Fujitsu Wins HPC Order From Japan’s Institute of Statistical Mathematics

Fujitsu Ltd. has received a supercomputer order from Japan’s Institute of Statistical Mathematics (ISM), an interuniversity research organization.

The order is for a hybrid system comprising two supercomputers – a distributed-memory supercomputer made up of 360 PRIMERGY RX200S5 PC servers, with a theoretical peak performance of 33.7 teraflops, and a shared-memory supercomputer made up of two SPARC Enterprise M9000 UNIX servers, with a theoretical peak performance of 4 teraflops.

The supercomputer system is due to begin regular operations in January 2010.

Since its founding in 1944 as Japan’s central research organization for statistical mathematics, the ISM has opened its doors to a diverse range of researchers, and has joined forces internationally to pursue accomplishments in numerous fields.

The ISM decided to deploy a new supercomputer system in order to develop new statistical methods for extracting information and processing data amid a rapidly expanding volume of data. The system will be used to research and develop new statistical techniques such as statistical modeling, large-scale data analysis, and data assimilation (a next-generation simulation technique). It is also hoped that the system will establish methods for forecasting and knowledge discovery based on large volumes of data, as well as quantitative risk analysis, all of which can benefit society.

About the New Supercomputer System
The new system is a hybrid consisting of two supercomputers, one with distributed memory and one with shared memory, a large-scale shared storage system, a physical random-number generator, and a visualization system.

The operating environment for the two supercomputers are integrated using Parallelnavi, Fujitsu’s high-performance computing (HPC) middleware, which allows users to operate the hybrid system seamlessly as a single supercomputer.

1. Distributed-memory supercomputer, 33.7 teraflops theoretical peak performance: This system is a cluster comprised of 360 Fujitsu PRIMERGY RX200S5 nodes, each equipped with 2.93 GHz Intel Xeon Processors X5570, resulting in a cluster powered by 720 CPUs (2,880 cores). Communications between nodes is facilitated by InfiniBand(R) QDR, the latest high-speed interconnect, enabling high-performance parallel computations.
2. Shared-memory supercomputer system, 4 teraflops theoretical peak performance: This supercomputer is comprised of two Fujitsu SPARC Enterprise M9000 UNIX nodes equipped with Fujitsu SPARC64 VII processors. The large memory volume allows the system to handle large-scale data-analysis tasks.
3. Large-scale shared storage system: The disk array system employs 35 Fujitsu ETERNUS DX80 storage systems, offering a usable capacity of 1.37 petabytes.
4. Fujitsu’s Parallelnavi series HPC middleware enables high usability

The Parallelnavi series covers all aspects of the supercomputing environments and offers a high degree of usability from scientific program development to job execution and operations management.



The middleware’s major features are as follows:

• Includes a program development environment with automatic parallel compilers for multi-core processors, performance tuning tools, and parallel math libraries; and a high-speed execution environment.
• Unifies heterogeneous architectures such as distributed and shared memory systems for integrated operations. Includes resource-management job scheduler that extracts maximum performance from each platform.

As a hybrid system, the large-scale shared storage system shares data between the two different supercomputers to maximize data usability, and enables high-speed file access.