Feasibility Studies on Next-Generation Supercomputing Infrastructures (Aug 2022- )

TOP Research Future Developments Feasibility Studies on Next-Generation Supercomputing Infrastructures (Aug 2022- )

RIKEN has been selected as one of the representative institutions for the Feasibility Studies on next-generation supercomputing infrastructures (system research) by the Ministry of Education, Culture, Sports, Science and Technology (MEXT).

The purpose of this project is to propose options for the next-generation computing infrastructure for the post-"Fugaku" era, taking into account the needs of science, industry, and society, and to propose a next-generation system that can serve as a research DX infrastructure with an advanced digital twin, in order to realize a platform for solving issues for the realization of broad SDGs and Society 5.0.

To this end, we will survey the architecture, system software libraries, and applications of the next-generation computing infrastructure, and examine elemental technologies to identify the performance and functional requirements for the next-generation computing infrastructure, while identifying technical issues and limiting factors.

Project Overview

The next-generation computational infrastructure is expected to become a platform for realizing SDGs and Society 5.0 by providing advanced digital twins that will bring "Research DX" in the science. Aiming to realize a versatile computing infrastructure that can execute entire workflow by making full use of wide range of computational methods, simulation techniques, and BigData at scale, we conduct a holistic investigation on architecture, system software and library technologies through co-design with applications.

As a basic principle of system design, we practice the "FLOPS to Byte" concept from architecture development to algorithm or application design to streamline data transfer and computation under power constraints, while taking necessary computing accuracy into consideration. Under the ALL JAPAN team composition, we will investigate system configurations and elementary technologies which improve effective performance of the next-generation computing infrastructure.

Subject of Investigation

Research on Architecture

Investigating technological possibilities (such as 3D stacked mem, accelerators, chip-to-chip direct optical link) and performance of the entire system or its components based on trends in semiconductor and packaging technologies
Predicting future system performance based on performance analysis of benchmark sets provided by Application Research Group, and feeding back to next-generation application development

Research on System Software and Library

Drawing roadmap for future system software development in Japan, specially considering data utilization enhancement, integration of AI technology with first-principles simulation, real-time data processing, and assurance of high security

Research on Applications

Building a broad benchmark set to evaluate multiple architecture choices while considering improvements in algorithms and parameters of application based on the results of architectural evaluations and exploratory "what-if" performance analysis
Investigating what classes of algorithms are expected to evolve significantly for future systems

Investigation Schedule

Investigation Schedule
Time	Architecture	System Software	Application
H2 2022 - H1 2023	Explore device/arch technology	Examine existing SW and its utilization	Examine existing apps and benchmark design
H2 2023 - H1 2024	Performance estimation with benchmarks	Identify requirement of SW development	Perf. analysis by benchmark evaluation
H1 2024 - H2 2024	Arch selection and their R&D	Draw roadmap	Study for target science

Organization Chart

Architecture Research Group

Architecture Research Group: RIKEN R-CCS
Architecture Research sub-G1: RIKEN BDR
Architecture Research sub-G2: Fujitsu Ltd.
Architecture Research sub-G3: Intel Corporation
Architecture Research sub-G4: AMD Inc.
Architecture Research sub-G5: NVIDIA Corporation
Architecture Research sub-G6: Hewlett Packard Enterprise
Architecture Research sub-G7: Arm Ltd.

System Software and Library Research Group

System Software and Library Research Group: RIKEN R-CCS
Support on Group Management: Nagoya Univ.
Scheduler / Runtime sub-G: Tohoku Univ.
IO / Storage / Filesystem sub-G: Univ. Tsukuba
Storage Archi Pattern Investigation: DDN Japan
OS / Virtualization / Cloud sub-G: National Institute of Informatics
HPC Env. Usage Investigation sub-G: Osaka Univ.
Communication Library sub-G: Kyushu Univ.
Compiler / Programming-model sub-G: RIKEN
Numerical Library sub-G: RIKEN
AI Framework sub-G: RIKEN

Application Research Group

Life Science App. Area sub-G: Yokohama City Univ.
Material and Energy App. Area sub-G: NIMS
Weather/Climate Sci. App. Area sub-G: JAMSTEC
Disaster Prevention App. Area sub-G: Univ. Tokyo
Manufacturing App. Area sub-G: RIKEN
Support on Manufacturing Apps: JAXA
Fundamental Science App. Area sub-G: RIKEN
Support on Space / Planet Sci. Apps: NAOJ
Social Science App. Area sub-G: RIKEN
Digital-twin / Society5.0 App. Area sub-G: Univ. Tokyo
Support on Digital-twin Apps: Japan Atomic Energy Agency
Weather Model Perf Analysis sub-G: RIKEN
Support on Weather Model Analysis: Meteorological Research Institute
Computational Science Algorithm sub-G: Univ. Tsukuba
Machine Learning Algorithm sub-G: TiTech
Benchmark Construction sub-G: RIKEN
Performance Modeling sub-G: RIKEN