Interdisciplinary large-scale computer system
 
   
 Our Center was providing nationwide services with the supercomputer system and the general-purpose computer system as a Joint Usage/Research Center. Subsequently those two computer systems were integrated as Interdisciplinary Large-scale Computing System. The integrated system has been utilized for various services since November 2011.
 
 The Interdisciplinary Large-scale Computing System consists of a supercomputer system, a cloud system, a network system and a system for the management of users and peripheral devices. In April 2014, we also introduced Petabyte-class Data Science Unified Cloud Storage System in order to reinforce the facilities and enlarge the storage capacity, which was intended for a big-data processing. All of the systems are integrally operated for providing computation, application, file and hosting services and many other services. As core computer systems in a national joint facility, the systems as a whole serve as a comprehensive information system that supports researchers in the private sectors as well as in the academic fields.

 

  
Supercomputer System
 
 The supercomputer system adopts HITACHI SR16000 Model M1 and the IBM Power 7 CPU, whose clock speed is 3.83 GHz. The server rack is as shown in the photo, and the entire system consists of three racks. The 22 physical processing units are divided into 8 arithmetic and logic units, which are referred to as “nodes” below, in order to provide computing services. The number of physical cores per node is 32, and the main storage unit has a capacity of 128 GB. Software assets built in the past system is possible to be used, and high-speed processing is realized.
 
 The total performance is 172 TFLOPS. To provide various services, 2 of the 176 nodes are used as TSS nodes and 166 are used as batch nodes. The remaining 8 nodes are used for system management. The TSS nodes are available for program development/editing, compiling and small-scale SMP/ MPI jobs. Of the 166 batch nodes, 38 are used to run SMP jobs and/or medium-scale MPI jobs to provide computing services whose maximum number of nodes used per job is 16 and maximum elapsed time is 72 hours. Another 128 batch nodes are used to provide large-scale ultrafast computing service on a steady basis by running large-scale MPI jobs. For such an MPI job, up to 128 nodes, or 4,096 physical cores, and a main storage capacity of 16 TB are used, and the maximum elapsed time per job is 24 hours. Such a large-scale computing services are unprecedented, and this is a distinguishing feature of computing services available with a supercomputer system.
 
 Compilers available include HITACHI optimizing FORTRAN as well as XL Fortran and C/C++ of IBM. The external storage device connected to the supercomputer system via a fiber channel is the HITACHI AMS2500, whose total effective capacity consists of 600 TB (SAS, RAID5) and 300 TB (SATA RAID6). Safe, secure storage of valuable data is ensured by RAID5 and also by continuous full backup of the SAS domain, where up to 400 TB of capacity is reserved for each user area.

 
Cloud System
 
 In the Hokkaido University Academic Cloud, full virtualization is adopted and a cloud middleware is introduced in order to realize the largest academic cloud system in Japan, one that enables more than 2,000 virtual machines to be managed.
 
Hardware
 
■ HITACHI Blade Symphony BS2000
 This blade server consists of 114 nodes. Each node has an Intel Xeon E7-8870 processor x 4, or 40 cores in total, shared memory of 128GB and a network interface of 40GbE x 2. To this blade server, a file system with the capacity of 260TB is connected via SAN (Storage Area Network) as a bootable storage. The large-scale cloud system with the peak performance of 43TFLOPS is comparable to a supercomputer.
 
■ HITACHI AMS2300
 The HITACHI AM2300 is a large storage system with a capacity of 500 TB, serving as shared storage for the cloud system. It is used as an additional file area for virtual machines, an online storage server, and a user area for the use of applications.
 
■ Citrix CloudStack / XenServer / VMWare vSphere
 Citrix CloudStack is adopted as an infrastructure middleware for the cloud system. At a web portal, users can apply for the use of, configure, and manage the operations of virtual machines (i.e. virtualized servers) in a unified way. A virtualized server is made available immediately after an application is filed.
 
 Infrastructure software for virtualization includes Citrix XenServer and VMWare vSphere. These make it possible to build many virtual machines in one physical server, ensure flexible operations management, reduce costs and substantinally improve energy efficiency.
 
Petabyte-class Data Science Unified Cloud Storage
 
 This system was introduced to strengthen the data federation/processing functions with supercomputers and other cloud systems by reinforcing the cloud service infrastructures for HPCI.
 
 The system accelerates processing of a huge amount of data accompanying a large-scale scientific computation, and supports R&D involving such data science as big-data processing.
 
 The system is divided into two subsystems: Data Science Cloud System and Cloud-integrated Storage System.
 
Hardware
 
■ Data Science Cloud System
 The Data Science Cloud System provides computational resources. It consists of 25 nodes of HITACHI HA8000/RS210, each of which has two Intel Xeon E5-2670v2 processors (2.5GHz, 10 cores), memory of 80GB and HDD of 2.4TB. In addition CPU-to-memory bandwidth and CPU-to-CPU bandwidth on each node are both 512Gbps.
 
 In order to realize a cloud computing infrastructure, Citrix XenServer 6.2 and Apache CloudStack 4.2 were adopted, the former is an infrastructure software for virtualization and the latter is a middleware for the cloud system. These software products make it possible to deploy and utilize various virtual machines on demand. These operations can be performed easily via a web browser.
 
■ Cloud-integrated Storage System
 The Cloud-integrated Storage System consists of WOS7000 appliances and a GridScaler NAS manufactured by DataDirect Networks, Inc. (DDN), and has a large capacity of 1.96PB.
 
 The system provides storage area for WebDAV, Gfarm file system and Amazon S3-compatible object storage, and NFS disk space for the Data Science Cloud System as well.
 
Hosting Server
 
 The hosting server services provide virtual machines that can be used for Web hosting, etc. Additionally, server packages on which content management systems such as MediaWiki, Movable Type, etc. are installed are provided. Because the server packages are available for immediate use, users do not have to install software. The adoption of High-Availability (HA) architecture ensures automatic switchover to a backup server in the event of a failure; thus, it improves the reliability of the hosting server.
 
Project Server
 
 The project server services provide virtual machines which users can use for research projects. With this server, customization at the OS level or at other levels is available to users. Virtual machine service levels are as shown below:
 
  • S server: 1 core, 3 GB memory, and 100 GB HDD
  • M server: 4 cores, 12 GB memory, and 100 GB HDD
  • L server: 10 cores, 30 GB memory, and 100 GB HDD
  • XL server (physical server): 40 cores, 128 GB memory, and 2 TB HDD
 In addition to a standard OS installation image, MPI and Hadoop are automatically installed in order to provide a cluster package in which multiple servers are configured as a cluster. Application for an additional disk is also available starting in units of 1TB (excluding XL server.)
 
Application Software
 
 Gaussian09 for the supercomputer system. Mathematica*, MATLAB*, ANSYS*, MD.Nastran*, Patran*, Marc*, Marc Mentat*, LS-DYNA, ANSYS FLUENT*, Gaussian09, Amber, COMSOL Multiphysics, SAS*, and AVS/Express Developer for Application servers.
 
*available only to researchers and students in Hokkaido University

 
Various Services & User Support
 
 The Interdisciplinary Large-scale Computing System, facilities for nationwide joint use, are available to graduate students and teachers at universities, colleges and institute of National Colleges of Technology for computing and information processing that is necessary for their academic research. Since April 2010, private companies also
 
  • Large-scale computing services
    Large-scale and high-speed numerical data processing is made available by the supercomputer.
  • Application services
    Various types of application software run on application servers can be executed via Web browsers.
  • Online storage services / WebDAV storage services
    Data can be transferred via Web browser, and the operation required makes it very easy to upload and download files.
  • Hosting services
    Virtual machines installed with MediaWiki or other content management systems are lent out.
  • Project services
    Virtual machines are lent out for research projects. Parallel computation with multiple machines is possible. Physical servers are also lent out.
  • Petabyte-class Data Science Unified Cloud Storage
    Virtual machines are lent out. Parallel computation with multiple machines is possible. Physical servers are also lent out. These services are mainly intended for researchers whose projects are adopted by HPCI/JHPCN.
  • Download services
    The formula manipulation software Mathematica and the data visualization software AVS/Express are available for down loading to PCs in laboratories (for use on campus only) .
  • User homepage services
    Users can disclose content of their websites online just by uploading the content to Web servers.
  • Printing services
    Color printing on A0-, A1- and B0-sized paper is possible. Printing on cloth is possible, as well as on plain and glossy paper.
  • User supports
    Users can inquire about the use of computer systems, the creation and updating of programs, and the like. Regarding large-scale computer systems, questions and requests are accepted by phone and e-mail. The questions and requests are disclosed on the Information Initiative Center’s website in question-and-answer format.