Before the build up to the final event in the 2018 Student Cluster Competition season (the SC18 competition in Dallas), I want to take a moment to write about one of the great inspirational stories of these competitions. I’m talking about the story behind the amazing success of the South African CHPC cluster team.

First some statistics: Team CHPC (South Africa’s Centre for High Performance Computing) has competed at six ISC Student Cluster Competitions, beginning in 2013. In three of these efforts, including their first two competitions, the team earned the Overall Championship. They earned second place finishes in two additional competitions and a close third place finish in the 2018 event.

It’s hard to overstate how high the deck was stacked against the South Africans in those first competitions. Their students were from universities scattered across the country – universities that didn’t have HPC classes or clusters for the kids to work out on.

The only HPC training and equipment available to the students was what the Cape Town based CHPC could provide.

That first CHPC team shocked the student cluster competition world by winning their first bout in 2012, beating teams that had many more resources, much more training, and much more cluster competition experience.

In December of 2013, I had the privilege to observe the CHPC team selection event and meet the kids competing for a spot on the national team. Their selection process is different than any other and their reasons for participating in the competitions are also unique.

Spreading the Knowledge

Happy Sithole, head of the CHPC, was (and still is) the driving force behind the organization’s participation in the student cluster competitions. His major goals are to build up HPC expertise in the country and get students interested in learning about HPC.

With this in mind, he laid down some rules about how the students will be selected for the national team. First, no student will compete in more than one competition. This is way different than other teams, the vast majority of whom send mostly the same students to every competition until they graduate and go professional. This policy ensures that the maximum number of students will get HPC training and the opportunity to compete at the ISC event in Germany.

The other rule is that there will be a national ‘play in’ South Africa competition where teams from South African universities compete in a mini student cluster competition. The winning team, along with outstanding students from other teams, are selected to compete on the CHPC national team. This is an annual event, taking place each December, and has become the foundation for an entire HPC conference in the country which, again, works to spread HPC expertise throughout South Africa.

Here’s a video interview with David McLeod, Happy Sithole’s right hand man in charge of their cluster competition teams. In the interview, we discuss how South Africa builds their teams and why they’ve been so successful. (Click on the upper left corner for the drop down menu and select “SA13 David McLeod Interview” video.)

When I arrived in Cape Town in 2013, I was surprised to see a number of familiar faces roaming around the competition venue. These were the students from the winning 2012 CHPC team, who returned to become mentors for the students on the various South African universities. They were very active mentors, showing the newbie teams how to set up their clusters, how to run applications, and how to wring the most performance out of their systems.

While the student mentors knew their way around clusters, as you’d expect from kids who beat the world at the ISC12 cluster competition, the students they were mentoring were very new to HPC. Almost none of them had any experience with HPC or even Linux prior to being trained by the CHPC. They had never used a multi-node system and had, of course, never had the opportunity to build a cluster from the hardware up. I asked many of the students the simple question “what did you know about HPC before this competition?” and filmed the responses. (Click on the upper left hand corner of the screen below and select “SA13 What I Knew About HPC” from the drop down menu.)

The CHPC had conducted some pre-competition training for the newbie students, but this was book learning, their real indoctrination was the baptism by fire that they received at the inter-South Africa mini competition. Mistakes were made, corrected, and new mistakes committed. But by the end of the competition, all the teams had gained much better technical skills, and, more importantly, were highly motivated to pursue a career in HPC. Here’s a link to all of my South Africa 2013 videos.

At the end of the three-day competition, the CHPC put together a gala reception with fantastic entertainment and food. The winning team, plus the outstanding students from other teams and alternates were named at the end of the evening. I’ve captured all of the proceedings on video – it was a fantastic party and I recorded a great interview with most of kids on the winning team. Here’s the video: (Click on the drop down link in the upper left corner of the screen below and select “SA13 Entertainment Team Reveal.”)

Fast forward six months to ISC14….and the CHPC cluster team shocks the world once again by repeating as Overall Champion.

So How Do They Do It?

Winning a cluster competition once could be a lucky lightning strike of inspiration. But winning it twice in a row is something different – it’s the result of hard work on the part of the students and coaches, plus an above average dedication to learning the ins and outs of HPC hardware, software, and how to optimize both.

I received an invitation to observe the South Africa intra-country cluster competition again, this time at Kruger National Park, which is a game preserve located in northeastern South Africa. It was certainly the most exotic location for a cluster competition and an inspired choice on the part of the organizers.

Eight student teams fiercely competed at the 2014 South Africa event and I captured video of all of them. Here’s a link to the YouTube playlist.

I think the secret of the South African success can be traced to a couple of different factors. First, they seem to outwork other teams both before and during the competition. Well in advance of the international ISC competition, the CHPC delivers introductory training to students at a large number of South African universities. The universities then form their own teams to compete at the annual December South Africa cluster challenge – where they receive even more training from team mentors.

Once the national team is selected, the HPC education goes into overdrive. CHPC personnel conduct meetings with their far-flung team members and continue to hone their skills in anticipation of the June ISC competition.

One factor that can’t be overstated is the participation and assistance of long-time sponsor Dell. The company has provided more support to the South African team than any other cluster competition sponsor.

They not only provide all of the equipment for the intra-country cluster competition, they’ve also collaborated with the CHPC to develop a unique way to help the kids understand the trade-offs inherent in putting together a cluster.

Each of the student teams is given an amount of virtual money to spend on their cluster. They then consult a special price list and select their components. If you watch the initial team video interviews, you’ll see that they have a wide range of gear. Some teams have lots of compute nodes (like eight or ten) but have spent less on memory or their interconnect. Other teams have spent more money on their interconnect and memory but have fewer nodes. It was fascinating to hear their rationale for their configurations.

Dell’s commitment to the team continues throughout. After the national team is selected, Dell flies the team to Dell HQ in Austin, Texas for a week of training, meetings, and BBQ. They even take the kids to TACC (Texas Advanced Computing Center) for meetings with HPC experts. According to the team members, this trip is hugely helpful in terms of HPC training and, moreover, something that few of them could ever imagine happening.

Dell also provides the hardware the team uses at the ISC competitions, which isn’t insignificant. The team usually sports an eight node cluster with plenty of GPUs, memory, and fast I/O. Dell gives the team what they need and also helps pay for their travel costs. In terms of their involvement and generosity, it’s hard to top what Dell does for the CHPC team.

Here’s an interview with Dell’s team mentor Vernon, where we discuss Dell’s participation in Team South Africa’s training and subsequent success. (Click on the upper left corner of the video for a drop down menu and select “SA14 Vernon From Dell Interview.”)

The Dynasty Continues

The team that was selected in December 2014 went on to win the second place at the ISC15 competition, narrowly edged out of first place by China’s powerhouse Tsinghua team – the most highly decorated team in Student Cluster Competition history.

CHPC came back in 2016 to win the Overall Championship again in 2016, grabbed another second place finish in 2017, and took home third place at the latest contest in 2018. Every time the South Africans compete, they end up with one of the top three medals – a truly enviable record.

Unfortunately, we won’t see South Africa at either the ASC or SC competition any time soon. The timing of the other two competitions (spring and fall) conflict with the South Africa academic schedule, plus they also don’t work with the South African selection and training schedules. The only way a CHPC team could compete at ASC or ISC is if they put together an all-star team of former students dedicated to the other two competitions. While this is certainly possible, it’s a lot to ask of a small organization like the CHPC. But put me down as someone who would love to see what the South Africans could do in the US or China.

In our next stories, we’ll be introducing the SC18 competitive field, the applications they’ll be tackling, and, of course, our on the spot comprehensive coverage. Stay tuned…