How do you compete in supercomputing? The students on the Texas A&M University Supercomputing Team know that’s going to be the first question before it’s even asked.
“Build. Run. Optimize,” said Dr. Jian Tao, advisor for the team and Texas A&M Engineering Experiment Station researcher. “You build the computer, you run programs on it and then you optimize the performance.”
They’re simple words, but the competition is anything but. Over 300 student teams from around the world entered the Asia Supercomputer Community’s Student Supercomputer Challenge (ASC18), but only 20 made the cut to the finals to be held in May in China. Texas A&M’s group is not only the lone team from the United States to make the finals, it’s the only team from the entire Western Hemisphere.
And it almost didn’t happen. Instead of just trying to form an extra-curricular team, Tao was encouraged by the Engineering Academic and Student Affairs office to create an AggiE-Challenge course for the competition. AggiE-Challenge courses give students the opportunity to work on projects related to engineering challenges facing our society. In this case, the challenge is building efficient high-performance computers.
Working against the clock, Tao was able to get the course approved during the middle of the last week where students can add and drop classes.
“I wasn’t sure we would even be able to get anyone, but they were able to announce the class and some students were interested,” he said. “We didn’t even have a room yet. We had to figure out where the best time and place to meet would be after everyone had already signed up.”
Four students will travel to China, but there are nine on the team. They’re a mix of underclassmen and upperclassmen coming from multiple departments within the Texas A&M College of Engineering. Like any team, figuring out each other’s strengths and weaknesses was one of the major challenges early on.
“We have members with different engineering backgrounds and skill levels,” said freshman Michael Lau. “Two of us are freshmen, so we’re learning some things we wouldn’t have learned for a couple of years otherwise. So that’s been cool also to see how everyone fits together on the team.”
At the competition, teams have to perform tasks, and to do them they have to build a supercomputer. (The team is quick to point out that when you say “supercomputer” it conjures up images in popular culture of a room of servers or IBM’s Watson, but these computers for the competition are just two nodes compared to large supercomputers that can have hundreds). To build that supercomputer, the team will pick from a list of available hardware before the competition. That hardware will be in boxes at the competition, and they have to assemble the computer. The key is that the power consumption cannot exceed 3,000 watts. So teams have to select what combination of parts will best accomplish the tasks while staying below that threshold.
The tasks involve running several different programs that use significant computing resources. Some of the programs take more central processing unit (CPU) power, and others more graphic processing unit (GPU) power. Then teams have to find ways to optimize the way those programs run on the systems.
“Should you put more CPUs in there or GPUs?” said Tao. “That’s what you have to decide. They just give you boxes, and you have to assemble it yourselves and install the software and optimize it yourself. It’s a very cool challenge.”
The wildcard round at the finals is a mystery challenge that teams have no way to prepare for. They know that the other tasks will be using the same programs they used in the preliminary round, but the mystery task will be what separates the teams at the finals.
“Everyone is going in blind,” said senior computer science major Mohammad Sagor. “It could be hit or miss. We all have our various things that we’re working on and how to optimize that, but that’s going to be a game changer. And on top of that it’s with hardware we’ve never used before, so that’s another layer.”
Adding yet another layer is the fact that the internet itself is more heavily censored in China than it is in the United States.
“There are some restrictions on the internet in China, so it’s not as simple as just looking something up here,” said Sheldon Wei, another freshman on the team. “That’s something we’re going to have to figure out on the fly because there’s no way to prepare for that really.”
Being able to look things up is a crucial part of the challenge according to junior computer engineering major Sean Dormiani, because the “installing the software” portion of the competition isn’t like installing a new program on your regular desktop computer.
“Installing software sounds simple, because you’re used to clicking next, but when you’re compiling it from the source it’s not the same,” he said. “It’s a very minimal front end, it’s all text, there are no buttons. “It’s like building a tower and you’re supposed to build the top, but you need everything underneath it, and sometimes it’s not documented what you need underneath it. So you’re like ‘I’m missing a floor here, what’s going on?’ and sometimes you end up finding one guy on one forum somewhere that has the same problem and they’ve found a solution.”
They’re also just excited to be the first Texas A&M team to ever compete at the event.
“We’re going to introduce Texas A&M to the competition and I’m excited about that,” said sophomore computer engineering major Shaina Le. “It's really inspiring to have a bunch of teams come together, and even though we're competing against each other and against the clock, we're all working toward the same end goal of sustainability, efficiency, and optimization. It's all about the bigger picture, and I think that's wonderful.”
The team has been supported by Texas A&M High Performance Research Computing, and members also include senior computer science major Isaac Decastro, senior electrical engineering majors Michael Hagaman, Samra Tariq and Alexander Ngo and freshman engineering student Zengyu Wei.