The massively parallel cluster PACS CS

[image]

The PACS-CS system is a PC Cluster system which consists of 2,560 nodes, connected by 20,480 Gigabit Ethernet cables. The system achieved 10.35 TFLOPS in the Linpack Benchmark, ranking 34th on the June 2006 Top 500 list. To support large-scale parallel computation of computational science, the system was designed in the bandwidth-aware way: a node is equipped with a single CPU for memory bandwidth, and the interconnection network for parallel processing is configured as a multi-dimensional Hyper-Crossbar Network based on trunking of Gigabit Ethernet for high network bandwidth and cost-performance. The project aims not just developing a new cluster computer for computational sciences, but it also tried to foster interdisciplinary collaborations.

T2K Tsukuba system

T2K-Tsukuba is a very large scale PC cluster with 10,368 cores to provide 95.4 TFLOPS of peak performance. To execute wide variety of science computation, each computation node is configured as a fat-style PC server with 16 cores and 32 GB main memory. A large scale shared file system is provided by Lustre with 800 TB of user space under RAID6 physical file system, which is connected to all computation nodes via Infiniband to provide 16 GB/s of I/O bandwidth. T2K-Tsukuba started its full operation from October 2008, and currently utilized for more than 30 of large scale computing projects. The system is ranked at 20th in Top 500 list on June 2008.

[image]

FIRST simulator

The FIRST simulator is a hybrid PC cluster, where a newly-developed board for gravity calculations, Blade-GRAPE X64, is embedded in each node through PCI-X bus in a PC cluster. The theoretical peak performance of Blade-GRAPE is 136.8 GFLOPS. The peak performance of FIRST simulator is 36.1 TFLOPS, where the host PC cluster is 3.1 TFLOPS and the Blade-GRAPEs are 33 TFLOPS. All nodes are connected uniformly with each other via multi-port Gigabit Ethernet interconnect switch.

[image]

[image]

Interdisciplinary Computational Science Promotion Programs

The Center of Computational Sciences is an inter-university facility, open to researchers of universities and other research institutes in Japan. The Center solicits proposals for "Interdisciplinary Computational Science Promotion Programs" to encourage research activities of computational sciences, making the computational power of the PACS-CS, T2K-Tsukuba and FIRST to researchers throughout Japan.

The interdisciplinary Computational Science Promotion Programs have three categories:

  1. Interdisciplinary collaboration program
    • The Interdisciplinary collaboration program is to promote interdisciplinary research activities of different disciplines.
  2. Large-scale scientific simulation program
    • The Large-scale scientific simulation program is to push forward the grand challenge of several fields in computational sciences by providing the computational power of the PACS-CS and T2K-Tsukuba.
  3. Joint research program
    • The Joint research program is to promote the development of the methodology of computational sciences and advanced researches with the staffs of the center for computational sciences.

Large Scale General Use

The Center solicits proposals for "Large-Scale General Use" to provide a half of the computational power of the T2K-Tsukuba to researchers throughout Japan.

The Large-scale general use has three categories:

  1. Priority use
    • The permitted number of nodes is assigned by top priority to each user group.
  2. Dedicated use
    • The specific number of nodes is always dedicated.
  3. Best effort use
    • Jobs are executed only when resources are available.
Center for Computational Sciences, University of Tsukuba