Prof Satoshi Matsuoka

Professor Satoshi Matsuoka
Professor, Ph.D. / High Performance Computing Systems Group
Global Scientific Information and Computing Center
& Department of Mathematical and Computing Sciences
Tokyo Institute of Technology

Research Activities:
My research is principally in system software for large scale supercomputers and similar
infrastructures such as Clouds for HPC, and more recently, convergence of Big Data / AI
with HPC, as well as investigating the Post-Moore Technologies towards 2025. Over the
years I have been involved and lead a number of large collaborative projects that worked
on basic elements that are now significant for the current and more importantly future
exascale systems, such as fault tolerance, low power, strong scalability, programmability,
as well as large-scale I/O. Some of the major projects I have lead are scalable
implementations of parallel object-oriented languages on 1000CPU-scale machines in the
early 90’s (Fujitsu AP1000 and ETL EM-4), GridRPC/Ninf project in collaboration with
ETL/AIST (1995-2005), the “Titech Campus Grid” project (2002-2006) which deployed
a 1300 CPU grid-of- clusters in production mode within the Tokyo Tech campus, the
NAREGI project (2003-2007) which is a national project to develop a deployable
advanced grid middleware for national supercomputing centers, and Tsubame1 (2006-
2010), a production supercomputer which was the #1 machine in Asia for 1.5 years.
Some of the recent projects include Ultra Low Power HPC (2007-2013) which aims to
achieve 1000-fold power/performance improvement in supercomputers, Info-Plosion
(2005-2011) which aims to develop basic system software technologies for large-scale
data and information processing, Tsubame2.0 (2010-2014) which became the 4 th fastest
supercomputer in the world and the “Greenest Production Supercomputer in the World”
on the Green 500 two consecutive times, heavily utilizing GPUs and other power
efficient technologies. The most recent awards are 1) the Billion-way fault tolerance in
supercomputers (2011-2015) which will investigate the key technologies for reliable
exascale in supercomputing; 2) Ultra Green Supercomputing (2011-2015) and it
successor Post-Green Project (2015-2019) where we investigate low power control and
cooling technologies to achieve exascale power requirements, the most recent result
being design and construction of the oil-immersed Tsubame-KFC supercomputer that
became No.1 in the world on the Green 500 two consecutive editions; and 3) Extreme
Big Data (2013-2018), where we aim to converge HPC/AI and Big Data technologies in
order to be able to handle extreme-scale data that can be only processed with
supercomputers. I am in the process of designing a new 100 PetaFlops-scale AI-dedicated
Cloud Supercomputer called ABCI for the AI Research Center hosted by AIST National
Institute. Finally, we are starting research on post-Moore technologies, where we claim
that, transition from FLOPS centric means of compute acceleration will be replaced by
data or BYTE centric computing, and as such, the convergence of HPC and Big Data/AI
are more fundamental, not just because of mutual needs.