lake City on November 12, our cluster was lost somewhere in the Midwest due to a shipping problem. We had no idea if it was going to make it to the conference in time for Monday morning, or if we were going to be able to compete at all. When we initially had learned the hardware wasn’t delivered on time, we packed up a couple systems we had at NUCAR and checked them as luggage on to the plane with us. We knew what we had brought wasn’t enough to be competitive, so we asked AMD, who had sponsored our original system, if they had any spare hardware for their booth at the conference. They gave us some extra parts and we quickly got to work trying to put together what we had, borrowing cables and other things we needed from other vendors at the conference and even from some of the other teams. Once we got the hardware up and running, we had to reinstall all of the software we needed to run the applications, as well as try to optimize the apps again for the new cluster. We worked non-stop day and night Saturday and Sunday to get the cluster to work for the start of the competition Monday. However, when Monday morning arrived, we still couldn’t get everything to work, and were worrying if we were going to even come close to the other teams’ scores.

The two applications that are run on Monday are the LINPACK and HPCG benchmarks, which are industry standards in the high performance computing world. Our thrown together cluster wasn’t quite ready to run such complicated applications, so we did our best to get a good score but didn’t come close to a lot of other teams. However, the competition has a power budget, and some teams had systems so powerful that they couldn’t get the benchmarks to run without going over the budget, which invalidates the score for that run. That meant that some teams didn’t even submit scores, so we were feeling hopeful for the remaining four applications that we had to run.

left to right: Spencer Hance, Selean Ridley, Zach Marcus, 
Jason Booth, Carter McCardwell, Tristan Sweeney

The competition itself is held on the giant conference floor, where vendors set up massive booths and demonstrations to show off their latest and greatest technologies. While we were waiting for applications to finish, we got to walk around the floor, meet a lot of smart people, and get a bunch of free stuff from the vendors. We also participated in a job fair specifically for students at the conference. Also, conference attendees and vendors would walk up to the SCC booths and ask us about our cluster and the competition. People were impressed with our ability to put together hardware in such a short time and still be running apps, and had a lot of questions about Northeastern, co-op and our engineering programs.

The last two days of the competition went a lot smoother than Monday had. Besides getting some sleep and walking around the conference, we finished a lot of the datasets for the applications and were starting to feel like we might not come in last place. At the end of the competition, we actually finished far from last, and received a special “MacGyver” award that the SCC committee had given us for our grit, intelligence and inventiveness. We have already started planning for next year’s competition, and are hoping to go back (without shipping problems!) and show how much better we can do with a real cluster!

Overall, SCC was a fantastic experience. It may not have been the success story we originally wanted, but we learned a lot, had fun, met some great people, represented Northeastern’s engineering program, and may have even gotten some co-op offers out of it. Now that it’s been over for two weeks, I’m already excited to start working towards next year’s win!

To learn more about the competition visit studentclusterchallenge.us 

To learn more about the conference visit sc16.supercomputing.org