Tokyo Institute of Technology - DOC by MicrosoftCaseStudies

VIEWS: 20 PAGES: 2

More Info
									                                              Microsoft Server Product Portfolio
                                              Customer Solution Case Study




                                              Technology Institute Expands Research
                                              Initiatives with High-Performance Computing


                                              “With Windows, there’s predictability in how the tools
Customer: Tokyo Institute of Technology
Web Site: www.titech.ac.jp
                                              look, feel, and work, which makes it easier to run jobs
Customer Size: 1,759 employees                and helps us accomplish more in the same amount of
Country or Region: Japan
Industry: Education—Higher education
                                              time.”
                                              Naoya Maruyama, Postdoctoral Research Fellow, Global Scientific Information and Computing Center,
Customer Profile                              Tokyo Institute of Technology
With a 125-year history, Tokyo Institute of
Technology is the largest science and
technology university in Japan. The
                                              Tokyo Institute of Technology uses high-performance computing
institute currently serves approximately      (HPC) to accelerate its research. It deployed a 32-node Windows®
10,000 students each year.
                                              HPC Server 2008 GPU-accelerated cluster to take advantage of
Software and Services                         enhanced tools to configure graphics processing units, which are
 Microsoft Server Product Portfolio
  − Windows HPC Server 2008
                                              critical to its research. The institute has found that the HPC
                                              solution provides intuitive, integrated tools for greater configuration
Hardware
 Custom assembly of hardware
                                              flexibility, straightforward deployment, and easy management.
  components, with:
  − AMD Phenom X4 9850 quad-core
    processors                                Business Needs                                     supercomputer, which consists of a 10,000-
  − 8 GB RAM                                  Tokyo Institute of Technology (Tokyo Tech)         core, 64-bit SUSE Linux cluster.
  − Four NVIDIA GeForce 8800 GPUs for         promotes advanced research across the
    each node                                 spectrum of scientific and technological           A significant research focus for the GSIC
  − GigE interconnects                        fields. Through its world-class research,          involves using HPC for new technologies,
  − DATA 500 GB hard drives                   Tokyo Tech works to pioneer new forms of           specifically, how to best use graphics
                                              scientific knowledge and technology.               processing units (GPUs) and how to reduce
                                                                                                 power consumption when using GPUs. The
                                              One of the research centers at Tokyo Tech is       center’s General Purpose GPU research
                                              the Global Scientific Information and              eventually will be applied to a range of
                                              Computing Center (GSIC), which supports            scientific applications.
                                              research and education by optimizing the
                                              most innovative information technologies.          To conduct its research, the GSIC needed to
For more information about other Microsoft    The GSIC maintains several supercomputers          experiment with the maximum number of
customer successes, please visit:
www.microsoft.com/casestudies                 and high-performance computing (HPC)               possible GPU configurations. However, it
                                              clusters. The largest is the TSUBAME               found that there was a lack of tools in the
market to support GPU configurations for the                            development of array-processing operations,           Deployment Services,” says Maruyama.
Linux operating system.                                                 such as those that frequently arise in image          “The graphical user interface makes it
                                                                        processing and scientific computing.                  faster and easier for me to handle tasks.
Solution                                                                                                                      Plus, I can see at a glance the status of
In September 2008, Tokyo Tech decided to                                Additionally, because of the success that the         our compute nodes with the solution’s
deploy a GPU-accelerated HPC cluster running                            research group has had with its new HPC               built-in monitoring tools, which helps me
on the Windows® HPC Server 2008 operating                               operating environment, the director of the            make proactive adjustments.”
system in collaboration with Microsoft®                                 institute’s supercomputer projects is
Research. The cluster uses a combination of                             considering the use of Windows HPC Server             Continues Maruyama, “Before this project,
hardware components that the institute’s                                2008 to run its next supercomputer.                   I had never used a Windows cluster, so I
researchers put together themselves. The                                                                                      wasn’t sure how well I’d do with it. I
institute utilized Windows Deployment                                   Benefits                                              expected to need quite a while to learn
Services to deploy 32 compute nodes, a                                  Through its use of Windows HPC Server                 how to be successful with it, but I’ve been
process that went smoothly after the GSIC                               2008, Tokyo Tech is broadening its research           surprised by how quickly I was able to use
resolved issues with network switches and                               and extending the value of HPC because the            the tools and be productive.”
verification of the proper drivers.                                     solution is so easy to learn. “Implementing
                                                                        Windows HPC Server 2008 has made it                  Solid performance. Tokyo Tech has
As of January 2009, a small research group                              easier to expand the power of HPC to the              discovered that the performance of its
within the GSIC uses the Windows HPC Server                             greater research community,” says Naoya               new cluster is comparable to that of its
2008 cluster, initially experimenting with                              Maruyama, Postdoctoral Research Fellow at             Linux HPC clusters. “We’ve seen
running a custom-developed bioinformatics                               the Global Scientific Information and                 equivalent times both for performance
application. The application supports the                               Computing Center at the Tokyo Institute of            and for the time that it takes to start jobs,”
maximum number of GPU configurations,                                   Technology.                                           notes Maruyama. “With Windows HPC
which makes it easy for researchers to truly                                                                                  Server 2008, we haven’t had to sacrifice
study the effectiveness of using GPUs versus                               Greater choice of tools. Tokyo Tech               performance to be able to derive the
standard central processing units (CPUs) with                               researchers at the GSIC now have access           benefits from such a wealth of tools.”
respect to performance and power usage.                                     to more configuration tools with Windows
                                                                            HPC Server 2008. “Only on Windows can            Increased user productivity. GSIC
Although the researchers had no prior                                       we find and change configuration settings         researchers have experienced a boost in
experience with Windows-based tools, the                                    so quickly and in so many ways,” says             productivity because of the integration
group was able to make immediate use of                                     Maruyama. “For instance, with Windows             among tools in Windows HPC Server
such features as the built-in Job Scheduler                                 HPC Server 2008, we can use a GPU                 2008. “When working on Linux, we
and graphical system-monitoring interface,                                  configuration tool from NVIDIA with which         needed to learn a variety of tools that
including the Heat Map feature for                                          we can modify GPU clock speeds to                 have been developed by a number of
visualization. On average, the researchers run                              possibly find a way to reduce GPU power           different vendors and manufacturers,”
hundreds of jobs a day.                                                     consumption without significantly affecting       says Maruyama. “But Microsoft provides
                                                                            application performance.”                         all the necessary components for running
In the future, the group expects to run                                                                                       the cluster. With Windows, there’s
additional applications for scientific                                     Intuitive deployment, use, and                    predictability in how the tools look, feel,
computation on the cluster. For instance, it                                management. The institute made a                  and work, which makes it easier to run
plans to port its applications to Microsoft                                 smooth transition to supporting a                 jobs and helps us accomplish more in the
Accelerator to study performance, ease of                                   Windows operating system in its HPC               same amount of time.”
programming, and other factors. Microsoft                                   environment. “I’ve used tools for Linux
Accelerator is a high-level data parallel library                           deployment and management in the past,
that is designed to help simplify the                                       but none that were anything like Windows




This case study is for informational purposes only. MICROSOFT MAKES NO WARRANTIES, EXPRESS OR IMPLIED, IN THIS SUMMARY.

Document published February 2009

								
To top