Of Digital Competitions And Actual Relationships

AsianScientist (Apr. 8, 2021) – From Monday to Friday, Liu Siyuan is a system administrator at a buying and selling agency. His is the area of mammoth machines that transfer information at lightning speeds. From software program updates to {hardware} upkeep, he spends his days taking good care of their each want.

However Liu is not any unusual system administrator. As captain of the Nanyang Technological College, Singapore (NTU) workforce that gained the Scholar Cluster Competitors (SCC) in 2017, he was as soon as a keeper of larger beasts. For a number of valuable hours every weekend, he relives this expertise as he trains a brand new workforce of juniors to take up the mantle.

“Once I first began, I didn’t have plenty of helpful supplies and needed to be taught plenty of issues the arduous means,” Liu recounted. “I would like my juniors to get a very good begin.”

The primary SCC was held in 2007, but it surely wasn’t till nearly a decade later that NTU began sending groups to participate. By then, the competitors had grown to contain greater than a dozen groups from nations all around the world, together with China, the US and Europe. Within the faculty’s second 12 months taking part, the workforce of six undergraduates led by Liu gained the highest prize.
For the love of supercomputing

A Chinese language nationwide, Liu moved to Singapore in 2013 to check laptop science at NTU. There, he discovered a house within the excessive efficiency computing (HPC) group, which was then a motley crew of simply ten college students. Although small, the casual membership was devoted to proving itself at competitions: workforce members would meet weekly throughout faculty time period, and spend greater than half their holidays making ready collectively.

“Lots of my good buddies now are from the workforce,” Liu remarked. “I believe we constructed up this relationship as a result of we spent a lot time collectively. Most of my college life I spent with this membership.”

Dr. Francis Lee, who revitalized NTU’s casual HPC membership when he took over as membership advisor in 2016, mentioned that the college’s HPC group has been rising since then.

“With our success in 2017, extra individuals acquired to find out about us and extra individuals have joined,” he mentioned. “That’s how our ecosystem works; we have now enthusiastic college students who wish to be taught, and seniors who’re keen to come back again to share.”

Right this moment, the membership is a various group the place college students of various nationalities have come collectively to bond over a shared love of supercomputing. The most recent SCC workforce, Supernova, is a combined bunch hailing from Indonesia, Malaysia, China and Vietnam.

“Though we’re from totally different nations, all of us converse English,” mentioned workforce chief Li Shenggui, a fourth-year laptop science pupil from China. “We’re all from NTU and we’re used to the best way NTU college students work collectively. We share a typical tradition.”

For 2020’s competitors, which was held in November, Li partnered Aurelio Jethro Prahara, an Indonesian pupil in his second 12 months at NTU with a background in information science and synthetic intelligence. Collectively, they tackled the reproducibility utility—considered one of 4 within the competitors—which required studying a analysis paper on computed tomography reconstruction and replicating its findings below totally different settings and datasets.
Each cloud has its silver lining

Whereas the functions had been just like previous years’ challenges, the 2020 competitors was the primary to be carried out just about. Prior to now, college students had three months to construct a cluster and ship it to a conference middle, with cloud computing taking part in a minor position. This time, with the competitors held on-line, all groups migrated utterly to cloud-based HPC with solely a month to organize. Earlier years’ energy caps had been additionally changed by price range limits in Microsoft Azure cloud—constraints the workforce’s cloud supervisor Pang Jin Hui fastidiously navigated.

“It’s a trade-off between efficiency and value,” defined Pang, a third-year laptop science pupil from Malaysia. “If you happen to spend extra, your efficiency shall be excellent. However you have got a restricted price range.”

Although the educational curve was steep, cloud computing additionally had its perks.

“In some sense, it’s very straightforward as a result of we don’t have to do plenty of configuration,” Pang mused.

In earlier years, current libraries within the {hardware} and outdated software program made it troublesome to put in new libraries bug-free. The cloud cluster, although, at all times began out empty, with comparatively new variations of each {hardware} and software program. And energy might be ramped up simply the place wanted, permitting customers to harness supercomputing energy with out investing in pricey {hardware}.

“Supercomputing goes to be the norm fairly quickly. Nevertheless it’s nonetheless acquired an extended method to go,” mentioned Lee. “Lots of people don’t know how you can optimize; they suppose that so long as it runs, it’s okay. I hope a few of these college students will go into the sector like their seniors have, change that atmosphere, and contribute to it.”

One group, many nodes

With adjustments to the competitors format and the extreme time crunch, Liu’s steering has grow to be much more helpful this 12 months. On a selected Saturday afternoon earlier than the competitors, the workforce gathers in a Zoom room for yet one more coaching session. All muted with video off, solely their names seem on the black display screen: Aurelio Jethro Prahara, Pang Jin Hui, Tan Jia Qing, Lengthy Do Xuan, Dong Yunxing and Li Shenggui.

These conferences aren’t like those Liu remembers—there isn’t a lot room for high quality time spent collectively in-person throughout a pandemic—however the spirit and camaraderie of the membership endure. After a 20-minute-long lesson on cloud software program, the workforce settles down to debate technique, peppering their professor and seniors with questions on how a lot time and energy they should allocate to every part. Because the afternoon wears on, the dialogue meanders from budgeting to benchmark efficiency earlier than veering off to analyzing particular parameters. It’s darkish once they lastly disperse.

After two months of intense preparation, D-day arrived on November 11. Racing towards 18 different groups working throughout six totally different time zones, workforce Supernova gave their all around the subsequent 72 hours, finishing the benchmarks, functions and the thriller utility introduced solely in the beginning of the competitors. Regardless of their finest efforts, the workforce finally misplaced out to Tsinghua College, which dominated the competitors by popping out tops for all three benchmarks and a lot of the functions.

A barely disappointing outcome maybe, however Liu is aware of there’ll be a subsequent 12 months, and he is aware of he’ll be round. HPC is rather more than profitable competitions, he mentioned.

“What’s cool about HPC is that we’re at all times striving for higher efficiency,” Liu defined, his enthusiasm rising with every phrase. “That’s what we do every competitors, and that’s what HPC facilities are doing all around the world. They construct very advanced machines and attempt to speed up every kind of scientific discovery. Getting as a lot efficiency as we will with what we have now—that’s what I actually take pleasure in.”

In the end, Liu concluded, HPC is a few journey to continuously outdo your self within the face of constraints. Alongside the best way, you impart data to newcomers and construct group. And as your cluster of nodes grows, so does your collective energy.

This text was first printed within the January 2021 print model of Supercomputing Asia.
Click on right here to subscribe to Asian Scientist Journal in print.

———

Copyright: Asian Scientist Journal; Illustration: Oi Keat Lam.
Disclaimer: This text doesn’t essentially mirror the views of AsianScientist or its workers.