The University of Tokyo, Kyushu University, Fujitsu, Hitachi, and NEC in cooperation with RIKEN AICS started a two-year project funded by MEXT (Ministry of Education, Culture, Sports, Science and Technologies), Japan, where a future supercomputer system towards an exascale supercomputer is studied. It is assumed that the first target system will be deployed around 2018 with up to 30 MW and 2000 m2 constraints. The purpose of this project is to design a system by 2018 and reveal what kind of researches and developments must be taken into account.tion
As the first step to study feasibility of a new architecture, applications of material and climate sciences were selected in 2012. Based on the K computer architecture, a new architecture has been designed and evaluated to cover a wide range of applications. At the same time, the system software stack for both the new architecture and commodity-based machines has been designed in order to assure general versatility.
Organization and Role
|The University of Tokyo||Yutaka Ishikawa||PI and Overall software design|
|Takahiro Katagiri||Tuning and evaluation of applications|
|Kei Hiraki||Co-PI for evaluation of architecture and compiler|
|Hiroshi Nakamura||Study on Low power technologies|
|Kyushu University||Mutsumi Aoyagi||Co-PI for providing network evaluation environment|
|Fujitsu||Naoki Shinjo||Co-PI for overall architecture and its software design|
|Hitachi||Tsuneo Iida||Co-PI for storage and system software design|
|NEC||Yuichi Nakamura||Co-PI for system software design|
|RIKEN AICS (Cooperation)||System software design and evaluation of applications|
Our study is being carried out by four intensively cooperative cross-organizational teams; Architecture Design, Application Tuning, Architecture Evaluation, and System Software Design.
The Architecture Design team provides its architectural parameters for the Application Tuning and System Software Teams. The Application Tuning team tunes the selected applications and measures performances using these architectural parameters. The System Software Team designs and develops the prototype of system software for both the new architecture and commodity-based machines. The Architecture Evaluation team examines the evaluation results and gives architectural feedback to the Architecture Design team for improvement. This evaluation process is a bimonthly cycle. Twelve application benchmarks will be eventually selected to evaluate the architecture.
The figure below shows how the Application Tuning team works. They will identify hotspot loops of the selected applications and collect performance accounting information by running them on Fujitsu FX10, a product using enhanced technologies implemented in the K computer. The target system performance is predicted using performance evaluation tools provided from the Architecture Design team.
Background of Feasibility Study
In 2011, the Japanese Government started to discuss future Japanese HPC infrastructure, so-called HPCI. MEXT organized two working groups under the Council on HPCI Plan and Promotion, an application team and an architecture team. The application team examined the science roadmap to challenge respective key socio-scientific problems in 2020 and the requirements for HPC systems to produce valuable results. The architecture team studied key technologies to build an HPC system usable in 2018 to satisfy the requirements in the science roadmap. The two teams cooperatively wrote up the Report on Strategic Direction/Development of HPC.
In the chart included in this report, referred above, required memory bandwidth and capacity for each application are plotted by ‘X’. Legends are GP (General purpose), CB (Capacity-bandwidth oriented), RM (Reduced memory), and CO (Compute oriented).We are studying a GP type system in this project.
Eight variations of the architecture have been designed and evaluated using some of twelve application benchmarks, and three variations of the architecture have been selected for this study. By March of 2014, the proposed architecture will be evaluated using all twelve application benchmarks.