By: Marcus Bond, Professor of Chemistry and Physics, Southeast Missouri State University
“It’s not a sprint, it’s a marathon” declared my dentist in the middle of a recent procedure. Good advice for surviving SC—and life in general.
SC16 was my first. This was a conference I never believed I would be able to attend. But thanks to the support of STEM-Trek have been able to attend most conferences since. As I vaguely recall from my SC16 blog, I had set a goal during a discussion at the outing to Benihana, to scale the small cluster I was operating in our department, at the time, up to a central research computing system for the university.
I did not meet that particular timeline or that particular goal. But with some patience and persistence I now involved in a NSF grant funded state-wide project that has seen 12 compute nodes dedicated to undergraduate use installed in the condo model as part on the new Hellbender cluster at the University of Missouri. So, the seed of that original idea has morphed into a larger vision. Success due, in part, to my continued involvement with STEM-Trek, which has supported my continued involvement in the SC conference and engagement with the supercomputing community. Over the years I have seen my interests at the meeting gradually transition from just trying to learn the basics to attending more sessions on how to teach supercomputing to others.
A travel delay prevented my attending the first half of the pre-conference workshop. A shame since I get to enjoy an informative program in a relaxed, personal setting before the intensity of the full conference. And this year it was intense with a record attendance of over 18,000, in sharp contrast to the abruptly online conference planned for Atlanta in 2020 (the phoenix rising from the ashes, as noted by the conference chair in his opening remarks).
In any case, as I walked into the preconference workshop room during the lunch break, I was excited to see DK Panda in the audience. I had attended his tutorial on Deep/Machine Learning at SC23, which gave me a useful base for understanding AI. I found almost immediate use for it as I attended a few talks about AI on the exhibition floor last year and was able to understand them. So it was nice to hear from him again about his project to democratize AI.
My use cases for HPC have been limited to simulation, but I am increasingly interested/worried about how to implement AI in my research. More frequently AI is being featured in domain specific conferences. For example, at the 2024 International School of Crystallography workshop on Powder Diffraction there were sessions on applications of AI with the comment made that AI will not replace scientists, but scientists who use AI will replace scientists who don’t us AI. At a later session we were asked to write a Python script to implement the Scherrer equation. I had AI do it for me instead. My first professional use of AI! And a desperate bid to avoid replacement
I bookended the preconference workshop talk on AI with my final session of SC24: the well-attended panel discussion “The Paper is Dead. Long Live the Paper?—How LLMs Change the Field”. The focus was more CS centric with concerns that AI could write the programs and then write the paper about the programs. Should AI be listed as an author? These concerns were moderated by acknowledging that there are already many tools that aided manuscript preparation and perhaps AI is just another tool.
As a domain scientist I have watched computer automation of experimental equipment vastly increase output of publishable data. Traditional publication methods have become too time consuming, in my opinion, to keep up with the flood of data. I personally see a role for AI in expediting the process of moving data from the lab to public presentation, and would welcome it.
One concern that has been expressed about AI is the relatively high energy cost compared to a database lookup. Thus I was pleasantly surprised by the first presentation at my first conference session of SC24 (Workshop on Education for High Performance Computing) by Valerie Taylor from Argonne on “Energy Efficiency: A Major Focus for Future HPC Systems”. She argues for including energy costs at various stages of computation in the curriculum and teach students to consider energy expenditures in implementing software—perhaps letting a job run a little longer or (if higher precision is not needed) running at lower precision to conserve energy.
Other highlights of the workshop:
–A suggestion to have a student cluster competition oriented more toward software than hardware.
— A parallel programming exercise involving penny sorting that used a metronome to mimic clock cycles on a CPU, i.e. students could only do a sorting step when the metronome ticked.
–A presentation comparing programming languages taught in CS classes, with Java and Python dominating, with those used in the Exascale Project, in which C++ led followed by Fortran. Not that it matters to me since I don’t teach CS classes. It just warmed the cockles of my fossilized dinosaur heart to see Fortran still making the cut. A small sliver also went to Julia, which leads to my tutorial dilemma.
Ever since Alan Edelman’s compelling Sidney Fernbach Award presentation at SC19 on Julia I have been interested in learning more about it. There was a tutorial on Julia, but the same time as the tutorial on the Munich Quantum Toolkit. Quantum Computing is quite interesting, although I have been to quite a few workshops on it already. But I do like to learn about resources I could use to help me learn on my own. So, Munich Quantum Toolkit it was.
As a sidenote, I did hear another compelling presentation on the parallel programming language Chapel from a developer at HPE while standing together in the very long line waiting to get into the Technical Program Reception. So add another language to the list!
I enjoyed the Keynote Address from Nicky Fox with a broad overview of NASA activities. Her first topic about gathering data about our Earth built on the talk I attended at the NASA boothon the exhibition floor at SC23. I also enjoyed her videos of the Voyager spacecraft trajectories since it is the same video I show my physics students to show an object in uniform linear motion continues in uniform linear motion (more or less) once it leaves the Solar System.
I skipped my talk at the NASA booth this SC. Not because I had my fill of NASA with the keynote talk, but because it was so loud on the exhibition floor. This is the largest exhibition I have seen at SC. I was having trouble understanding some of the talks at booths simply because of everything else going on. Plus, a lot of these talks were packed. I had to wait until the crowds thinned out Thursday morning before I could finally get into a SLURM talk.
One thing I like to do at the exhibition is to meet vendors who are new to SC. Some notable visits:
I spoke with Dan at Dry Coolers. He had a booth along the periphery of the exhibit, where many of the new vendors were located. As a physical chemist I enjoy talking about the science behind the various technologies. Dry Coolers essentially provide swamp coolers, a more environmentally friendly option. Reminded me of the cooling arrangement at NCAR which I visited as part of the Linux Cluster Institute in 2017.
I stopped by the sinewave booth just because it caught my eye. I was teaching waves to my physics students right before I left for the conference. I thought it would be cool to pick up a promotional item to give to a student. Actually, I pick up quite a few promotional items from various booths to give mostly to students. Mostly I return from the conference to a Monday morning class right before Thanksgiving Break. My inducement to get them to come to class Monday is the promise that I will bring something back from the conference for them.
Tennessee Tech made their first appearance as a vendor. They are in the process of building a brand-new data center, named ASCEND, which seems quite ambitious. They are actually in the same athletic conference as Southeast Missouri State. We beat them 34-3 in football this season. But they have the booth at SC. I think they win.
I met Bruns-Pak as a first-time exhibitor last year, but I like their booth so much that I include them here. They design data centers! Which I think is a great addition to the exhibition. Great guys. They also have a fabulous raffle that few people attend. I hope they are back for SC25. If they are, check out their raffle.
Advanced Clustering Technologies is a long-standing vendor at SC. I include them here since this is where my HPC journey really started. I talked to Jim Paugh at the ACT booth at the American Chemical Society meeting in 2011 or so. He gave me a good rundown on what to look for in a cluster, even though I was not in the market to buy one at the moment. A couple of years later I was awarded internal money to buy a small cluster and ordered from ACT. A couple of years later he suggested I attend the Oklahoma Supercomputing Symposium. At which point I ended up on Henry Neeman’s email list, and here I am.
Georgia Tech brought a singing, vibraphone playing robot that performed with a full musical group or, when alone, would mimic the notes you tapped out on a keyboard next to it, while HRLS had an airplane parked at their booth.
I stuck around the exhibition floor past closing time so I could attend the Student Cluster Competition informational sessions. I have been interested in developing a state-wide team, but didn’t even know where to start. The panel discussion was especially useful in clearing up many of my questions. I am still in very early stages, but have started to float the idea with others in the state.
As I left the session I walked through the exhibition as the booths were being taken down. A completely new experience to me. It felt like a forbidden pleasure.
