National Tsing Hua University, Taiwan
2010 Champion returns to defend crown
SCC team profile Student Cluster Competition time is fast approaching. Below is our first competitor profile and check out our introduction here.
Last year, the team from Taiwan’s National Tsing Hua University (NTHU) overcame some adversity to take the 2010 Student Cluster Challenge (SCC) crown* as the overall winner. The team also won a blue ribbon** for being one of the first three teams to break through the Teraflop barrier during the SCC.
They achieved these feats despite a sponsor change, with Acer stepping in to provide equipment at the last minute. This gave the team less opportunity to get to know their hardware; and Acer, while very helpful, isn’t a power in high performance computing (HPC) and thus doesn’t have the resources that other vendors can offer.
Combine that with the language barrier, a tight travel schedule, and jet lag, and you’d think that the NTHU team would be at a distinct disadvantage.
But they overcame these problems and kicked ass. Of the six categories used to determine the overall winner, NTHU took first place in four of them and second place in another. It wasn’t a runaway victory; the next closest competitor was only about 10 per cebt behind them.
While they (and all the teams) obviously have solid technical skills, one of the reasons for NTHU’s win was pretty basic – something that almost any layman could have figured out. The story really illustrates how the competition works, and why it’s so interesting to follow.
One of the major decisions the teams have to make is how to approach each of the computational problems they’re assigned. There are four different workloads, each relating to a different scientific discipline and each with its own unique computational characteristics. The students can decide how and when to run the workloads and how much resource to devote to each.
One last year’s tasks was to hack four huge datasets of passwords. This was a computationally intense task, and the datasets were large enough that it was highly improbable that any team could complete them all. While the rest of the teams simply loaded up the data and started chugging through the files, the NTHU team took a different approach.
They sorted the data first, reasoning (correctly) that shorter passwords would be easier to crack, and then attacked them in order of length. As a result, they almost maxed out this task, receiving top scores on three data sets and 80 per cent on the fourth. This put them ahead of their closest competitor by 20 per cent and accounted for much of their margin of victory.
This year, the team is again sponsored by Acer, but its members are almost entirely new, and yet to be tested in the SCC pressure cooker. The team leader is the sole holdover from the 2010 championship team.
Their approach is much like last year’s; in fact, their application is almost a carbon copy of the one they submitted last year. I have to admire the efficiency – why come up with entirely new words to say essentially the same thing?
Each team member is responsible for at least one SCC application, meaning that they’ll be the expert in running and optimizing it. As they put it, “A detailed training schedule is set,” which I take to mean, “We each have an incredibly long list of tasks to master, and we will devote many, many hours to doing so.”
While this version of their team hasn’t seen SCC before, they have been tested in computing competitions. In April, they entered a similar competition sponsored by the Taiwan National Center for HPC. It was a timed test on how quickly they could build and tune a Linux PHC system. Not surprisingly (to them) they won, and are counting on this experience to help them win again at the 2011 SCC.
They also had some opinions how their team is coming together, what they expect to see
I can't deny that facing some frustration indeed, but totally, I get a lot of fun. Not only the way our teammates get along with but also the sense of accomplishment from solving problems.
-Since I participated in this competition (and the training process) last year, I can say I am pretty familiar with these preparing work. I think my teammates might suffer a little bit more than I do. (Since it's their first time participating in this competition)
- It's more fun - because I have learned a lot of knowledge and system management. It's a great opportunity for me to learn more.
- I'm worried of being the one pulling our team back, so I have to work very very hard!
- Sometimes we need to stay in the lab very late to set up the machines (install the OS, system utilities, things like that). To keep myself and my teammates awake, I tend to make noises (I can do some BeatBoxing noises) to keep my mouth and their ears busy so we don't accidentally fall asleep. I hope my teammates don't complain me for being too noisy. :p (making a note right to capture that on video during one of my visits – I’ll post it in a video blog, gotta see this….And noooo….I can’t imagine that being annoying at all, ever after hours and hours of being in a high stress situation…)
- We have prepared this since spring; we (have) spent a lot of time trying to simulate the computations. We will get the champion!!
One thing I didn’t ask is if they’ll be bringing their temporary tattoos again this year. In 2010, they had their school seal on a temporary tattoo. I applied one and it lasted from mid-November until somewhere around mid-March – and that was with me showering at least once a week!
- On a more serious note, I’m really looking forward to seeing what Team Taiwan can do this year. They quietly go about their business without much noise or drama – their booth was an island of serenity and calm. But they definitely know what they’re doing and they want to become the first repeat SCC Champions. Right now, I think that they have to be considered the team to beat.
* There is no actual SCC crown, sad to say. I’m betting that someone could build a hell of a cool crown using discarded computer parts.
** The SCC doesn’t award ribbons, or even ribbon cables.