SYSTEMS

Fujitsu Limited today announced that it has won an order to provide a new supercomputer system to the Information Technology Center at Nagoya University.

The supercomputer will be designed as a hybrid computation server system comprised of three different computer architectures: the SPARC Enterprise M9000 UNIX server, the HX600 technical computing server, and the FX1 high-end technical computing server. The combined theoretical peak performance of these three systems is 60 teraflops.

The supercomputer will be a shared system used by Nagoya University as well as other research institutions and corporations, and will begin operation in two stages, the first in May 2009 and the second in October 2009.

Decision to Deploy the New System
The Information Technology Center at Nagoya University is an academic research facility available to the university faculty along with researchers from other research institutions and corporations throughout Japan.

Up until now, the center has been using Fujitsu's PRIMEPOWER HPC2500, which was deployed in March 2005 with Japan's largest memory capacity at the time. In the past few years, however, two different types of needs have emerged: the need for more memory required for massive computations, and the need for a larger number of CPUs required for parallel computations. To accommodate both sets of needs, the Information Technology Center decided to deploy a new hybrid system comprising three different kinds of architectures, enabling it to both boost computation capacity and broaden the reach of parallel computing to more users.

Features of the New System
The new supercomputer is a hybrid system comprising three different types of computation servers. It will use Fujitsu's Parallelnavi HPC middleware, HPC Portal, and Management Portal to integrate the servers so that users experience them as a seamless, single system. The key features of the servers are as follows.

Large-scale SMP computation server (System 1)

Hardware: SPARC Enterprise M9000 UNIX server
Nodes: 3 (96 CPUs, 384 cores)
Total memory: 3 terabytes
Theoretical peak performance: 3.84 teraflops
This system is the successor to the current PRIMEPOWER HPC2500, using SPARC64(TM) VII quad-core processors developed by Fujitsu in a large-scale SMP configuration. Users of existing systems will be able to easily migrate their applications to the new system. With one terabyte of shared memory per node, the new system can perform extremely large-scale computations.

Large-scale distributed Linux computation server (System 2)

Hardware: HX600 technical computing server
Nodes: 160 (640 CPUs, 2,560 cores)
Total memory: 10 terabytes
Theoretical peak performance: 25.6 teraflops
This is a Linux-based, large-scale PC cluster based on "open supercomputer" specifications using off-the-shelf technology within an x86 architecture. As a general-purpose system, it can run a wide range of application software, making it accessible to many researchers and broadening its reach.

Large-scale distributed UNIX computation server (System 3)

Hardware: FX1 high-end technical computing server
Nodes: 768 (768 CPUs, 3,072 cores)
Total memory: 24 terabytes
Theoretical peak performance: 30.72 teraflops
This is a large-scale cluster system using the FX1 high-end technical computing server. With a chipset designed for high-memory bandwidth suited for SPARC64 VII and the high-performance compiler, the system is able to reach an extraordinarily high level of computing efficiency that is difficult to achieve with conventional multi-core systems. The system will be used for developing applications required for future high-performance computing such as multi-core support and high degrees of parallelism.

Additionally, the supercomputer will include an ETERNUS2000 model 200 disk array storage system with 1.15 petabytes of physical storage.

Katsuya Ishii, Professor, Information Technology Center, Nagoya University: "To meet the diverse needs of users from various research fields, Nagoya University decided to install the hybrid supercomputer system comprising three different architectures. Similar to our previous system built on Fujitsu's PRIMEPOWER HPC2500, the new system is equipped with high processing capacity as well as large memory and able to perform massive computation and large-scale data processing.

We trust Fujitsu's experience in hardware and software development and look forward to their continued support to ensure the smooth functioning of the hybrid system. Through the new supercomputer system, the Information Technology Center hopes to make further contributions to the advancement of science and technology as well as to the education of young researchers."