SYSTEMS

Massively parallel system will require only 15% the watts per teraflop of the current system: Fujitsu Limited today announced that it has received an order from Kyoto University to construct a next-generation supercomputer comprising a large-scale cluster system using the Fujitsu HX600 high-performance computing (HPC) server. The system complies with the T2K Open Supercomputer Specification. The main part will be a massively parallel cluster that connects 416 nodes of the HX600 HPC server. The cluster will possess a theoretical peak performance of 61.2 teraflops. In addition to a significant increase in sheer processing power, the new system will require only 15% the watts per teraflop of the current system while its overall power consumption will remain the same. The supercomputer is one of many new network and computing products offered under Fujitsu's Green Policy Innovation project. Announced in December 2007, the project aims to help customers significantly lessen their environmental burden. Kyoto University plans to use the system as its core computational platform to serve a variety of needs—not only for the scientific computations that its current system is used for, but also for massive genomic information processing, ultra-large archive searches, computer graphics, and other new areas of research. Furthermore, the system will be utilized to facilitate grid technology research which is carried out collaboratively by Kyoto University, the University of Tokyo, and the University of Tsukuba. The Road to the New Supercomputer Kyoto University's Academic Center for Computing and Media Studies is one of the seven members of the National Joint Information Infrastructure Center. Members make their computing resources available to academic and research institutions throughout Japan. The university had recently committed to updating its system in order to respond to the growing need for computing power from inside the university as well as from other universities and research institutions. The current supercomputer, a cluster using the Fujitsu PRIMEPOWER HPC2500, is to be replaced by a large cluster, which will reach a maximum theoretical performance of more than seven times the current system, with triple the memory capacity and more than one hundred times the total storage capacity. Overview of the New Supercomputer The features of Kyoto University's new supercomputer are outlined below:

1. Massively Parallel Computation Server: Each HX600 HPC server is loaded with four quad-core AMD Opteron processors. With 416 nodes, the system is capable of massively parallel computations, enabling it to handle the computational demands required by new applications. In addition, Fujitsu's Parallelnavi parallel-processing software, which provides the development and execution environments for parallel programs, will enable integrated monitoring and control over the entire system with a unified interface, thereby reducing the customer's administrative overhead and providing fine-grained control over unused resources—not merely at the server level but at the CPU/CPU core level—making it possible to put these resources into standby mode to conserve power.
2. Large-scale Symmetric Multiprocessing Server Cluster: The system will also include a large-scale symmetric multiprocessing server cluster consisting of seven SPARC Enterprise Unix servers, each of which includes SPARC64 processors (128 CPU cores) and 1 TB of memory, giving a massive memory space for large-scale computing requirements.
3. Large Storage: The disk array consists of 32 Fujitsu ETERNUS2000 model 200s, providing a total storage capacity of 800 TB. A PRIMEQUEST 580 IA server is used for the file I/O controller, and it can handle rapid transfers with high reliability.
4. High Speed Data Network: The computation servers and the large storage are connected using high speed InfiniBand interconnects. Multi-linked connections result in a redundant data network with fast connections.

Comment from Hiroshi Nakashima, Professor at Kyoto University's Academic Center for Computing and Media Studies: "This new supercomputer system upgrade represents a significant departure from our past procurement practices. This time we laid out the basic concept of the new supercomputer along with general specifications based on the T2K Open Supercomputer Specification jointly created with the University of Tokyo and University of Tsukuba, and we asked vendors to develop systems and submit proposals to meet our requirements. In this sense, I think it was more difficult this time to develop systems proposals that met our requirements, but this HX600 cluster exceeds our baseline requirements for performance and functionality, so we're very pleased with the outcome. The SPARC Enterprise M9000 sub-system will be able to directly utilize our existing software assets, and the high-capacity storage system can also be configured to exceed our base requirements. The next step, of course, is to get the system installed and running smoothly, and then to leverage our supercomputer to generate outstanding research results and give us new educational insights. We look forward to the ongoing support of Fujitsu."