Basic concepts, tools, terminology covered in eight week study
From clusters and multicores to parallelism and petaflops, the world of high-performance computing (HPC) can be intimidating to newcomers. That’s why, for the fourth consecutive year, the Oak Ridge Leadership Computing Facility (OLCF), a Department of Energy Office of Science user facility located at Oak Ridge National Laboratory (ORNL), offered a class over the summer called HPC Fundamentals in collaboration with the National Institute for Computational Sciences (NICS) at ORNL.
The 8-week course was open to anyone who wanted to learn some of the basic concepts, tools, and jargon associated with the field.
Bobby Whitten, group leader for the User Assistance Group at NICS, ran the class with assistance from OLCF staff members, including Suzanne Parete-Koon, Robert French and Adam Simpson. According to Whitten the goal of HPC Fundamentals was “to give people an introduction to what supercomputing is all about so they at least have a flavor of what we’re talking about—to demystify it just a little bit.”
Whitten said that understanding the basics of supercomputing is important, even for interns or staff who are not directly involved in the computational science community.
“Most of those folks out here at ORNL probably are going to go into science careers, and most science domains are now using supercomputers in some form or another. There are fewer and fewer that do strictly laboratory work.”
Suzanne Parete-Koon, a user support specialist for OLCF who taught a few of the sessions over the summer, stressed the necessity of introductory education.
“I think there is a need because whenever we do really basic programming, we get huge numbers; for instance, the Crash Course had really great attendance. If you look at the OLCF website, the programming tutorials get a lot of traffic. Are those our users? Maybe someday.”
The classes covered topics such as the role of HPC in science today, an introduction to UNIX/LINUX, and programming in Fortran. While even the outline might seem overwhelming to absolute beginners, the emphasis is on giving participants an overview of basic HPC concepts. Supercomputing is a complex and still-foreign scientific domain to many; it would be impossible to even mention everything in an 8-week course.
“We’re just skipping stones across the pond of HPC fundamentals,” said Whitten.
The focus on breadth instead of depth is a good strategy for absolute beginners, but what happens after they complete the class?
“A lot of times once they get through HPC Fundamentals, they go onto more specific training like what are the particular compilers? Or, let’s train you on how to program a particular architecture like GPUs. How do you run your programs on different kinds of hardware?”
While this kind of training is very useful for researchers in their individual positions, it does come with some drawbacks. It goes from a very broad introduction to specific, purpose-driven education, which leaves a gap for intermediate students.
Whitten would like to offer more extended training opportunities to anyone on campus, but striking the right balance between current and future needs can be tough.
“You’ve got to focus on the researchers and scientists who are getting science done on these machines—that’s really what you’re getting funded to do—but there is also a realization that you also have to train a broader group of people or you won’t have these scientists in the future. Where’s that intermediate phase? Where does that fit in?”
To try to reach such groups, the center offers some other courses such as the 1-day Crash Course in Supercomputing as well as outreach-oriented educational events for high school and college students. However, Whitten and Parete-Koon both say the demand is high for more courses at intermediate levels.
About 200-300 people attended the most recent 1-day crash course, according to Parete-Koon. “It was a huge showing, so that might tell you kind of what the need is. And when we had it, there was somebody who had never tried command-line terminal, and we had other people who had done advanced parallel techniques like MPS [massively parallel sequencing],”she said. “So how do you teach that range?”
Whitten hopes to address these issues in the future.
“The one shortcoming is that [the training] does kind of just end. There’s a whole lot more that goes on beyond that, and that’s probably the next thing we’ll start working on,” he said. “When you get through all the beginning fundamental stuff, what’s next? This is 101 Supercomputing. What’s the 102? What’s the 103? What’s the 503 of supercomputing?”
“We’ll get there,” he stated. —Christie Thiessen