Team TUMuchPhun: A Chronology of Events
While the Oktoberfest in Munich is in full swing, people in traditional clothing are pouring through the city and millions of liters of beer are sold, team TUMuchPhun is completely sober and working hard on the applications for this year's Student Cluster Competition.
Team TUMuchPhun – that is six computer science students from the Technische Universität München, who are highly interested in high performance computing. The team members Michael Zellner, Gregor Matl, Felix Thimm, Daniel Gallenberger, Felix Spaeth, Sharru Moeller (from left to right) shown in the picture above at the university's parabola slide.
The story of our team started in March this year, when the local chair for scientific computing introduced students to the competition for the very first time. We were quickly intrigued by the possibilities of this competition, by the knowledge we will gain about high performance computing, the possibility to work on actual scientific codes and on supercomputers, and maybe to some extent by the fact that the competition will take place in Austin, Texas.
The following weeks were dominated by plenty of meetings, a tremendous number of mails and our first photo shoot, all of which resulted in our application for the competition. Then, one month of suspense followed, until in May our team was finally accepted.
At this point of time, the actual technical work started. Every team member was assigned two or three of the five applications required for the competition, tasked to gain a deeper understanding of it and make it run on the Leibniz Rechenzentrum (LRZ), one of the worlds biggest HPC clusters.
Our secret weapon for the contest are our Xeon Phi coprocessor cards, multiple of which can be combined into one node that will allow us to crunch huge amounts of data with high efficiency, especially with respect to power consumption. Luckily, the Leibniz Rechenzentrum contains a Xeon Phi cluster which is very similar to the hardware we will bring to the competition and is right adjoined to our university, therefore it makes an amazing training ground.
In biweekly meetings we tried to solve the problems we encountered (which were plenty) and shared the tricks and hacks we discovered, compared our results and distributed upcoming tasks. So far, we encountered the biggest trouble with Trinity, because one of its components requires java, which our architecture does not offer, yet. On the other hand, MILC is our most favorite algorithm, since even on Xeon Phi nodes it runs very smoothly without any major adjustments. However, our motivation for the project is very strong, the more we are trying out on our systems, the more interested we become.
Our new team logo depicts all five applications.
As the competition is coming closer, we are currently engaged executing the algorithms in parallel and gaining some performance data. Also, we are steadily working on visualization techniques and polishing our media presence, e.g. with our new team logo.
In addition we might pay a visit to the Oktoberfest some time soon :)
About the Technische Universität München Team:
- All members of the team are studying Computer Science because:
- the joy of building something new, combined with an interest in computers since childhood.
- enjoyment creating new things while using innovative computer systems.
- enjoyment trying to solve problems and I'm generally interested in innovative computer systems.
- it really influences daily life and most of all, it is fascinating!
- fascination with computers and what you can do with them since childhood. Enjoyment creating new stuff and doing projects in general.
- In what other events like this have you competed? Was that as a team or individually? How did you do?
One student has previously competed in the German Federal Competition of Computer Science, which deals with algorithms in a rather theoretical way, and the International Olympics of Informatics (IOI), which also requires the algorithms implementations. Of those two competitions, he won the first one and became a bronze medalist in the latter. Also, he competed in the (International Collegiate Programming Contest) ICPC on regional level (as a team) and other programming challenges like the google codejam (Individually, advanced to top 500).
- What training routine are you following to prepare for this year's SCC15 competition?
We are meeting every other week and discuss our progress with respect to the codes. Furthermore, we have the unique possibility to access the LRZ SuperMIC cluster.
- The team distributed application responsibilities based on experience and interest. The student's report:
- My primary role is the RepastHPC (Zombie) code and the showcase, because I have already gained experience with OpenGL and Paraview.
- I am responsible for WRF and post-processing. The first one because I have a great experience with paraview and blender. My responsibly for WFR evolves from my interest in climatic simulations, which never stop to challenge me.
- I'm mainly focusing on Trinity and Linpack. Further, I am also focusing on system administration tasks with a special focus on tuning the overall system for maximum efficiency.
- My focus is on the MILC and the Trinity applications, because I'm interested in those algorithms.
- We distributed the responsibilities for the different programs across our team. I concentrate on the WRF and MILC codes.
- The student's report challenging assignments preparing for this competition.
- Getting all of WRFs additional libraries to compile fine with the intel compiler has proven to be a challenge. We're making progress, though!
- Compiling the library's for the Repast Code with the intel compiler for the xeon phi cluster.
- Making the post-processing of the WRF code was quiet challenging. Trying not to give too much away yet but this could quite possibly lead us to an advantage in the competition.
- So far the most challenging assignment is getting the programs to run on the LRZ SuperMIC.
- Installing all the trinity plugins on the SuperMIC has been the hardest part till now, but this way we now know those plugins structures and dependencies inside out.
- Some things have been easy for them:
- The easiest part was getting the Zombie code running on my laptop. It just worked as described in the ReadMe.
- I must admit getting WRF to work was easy for there were very detailed documentation on the Internet.
- Getting the programs to work on our own computers was a lot easier, since the documentation of most programs was really good.
- Running the MILC application really was a piece of cake. It just works out of the box.
- What architectural details can you share that will put fear into the hearts of your competitors?
Our system will have multiple Xeon Phi coprocessor cards per node which will allow us to crunch huge amounts of data with high efficiency, especially with respect to power consumption.
- What do you like about mysteries? How are your preparing for the mystery application?
It is the unknown people fear most, that is what makes the mystery application the real challenge. Not knowing what preparation is necessary, but believing that we can do it. It gives a chance to demonstrate the knowledge about our machine as well as our flexibility of our team. We are preparing for the mystery application by trying to make applications from previous years student cluster challenges. Further, we try to find other popular codes with interesting properties or system requirements through HPC related news and conferences. By doing this, we get more experience in working with unknown applications in general.
- Why do you want to win? What are you willing to do to win?
Our advisors and sponsors expect us to do our best. We can't let them down! I'm prepared for long hours in front of our available HPC systems to be able to utilize our final cluster setup as efficient as possible.
- What are you planning to do after the competition is over?
A few of them are going to stay in the US a bit and will get to know Texas! They're planning to visit Space Center Houston. And, besides school, there are many more competitions waiting for them!