CMD®20 Supercomputer Course

Tomoya Ono

Assistant Professor,
Research Center for Ultra-Precision Science & Technology, Osaka University

The Supercomputer Course in the CMD® Workshop is a relatively new course, which is the fourth in the 20th CMD® Workshop on March 2012. 
Out of consideration for computing resources, the supercomputer course is limited to three to four participants. It is designed to enable participants to learn through one-on-one instruction in an environment with resources that closely approximate the one in which researchers actually conduct research. Applicants to attend the course have included graduate students, post-doctoral researchers and researchers at private companies ̶ the people who will be leaders in the next generation of computational science. From among these applicants, the participants are selected based on their participation in the beginner’ s and advanced courses, their past coursework and so on.

The instructors are Ono (Osaka University) and Yoshiyuki Egami (Nagasaki University), in addition to Shoichiro Saito (Osaka University) as teaching assistant. In the Supercomputer Course, RSPACE code and STATE code are used alternately, and this time RSPACE was used. Usually the participants are asked in advance about what computing environment they would like, and the lecturers make the decision after consulting the participants. However, since the participants themselves read the academic papers prepared using RSPACE, they were able to come up with computing topics that were pertinent. Up to now, the supercomputer used for hands-on training was SX-9 at the Cybermedia Center, Osaka University, but this time Altix ICE at the Institute for Solid State Physics, the University of Tokyo was used.

The first day of the hands-on training began with an introduction to the computer. After that, the participants began hands-on training using computer code. As many of the participants had never used computer code before, initially they pursued the work while consulting the combined manual and text. RSPACE code is parallelized with both process parallelization using MPI and thread parallelization using OpenMP. Altix ICE is designed for multi-node use, so the hands-on training was conducted so as to emphasize an awareness of process parallization using MPI. As RSPACE enables computing time to be reduced by increasing the number of cores used, undoubtedly the participants were able to get a real sense of the importance of massively parallel algorithms of the kind used in next-generation supercomputers.
Judging from the participant surveys conducted after the workshop, the students had difficulty understanding nodes, CPUs and cores and other elements of computer structure, and the role of process parallelization and thread parallelization in relation to these elements. We were made keenly aware that future workshops using massively parallel computers will need to have content explaining the structure of computers in addition to instruction in physics and chemistry.
From the second day on, the mood in the workshop was like that of an actual research lab. As in other courses, there were set break times, but the participants were completely immersed in their computing tasks. Some participants took breaks between jobs, while others worked straight through without taking any breaks. It seemed that the workshop ended just as the participants had begun to get used to the computing work. We regret that there were some participants who were unable to achieve what they wanted to do in the workshop. However, in the participant surveys submitted after the Workshop, some participants were exuberant in expressing their desire to pursue joint research in the future. From the lecturers’ perspective, this made it all worthwhile.