Tokyo, January 12, 2007
Fujitsu Limited today announced that it has won an order from the Kamioka Observatory of the Institute for Cosmic Ray Research (ICRR) of the University of Tokyo in an open tendering, to supply a new data analysis system that analyzes the structure of outer space by utilizing a cosmic-particle observation detector known as Super-Kamiokande.
The new system will accumulate and analyze data regarding neutrinos, based on neutrino observation data from Super-Kamiokande.
The new system will consist mainly of the following from Fujitsu: a PC cluster of PRIMERGY BX620 S3 blade servers, a PRIMEQUEST mission-critical Intel Architecture (IA) server, the ETERNUS storage system, and Parallelnavi SRFS for Linux version 1.0, a high-speed distribution file system. The PC cluster is designed to achieve computing performance levels that are 35 times greater than the existing system (see note). The new system is scheduled to be operational in March 2007.
Background on the System Overhaul
Kamioka Observatory uses Super-Kamiokande - the world's largest water Cherenkov detector for cosmic particles, installed underground in the Kamioka region of Hida city, located in Gifu prefecture of central Japan - to capture particles known as neutrinos, a type of cosmic rays that are perpetually falling from outer space to earth. The observatory is one of the world's foremost research facilities that conducts research of outer space and elementary particles, based on observation data of neutrinos. Kamioka Observatory is known for a significant number of past scientific findings, including discovering the finite mass of neutrinos.
The new data analysis system will be closely related to the Super-Kamiokande detector. Partially due to the fact that neutrinos are very difficult to capture, observation by Super-Kamiokande is perpetual, 24 hours a day for 365 days a year - thus, there is a need to e observation data as well.
In considering a new system, in addition to existing operations, Kamioka Observatory took into consideration such factors as the need to be able to transfer the system without interrupting the observations currently being run, transfer of data currently stored in a magnetic tape library to a high-density disk storage system, sufficient CPU performance, high-speed networking, high-speed high-density data storage, and whether 24-hour quick-response support could be provided.
As a result of the open tendering, Fujitsu's system consisting mainly of the following Fujitsu products was selected: a PC cluster of PRIMERGY BX620 S3 blade servers, a PRIMEQUEST mission critical IA server, ETERNUS storage system, and Parallelnavi SRFS for Linux v1.0" high-speed distributed file system.
Overview of the New System
Computation server:
PC cluster consisting of 270 “PRIMERGY BX620 S3” blade servers (540 processors, 1080 cores). The combined power of 540 processors and1080 cores enables computing performance that is 35 times that of the existing system.
Storage:
ETERNUS 4000 mid-range disk array and the ETERNUS LT270 tape library. Analyzed data that is frequently accessed is stored on disks in ETERNUS 4000, while observation data is stored in cartridge tapes in the tape library ETERNUS LT720. By using different storage methods for depending on the data purpose and needs, the system enables task efficiency.
File System Management Server:
Three PRIMEQUEST 520 mission-critical IA servers.
Parallelnavi SRFS for Linux v1.0 distributed file system is embedded, thereby enabling high-speed transfer of large volumes of data. This server realizes a high-reliability file system and high performance that is ideal for the science and technology computation community.
File Transfer Software:
Parallelnavi SRFS for Linux v1.0” high-speed distribution file system. In order to support simultaneous access all at once from aligned servers consisting of 540 processors and 1,080 cores, this software will enable data transfer performance of 750 megabytes (MB) per second, roughly twice that of the existing system. This will enable the system to realize efficiency of analysis tasks, as volume of analysis data continues to grow.
*Note: Performance increased by 35 times compared to the existing system, as measured using SPECint_rate2000.