理化学研究所 計算科学研究機構

Events/Documents イベント・広報

International HPC Summer School 2017開催報告(6月25日~6月30日)

2016年に引き続き、理化学研究所計算科学研究機構では、若手研究者を対象としたHPC分野における国際サマースクール“8th International Summer School on HPC Challenges in Computational Sciences”を開催しました。本スクールはHPC関連分野において国際的に活躍できる人材を育成することを目的に、各サイエンス分野における最先端の講演と、実際に大型計算機を利用してプログラミングや可視化などを行うハンズオンセッションが行われました。





共通のテーマにおいて計算速度を競うParallel Programming Challengeでは、日本からの参加者のJiachao Zhangさんが、複数のGPUを利用しOpenACCとMPIを組み合わせたプログラムでGold Winnerに輝きました。







参加者の感想 ― 並列プログラミングチャレンジ (Parallel Programming Challenge)のGold Winner

Comments on winning the 3rd Annual IHPCSS Challenge
ZHANG Jiachao

During this summer school, the 3rd Annual IHPCSS Challenge program was held, and I was fortunate to win the award this time.

The challenge program involves performing hybrid parallelism, using MPI plus OpenMP or OpenACC, to speed up a typical Laplace Solver program under the limitation of assigned nodes. I worked as a team with Stefan Hofmeister, a friend from Germany who I met at this summer school. Since there are four nodes available and each node equips with two GPUs, we decided to use MPI and OpenACC to accelerate this program. The efforts included utilizing all eight GPUs via OpenACC compiled by PGI compiler; utilizing the pgi_mvapich library, which perhaps supports more advanced interconnection; utilizing overlapping of communication and computing and some revision in MPI. Finally, the performance was around 6.5 seconds in the challenge and this result was enough to win the championship.

Through this challenge, I benefited a lot. Firstly, it enabled me to understand the meaning of general-purpose acceleration, which is applicable to various applications in other fields. It encouraged me to focus on strengthening my foundation knowledge in computer science, as I hope to devote more time to HPC projects in future. Secondly, I experienced using batch processing to debug programs in a supercomputer. By experiencing the advantages and disadvantages of batch processing, I can gain a better understanding of my own research, interactive simulation systems, which assume computing resource is used in real-time. Thirdly, it reminded me to expect the unexpected and of the importance of team work. I realized it’s often necessary to just try and don’t give up. Although free time is very limited for this challenge in summer school, spend a bit more time because it’s possible to optimize a little more for the purpose of studying.

I really appreciate the precious opportunity to attend this summer school supported by RIKEN, JAPAN. I learned a lot from the presentations, lectures, mentoring and other activities. And the most important part is that I had the chance to talk with many experienced professors and talented young friends. The things I learned will last with me for a long time and will encourage me to try my best to study and work in HPC in the future.