Our student teams get a lot of real-world HPC experience in the course of the three-month Winter Classic cluster competition. They get to run on five different supercomputers, are trained by HPC experts, and run/optimize real HPC/AI workloads.
This is all great and it’s attracting new talent to our industry. But are they really learning what they need to know to excel in an internship or entry-level position? Does a computer science degree plus our competition prepare them well enough Short answer: No. Less short answer: Kind of.
What I’ve found in doing my own competition is that few schools require (or even offer) hands-on Linux courses. Fewer schools have more than a class or two on anything relating to HPC. Students can get some experience via internships or working on academic research projects, or even just pursuing it on their own.
My goal this year was to provide a self-contained set of curated training that would give any motivated student enough Linux and HPC skills/knowledge to compete successfully in the Winter Classic – regardless of the curriculum at their university.
Progress has been made
We’re most of the way there now. We added a curated introductory Linux course offered by the Linux Foundation, which also has hand-on exercises. And after a long, long, search, I found a great set of HPC tutorials from the Pittsburgh Supercomputing Center. Next year, students will be able to not just watch the tutorials, but actually perform the exercises on the same system and using the same environment/tools – a very cool thing.
The HPC training component arrived a little late for this year’s crop of competitors. But then fate intervened when a scheduling mix up (my fault) meant that NASA Ames couldn’t mentor the students in their customary second module slot.
Problem Morphs into Opportunity (to torture students)
So I had an empty week on the cluster competition schedule. It was too late to find another mentor for that week and too late the shuffle the order. I could have just given the students the week off, sure, but that’s not me. What to do?
How about making them learn HPC basics? With that idea, the HPC Crash Course challenge quickly took shape. Their task: watch the 16+ hours of the PSC HPC tutorials. They can split the watch-load up among their teams however they’d like. At the end of the week, they’d have to take an exam – as a team collaboration. A timed online test with multiple choice, short answer, and essay questions covering the material. Google searches alone wouldn’t correctly answer the questions I posed; they had to actually watch and learn the material to maximize their score.
I was surprised at how well the teams did. It wasn’t a hugely long test but it was highly detailed. Twelve of thirteen active teams submitted exams, eight of them had scores above 90%. All of them learned something, making it a good week.
Check out the video below to see how it all turned out. Henry Newman and I discuss the exam (his input on the scoring was critical), show some of the questions, and reveal the scores and then update the ‘Big Board’ so see how the exam results changed the standings.
A few tidbits, though:
The battles for campus dominance are heating up at Texas Tech and Cal State Channel Islands. The races between teams at University of New Mexico and Fayetteville State U are becoming a bit more defined. But it’s still early and it’s still anyone’s game.
Next up we have the Oak Ridge Challenge and it’s a big one. For the first time in any student cluster competition, students will work on both a traditional supercomputer (Oda, a smaller Frontier clone) and…wait for it….a real quantum system provided by IQM. This is a pretty big deal, right? Stay tuned….