Student Cluster Competition: The runners and riders
Checking out the form book
SC10 Eight university teams - six from the US, one from Russia and one from Taiwan descend on the SC10 supercomputing show in New Orleans next week to take part in the Student Cluster Competition (SCC).
The students bring their own self-designed and built clusters to the show, re-assemble them, and race to complete a set of benchmarks and workloads in the quickest time.
The competition tests their system design skills, their aptitude for learning new programs and new methods, and ability to optimize code to produce more (and better) output than their rivals.
El Reg is covering the competition from the show floor. To give you a taster, we have compiled profiles of the eight teams.
Experience + Hunger = SCC Success?
Hear the Purdue fight song as you read about the team.
If the Student Cluster Competition (SCC) were a table, then the team from Purdue would bring a lot of experience to it. But, then again, the table might not survive, because the Boilermakers will be bringing their trademark hard hat and sledgehammer to their SCC booth in New Orleans.
Purdue, along with Colorado, have competed in every SCC - four in all. But in each bout, they failed to win the big prize. In 2009, they won recognition for achieving the most flops per watt; but this year in New Orleans, they're looking to win the grand prize.
They've boxed up their system, and it's now on the way to New Orleans. This box is 'significantly bigger' than the system they took to Portland last year - large enough and fast enough to give them confidence that this time they could walk away with all the marbles (though there aren't any actual marbles to walk away with).
They're one of two teams co-sponsored by AMD, so I assume they are bringing the latest and greatest AMD processors. The Purdue boys (yep -all guys) will face off against teams armed with Intel's best. We'll see what happens in little more than a week.
As we heard from former competitor Dustin in our recent SCC overview , the SCC isn't just about the hardware. Knowing your system, knowing your benchmarks, using your time well, and having a solid team makes the difference between being competitive and winning it all. This is where Purdue might have an advantage.
One team member has participated in all four SCCs, beginning as a freshman in 2007. Other members work part time in Purdue's research data center, where they have daily hands-on contact with the same HP gear as they are running at the show. The team has also made good use of access to professors and other researchers for tips and tricks to maximize performance for specific benchmarks.
The Boilermakers are also organized. Each benchmark has been assigned to a senior team leader who works with younger students to wring out maximum performance. This helps in a couple of ways: First, the team leads can concentrate on optimizing a single benchmark rather than having to spread their focus. It also helps the younger members learn from the older, more experienced leads -a strategy that may help Purdue build an SCC dynasty. (Okay, they didn't actually say they wanted to build a dynasty -that's my hyperbole.)
Purdue's SCC application reads like an academic paper (it even has footnotes!), but the program has a lighter side. In 2007 they posted an innovative comic book to publicize their participation in the SC07 Bandwidth Challenge and the first Student Cluster Competition. So does Purdue have what it takes this year? Will they bring that golden trophy home to West Lafayette? (No, there isn't really a trophy, golden or otherwise.)
Louisiana State University
Home Town Tigers Tested
Here’s a link to the LSU fight song for your listening pleasure. It’s a nice accompaniment to the article below.
Louisiana State University, a mere 84 miles away from the site of SC10 in New Orleans, is competing to "bring pride to LSU and the state of Louisiana by showcasing the skills and knowledge that we will acquire through months of vigorous training for the contest:(as they put it). The LSU team is new to the SCC and, for the most part, new to HPC clustering. Like some of the other new teams, LSU has put in extra time to prepare -they started in January.
One of their first tasks was to build a Beowulf cluster and install some HPC apps on it for learning purposes. Judging by their sponsors, they’ll pack HP gear fueled by AMD processors with some Infiniband to tie it together. They didn’t specify much in their contest application about the clustering experience they have, or whether they have access to large clustered systems beyond what they’ve built with their own two hands (well, more like 14 or 16 hands).
Looking at the team you see that four of the six members have provided detailed biographical information including their computing interests and experience. One of the team cites Dr. Ian Malcolm from Jurassic Park as an inspiration -which makes me feel Jurassic Park-like, given that the movie came out when I was entering the industry. Also, it’s nice to see that these guys picked up a lot of computing knowledge in high school. The closest I came to a computer in high school was video arcade games, with my technical interest limited to: "I wonder if these games would accept a quarter-sized piece of metal instead of a real quarter?" They didn’t.
There are two team members who have blanks next to their names. I’m not sure if this means that they just didn’t hit the deadline for submitting their pics and info, or if their identities are being withheld for competitive reasons. Hmm... hard to imagine a college student missing a deadline. So I’m going to go with the more dramatic and conspiratorial explanation for the missing data. Until I’m proven wrong, these two guys are clustering prodigies who built their first cluster out of a junked set of Nintendo Game Boy handhelds. Sure, it wasn’t a great LINPACK, but they got a number.
The bottom line on the Tiger team is that they’re enthusiastic and motivated, plus they probably feel a bit more pressure due to being the hometown favorites. Is this enough to drive them past the other teams and into the winner’s circle? (Note: there isn’t an actual winner’s circle at the SCC.) Time will tell..
Florida A&M Rattlers
Compu-Rattlers and cluster battlers
For your listening pleasure while reading this story, here is a YouTube video of the FAMU marching band.
Another newcomer to the competition this year is a team from Florida A&M. Located in Tallahassee, Florida Aamp;M has an enrollment of 10,200 students, and their mascot is the Rattler -a rattlesnake. This will be the first time the Compu-Rattlers (going to go ahead and coin that nickname for them) will be competing in the SCC.
The Compu-Rattler team is composed of junior and senior computer science students who have a burning (well, maybe not burning) desire to learn about HPC. While they know computing, they don’t have a lot of experience with the care and feeding of clusters.
To help make up for this, they have prepared for the competition since January (longer than many teams). Their training regimen included early experience clustering together Mac Minis and older Power-based Apple servers. The school recently introduced a parallel scientific computing course, complete with its own 26-node cluster for departmental use.
What really comes through with the Compu-Rattlers is their enthusiasm for HPC and the competition. They’ve met weekly since January to work on cluster building and to learn how to use cluster tools so that it will become second nature by competition time. It’s also clear that they are approaching the SCC with the mindset that they aren’t just happy to be there -they want to win.
So will their drive and enthusiasm be enough to top the more experienced competitors? We’ll find out soon enough, as the inexorable march of time takes us ever closer to the crucible that is the SCC.
Real world, not book learnin’
The musical selection for this story is the national anthem for the Republic of China (Taiwan). It takes a while to build but finishes quite nicely.
One quote jumped out at me in the Student Cluster Competition application for the National Tsing Hua University Computer Science (NTHUCS) team team.
In the first paragraph they say: "During the past 2-3 years, we have learned some background knowledge from textbooks. However, instead of sticking our noses on text books, we believe that we should have some experiences on real world applications or implementations." They’ve chosen SC10 and the SCC as their proving ground.
The university competed in the SCC in 2007 and 2008, winning the LINPACK crown in 2007. But the 2010 team doesn’t include any of the students from those teams. Also, as they note in their application, no one on their team has any experience with HPC or clusters. They do have two former competitors to advise and train them during the run-up to the show.
Everyone on their current roster is deep in programming, and one member of is the systems administrator for all of the servers in their department. This could turn out to be a key position, given that teams have been derailed by hardware problems in the past.
Each benchmark has been assigned to primary and secondary application managers, who are responsible for understanding the app inside and out. With this knowledge, they can optimize their schedule to maximize throughput during the marathon 46-hour competition.
With HP as their major sponsor, we can assume that they’ll bring some blade-y type gear to the competition. Their location in Taiwan might be helpful as so much hardware manufacturing is located there; they might be able to snag some advanced parts right off the production lines. The major vendors, including HP, all have research outposts in Taiwan -experts who could help them squeeze every last bit of performance out of the hardware and software.
NTHU has fielded competitive SCC teams in the past and they’re certainly due for a win- but HPC and clustering is a new world to this 2010 team. Have they learned enough over the last 10 months to make 2010 the year they bring the SCC crown to Taiwan? (There isn’t an actual crown, but probably should be.) We’ll soon know. Here are some more details about NTHU’s team and goals for SCC.
University of Colorado
Not buffaloed by clusters and stuff
Here’s the CU fight song, with lyrics, to help put you in a Buffalo mindset as you read this story.
Colorado has competed at every Student Cluster Competition, beginning with the first in 2007. They posted the best LINPACK result in 2009, but have yet to grab the golden ring of ultimate victory. (There is no SCC Ring of Ultimate Victory, golden or otherwise.) They aim to turn that around this year by fielding an experienced team that has two returning members from their LINPACK-topping effort in Portland.
It’s also a deep team. When they wrote their SCC application, they had 13 members total and a traveling squad of eight: six primary team members and two alternates. The alternates, I guess, are there in case a primary team member gets injured in the course of the competition, or perhaps cracks under the merciless and unrelenting pressure to perform. CU has put together a diversified lineup that is numbers-centric -several members sport applied math majors. The balance of the team are computer science majors.
CU as an institution is providing a lot of support to the Buffalo team. They have tailor-made a computer science course for the SCC team that focuses on learning to use and optimize the system they’re going to be bringing to New Orleans. The campus has more than 30 departmental clusters and is close to completing a 1,368-node system. We assume this gives the CU team plenty of opportunities to hone their clustering skills.
From their application, it’s obvious that they already have a solid understanding of cluster technology and some of the tasks that they’ll be asked to perform at SC10. This is the payoff from having competed in the SCC before and also having a robust and supportive computer science department.
The CU Buffaloes look to be a very strong team. They have a lot of advantages going into the competition -strong university and faculty support, SCC experience, and a deep bench of motivated teammates. They also have an enviable slate of corporate sponsors in Dell, AMD, Mellanox and Fusion-io (via the HPC Advisory Council). Can they turn these advantages into a win at SC10? Is this the Year of the Buffalo? We’ll soon find out.
University of Texas, Austin
The eyes of Texas will be upon them/
To get into a Texas frame of mind, listen to their fight song as you read the article. Sing along if you have a mind to: the lyrics are there too.
This is the first time the Texas Longhorns have competed at the Student Cluster Competition. The inspiration for mounting the challenge came from a couple of team members who attended SC09 in Portland. They observed the SCC, which sparked an overwhelming urge to master cluster computing and test themselves at the world’s premiere HPC cluster competition. (It’s also the world’s only HPC cluster competition.)
I was surprised to see that UT hasn’t been in the competition before. The school has a solid computer science reputation, and their Texas Advanced Computing Center (TACC) is, well, a center of advanced computing. And if you’ve ever met a true Texan, you know that they have a bit of a competitive streak. So why has it taken them so long to jump into the fray? I’m baffled.
But they’re here now. So what are they bringing to the competition? First of all, they’re packing a rigorously trained team and steely-eyed determination to win. (I haven’t actually seen their eyes, but I do like the phrase). The Longhorn team sports members who are deep in network, visualization, and parallel programming.
They’re also making the most of their TACC links to give them the HPC chops they’ll need to successfully compete in New Orleans. In fact, they’ll be assembling and tuning their cluster at TACC facilities - and, we guess, getting expert TACC advice along the way.
The team is also studying past SCCs and have gained a lot of insight into what makes for a winning effort. I’m envisioning them in a screening room looking at film of past competitions, watching teams assemble hardware and run benchmarks -looking for any competitive edge. (It’s possible that I’m watching too much college football).
Rounding out the UT team are sponsors Dell and Microsoft. Microsoft is sponsoring at least two teams, setting up a Windows HPC vs. Linux angle to the competition.
I expect to see Texas become a perennial SCC participant. This competition is just too good a fit for the Longhorns, given their location, high tech heritage, and competitive nature. They’re all about making things bigger in Texas, and that extends to clusters just as much as cattle ranches.
Nizhni Novgorod State University
Russians fly 12,000 miles to display cluster prowess
I suggest you set the mood by listening to the Russian national anthem.
The team from Russia is traveling perhaps the most difficult road to the Student Cluster Competition (SCC). They face a unique set of challenges that serve to make the hill they’re climbing just a little bit steeper. While their university has a solid computing and HPC heritage and curriculum, this is their first SCC entry, and they’re learning as they go. The educational system is set up a bit differently in Russia than in the West. Skills are taught in different ways that don’t necessarily line up with what the SCC requires.
The NNSU team, for example, is composed of very bright students who are deeper in science and math than in computer science. So they begin with a solid foundation in the applications and how they work, but haven’t had nearly the hands-on experience with hardware as many of the other teams. Most of the Western teams are comp sci heavy and have members who are reasonably familiar with large systems, which may give them a leg up in this type of competition.
Most of the Russian students also have outside jobs in addition to their full-time course work; making the SCC team is an extra workload piled on top of their already demanding academic and work lives. But they’re enthusiastic and committed, and they’ve come up with a system design that should be very competitive.
To prepare for the competition, the team has been using a cluster located at the Redmond campus of sponsor Microsoft. While you can accomplish a lot via remote connections, there are still times when you need or want to physically interact with the system. The Lobachevsky State team has to rely on others to do this, adding time and extra steps to their tasks. As far as I know, they haven’t yet had access to a fully configured cluster like the one they’ll be using in New Orleans. Is this a big handicap? Maybe. But from what I can tell, it hasn’t dampened the team’s confidence or enthusiasm.
They’re going to spend a few days at Microsoft prior to SC10 so the students can get more familiar with the system and become acclimated to the time zone -and to American culture, too. I think that this will be the first international trip for the team members, adding just a bit more stress to what is already a stressful situation.
They’re going to be working against a language barrier as well, since most of the students either don’t speak English or aren’t fluent. But they will be very familiar with the discomfort of flying, given the 12,000 air miles they’ll log in traveling from Novgorod to Redmond to New Orleans and back.
But the NNSU team also has some factors playing in their favor. They’ve taken advantage of their access to Russian computing and subject matter experts to help them on their system design, benchmark optimization, and overall strategy. They’re making the most of their assets and bringing a strong entry to the competition. They don’t complain about their disadvantages; in fact, they barely mention them in passing as they talk about their progress.
NNSU is focused on showing what they can do and proving they deserve to be in the competition. After learning more about this team, I think that their effort is going to raise some eyebrows at SCC and that they will prove to be a very strong competitor.
Stony Brook University
Fighting Seawolves look to rock and rule
To increase your reading pleasure, I suggest listening to the Stony Brook fight song as you read the following story. They lyrics are interesting and include the phrase “Don’t let go until you hear them scream”.
This is the third Student Cluster Competition for Stony Brook. They went to the inaugural competition at SC07, skipped it in 2008, and then returned in '09 to snag the Overall Winner crown. (There is no actual crown, which is too bad.) They’re back again this year and looking to repeat.
Working with sole sponsor Cray, Stony Brook has put together a formidable effort. At least a few members from their winning '09 team have returned to help lead the charge at SC10. According to their SCC application, most of the team has significant experience in writing (and tuning, we suspect) highly threaded parallel code.
Every team has some mixture of computer science undergraduates and assorted science and math types. Stony Brook has these too, but also team members who have some pretty specific focus areas like compiler optimization and MPI coding. Other members have significant real world HPC work experience in highly demanding environments. The varied backgrounds and the ability to work with time constraints should pay off in the competition.
The team believes their real strength isn’t in hardware or technical skills alone. They cite their ability to work as a team, along with access to large clusters on Stony Brook’s campus, as key factors in their success last year. They’re looking to use their 2009 win as a springboard to taking the roses again this year. (There are no real roses.)
On paper, Stony Brook puts together a persuasive case. They have to be considered the favorite at this point. But anything can happen and there are definitely going to be some surprises at the competition this year. So will Stony Brook be the first team to repeat as SCC champions? Or will someone else emerge to take the crown? (There still isn’t a crown and I’m still bummed by that.) We’ll find out in a little more than a week from now.