OpenHPC member RIKEN is Japan’s largest research institution. It has approximately 3000 scientists on seven campuses across Japan, the main one just outside of Tokyo. The full Japanese name of RIKEN is Rikagaku Kenkyūsho (理化学研究所), which means “The Institute of Physical and Chemical Research,” though it now also conducts research in biology and other fields. It’s commonly abbreviated as RIKEN (理研). It is almost entirely funded by the Japanese government, and its annual budget is approximately ¥88 billion (US$760 million).
In June 2011, TOP500 ranked K the world’s fastest supercomputer, with a computation speed of over 8 petaflops. And in November 2011, K became the first computer to top 10 petaflops. In June 2012, K was superseded as the world’s fastest supercomputer by the American IBM Sequoia, and as of November 2015, K is the world’s fourth-fastest computer.
We recently had a chance to talk to Yutaka Ishikawa about RIKEN. Ishikawa is the leader of the FLAGSHIP 2020 Project at RIKEN Advanced Institute for Computational Science (AICS). He received PhD degree in electrical engineering from Keio University and has taught at the University of Tokyo for many years and run multiple supercomputing projects in Japan. From 2011 to 2014, he was the director of Information Technology Center at University of Tokyo, where two supercomputers are installed to serve academic and industrial users. He moved to RIKEN AICS in October 2014.
What is your experience or background in HPC?
I’ve worked in quite a few different areas in HPC including the SCore cluster system, designing commodity-based supercomputers, developing GridMPI, working on national infrastructure projects, and building a manycore-based cluster system whose peak performance will be more than 20PF.
(1) Cluster System Software from 1993 to early 2000s:
From 1993 to 2001, I was the chief of Parallel and Distributed System Software Laboratory at the Real World Computing Partnership where a national project for high performance computing was performed. I led development of cluster system software called SCore, which was used in several large PC cluster systems around 2004, e.g., University of Tsukuba, RIKEN, and TITECH.
(2) Design of Commodity-based Supercomputers
From 2002 to 2014, I was a professor at the University of Tokyo. I led a project to design a commodity-based supercomputer called T2K open supercomputer. As a result, three universities, Tsukuba, Tokyo, and Kyoto, obtained each supercomputer based on the specifications in 2008.
(3) Development of MPI library for Grid (2002-2007)
GridMPI is an MPI-2 implementation for the Grid environment where both network latency and bandwidth have to be taken into account to provide the high performance communication environment. GridMPI has a layer called the latency-aware communication, which optimizes communication with non-uniform latency and hides the various lower-level communication libraries.
(4) National HPC Infrastructure
I led the design and implementation of HPCI, the national High Performance Computing Infrastructure in Japan, from 2010 to 2012. HPCI has been officially operating since September of 2012. The K computer and more than 9 domestic supercomputers located at various universities and laboratories, and a distributed shared storage system are accessed with the single sign-on environment.
(5) Manycore-based supercomputer and its OS
I have been involved with the design of the post T2K machine since 2013, which will be operated by the University of Tokyo and University of Tsukuba in 2016. It will be a manycore-based cluster system whose peak performance will be more than 20PF. The operating system is based on a Linux kernel with a lightweight microkernel, called McKernel [M-C-Kernel]. McKernel is based on my Ph.D student work, and it has been taken over by RIKEN AICS. McKernel runs on Xeon, Xeon Phi, and Fujitsu FX10.
Please tell us about RIKEN’s mission.
RIKEN is in charge of the development of the successor to the K computer, called “post K,” scheduled to begin operation in April 2020. RIKEN has selected Fujitsu as the hardware vendor. RIKEN is responsible for some parts of the system software stack for post K including McKernel, MPICH implementation, file I/O middleware, XMP programming language, and domain specific libraries. Software developed at RIKEN is open source.
To understand what our group has been focusing on, please visit the following URL:
This is my invited talk describing our system software development from SC15.
Why did RIKEN want to participate in OpenHPC?
We believe that open source software plays a very important part in supercomputing progress. We would like to not only use open source software, but to also contribute to the open source community by providing our open source software running on large-scale supercomputers. It’s important to give back to the community.
What do you do on weekends?
My hometown is Tsukuba, but I am now living in Kobe because RIKEN AICS is located there. We have a Tokyo office, and I often commute from Kobe because the government and Fujitsu offices are in Tokyo. It takes about 3 hours!
When I am in Tsukuba, I’m gardening. During winter, it’s cabbages, Chinese cabbages, and onions. Unfortunately, because I have been mainly in Kobe, I have not had a chance to eat my cabbages yet. The onions are still growing. I expect to be able to cook and eat them this May.
Kobe faces both the ocean and the mountains and has many temples. I enjoy hiking or walking around the Kobe area to see temples and traditional Japanese streets.