Supercomputer Office, Security and Information Systems Department, Japan Aerospace Exploration Agency (JAXA)
Supercomputer Office, Security and Information Systems Department, Japan Aerospace Exploration Agency (JAXA)
Supercomputer Office, Security and Information Systems Department, Japan Aerospace Exploration Agency (JAXA)
出版者
宇宙航空研究開発機構(JAXA)
出版者（英）
Japan Aerospace Exploration Agency (JAXA)
雑誌名
宇宙航空研究開発機構研究開発資料
雑誌名(英)
JAXA Research and Development Memorandum
巻
JAXA-RM-14-011E
ページ
1 - 12
発行年
2015-03-19
抄録（英）
Multicore processing has become a major trend in high-performance computing. However, there are two major issues with multicore technologies: 1) how to make full use of modern multicore CPU capability with reduced user workload, and 2) how to achieve better scalability up to a high degree of parallelism for memory-intensive applications such as computational fluid dynamics. First, the multicore-based massively parallel cluster of the Japan Aerospace Exploration Agency (JAXA) supercomputer system, which uses the Fujitsu FX1 as the core computer, is described. Notable features such as a high-speed barrier network, enhanced memory bandwidth, and an integrated mechanism to hide latency, i.e., the integrated multicore parallel architecture, which comprises a hardware barrier, L2 cache shared among cores, and an associated automated parallelization complier, are described. Second, the process to achieve significantly high sustained performance (>90%, high-performance Linpack benchmark) on the multicore CPU cluster system is explained. Finally, performance measurement results for JAXA applications are provided.
内容記述
形態: カラー図版あり
内容記述（英）
Physical characteristics: Original contains color illustrations
High Sustained Performance and Scalability on a Multicore-Based Massively Parallel Cluster of JAXA Supercomputer System
AA1530011000.pdf
(1.26MB)
