University of Arizona - Download Now DOC

Document Sample
University of Arizona - Download Now DOC Powered By Docstoc
					                                             Microsoft Server Product Portfolio
                                             Customer Solution Case Study

                                             University Finds Windows HPC Server 26%
                                             Faster than Linux, More Cost-Effective

                                             “Showing that Windows is faster than Linux for [a]
Customer: University of Arizona
Web Site:
                                             widely used application is very important. It shows
Customer Size: 14,500 employees              that researchers can be more productive on a system
Country or Region: United States
Industry: Education
                                             that is also more cost-effective to run.”
                                             Dr. Salim Hariri, Director, NSF Center for Autonomic Computing, University of Arizona
Customer Profile
University of Arizona, based in Tucson, is
the leading public research university in
                                             When researchers at the NSF Center for Autonomic Computing at
the American Southwest, producing more       the University of Arizona investigated the technology to set up a
than U.S.$530 million in annual research.
                                             cluster for autonomic computing research that could largely
Software and Services                        monitor, manage, and repair itself, they faced a choice between
 Microsoft Server Product Portfolio
  − Windows HPC Server 2008
                                             Linux and Windows® HPC Server 2008. They chose Windows.
 Technologies                               Benchmark testing has confirmed that choice, showing Windows
  − Active Directory
                                             HPC Server to be 26 percent faster than Linux.

                                             Business Needs                                        structures and applications that largely
                                             According to Dr. Salim Hariri, Director of the        monitor, manage, and repair themselves.
                                             National Science Foundation (NSF) Center
                                             for Autonomic Computing (CAC) at the                  One of the first questions that Hariri and his
                                             University of Arizona, the increasing size,           colleagues faced at the center was choosing
                                             diversity, and complexity of computer                 the platform on which to conduct their
                                             systems, combined with the relative                   research. Linux, the traditional answer for
                                             shortage of skilled IT workers, has resulted          large, distributed parallel computing sys-
                                             in systems whose control and timely                   tems, seemed the obvious choice—but
                                             management exceeds human capacity.                    perhaps not the right one.

                                             Hariri and his colleagues want to do                  “Autonomic computing requires significant
For more information about other Microsoft   something about that. Their NSF CAC is                amounts of control and management
customer successes, please visit:            dedicated to advancing the science of                 functionality to be added to a typical                autonomic computing; that is, IT infra-               computing platform,” says Hariri. “Linux
may still be thought of as the faster system,                           simulation software system and a Kepler           Computing Manager, NSF CAC, “and we
but it doesn‟t offer the integrated services                            workflow application. The EM application is       couldn‟t have set up a similarly sized Linux
that would give us a leg up on building and                             designed for distributed parallel computa-        cluster in a day. Windows HPC Server gives
testing autonomic systems. Moreover, if we                              tions of large EM problems such as those          us deployment and administration tools not
want to have the greatest impact on Industry                            involved in aircraft design. The Kepler work-     available in Linux, and they‟re the familiar
and government, we need to base our                                     flow application is popular with scientists       tools we already know and use in Windows
research on systems popular in those                                    and engineers, who can create large-scale         Server® environments. Basically, we did
sectors—platforms that can be adopted                                   applications by using block diagrams instead      everything we needed to from the Windows
without a major need to „rip and replace‟                               of writing line-by-line code.                     HPC Server console, instead of having to go
what those users already know and use.”                                                                                   out to the individual computers. The console
                                                                        To confirm its operating system choice, the       really helped—and we wouldn‟t have had it
Solution                                                                NSF CAC benchmark tested the Kepler               with any other technology.
To meet these requirements, the NSF CAC                                 workflow application on Windows HPC Server
researchers at the University of Arizona                                against a similarly configured Linux system.      “We also benefit greatly by being able to use
built their autonomic cluster based on the                              The Kepler workflow application ran an            the same Active Directory services that we
Windows® HPC Server 2008 operating sys-                                 average of 26 percent faster on Windows           already have in place, and our researchers
tem, Microsoft‟s high-performance computing                             HPC Server than on Linux. “There‟s a wide-        benefit because they don‟t have to sign on
platform. “In Windows HPC Server 2008, we                               spread perception—or misperception—that           specially for the compute cluster. If they‟ve
have a computing environment that provides                              Linux is faster than Windows,” says Hariri.       signed on to the network, the compute
significant elements of functionality needed                            “Showing that Windows is faster than Linux        cluster acts as just another resource.”
for autonomic computing, so the effort to                               for this widely used application is very impor-
build out the computing environment is                                  tant. It shows that researchers can be more       Windows HPC Sever also beat Linux when it
significantly less,” says Hariri.                                       productive on a system that is also more          came to the availability of documentation
                                                                        cost-effective to run.”                           and resources, making it easier to address
The center is testing its theories about                                                                                  problems that arose in the system. Enfield
autonomic computing on a compute cluster                                Similarly, the power of a Windows HPC             estimates that a similarly configured Linux
system with 48 cores, including an 8-core HP                            Server cluster was demonstrated by testing        cluster would require at least 20 percent
380 head node and five additional HP 360                                the EM solution against a single-node sys-        more administrative time than he spends on
compute-node computers with 8 cores each.                               tem. The benchmark application ran in 18          the Windows HPC Server cluster.
The compute nodes were configured from                                  hours, versus a much smaller application
Windows HPC Server templates created on                                 (just one-sixteenth the size of the larger        The faster and easier administration of the
the head node and pushed out to the                                     application) that took 30 hours to run on a       cluster also changes the nature of the
compute nodes using the highly visual                                   single-core system.                               researchers‟ work, according to Enfield. “You
Windows HPC Server console.                                                                                               can experiment, decide something‟s not
                                                                        Benefits                                          right, and go back and change it without
The Windows HPC Server 2008 compute                                     “Researchers want their applications to run       worrying how it will change the base image.
cluster was deployed in a day, with the univer-                         as quickly as possible, and our benchmark         That means researchers can work more
sity‟s existing Active Directory® services                              testing showed that for high-performance,         freely and try more iterations of their work.”
structure extended to give researchers                                  distributed parallel computing, Windows HPC
appropriate single sign-on rights to the new                            Server outperformed Linux,” says Hariri.          Some benefits are less staid. “I like seeing
computing resource.                                                                                                       Microsoft getting into the high-performance
                                                                        The advantages of Windows HPC Server              space, handling the same challenges we
Two notable applications running on the                                 extend to the setup and administration of         typically handle in Linux” says Enfield. “This
Windows HPC Server cluster are a 3-D high-                              the cluster, as well. “I have extensive           is one of the more fun projects I‟ve done.”
performance parallel electromagnetic (EM)                               experience with Linux,” says Leo Enfield,

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

Document published June 2009

Description: When researchers at the NSF Center for Autonomic Computing at the University of Arizona investigated the technology to set up a cluster for autonomic computing research that could largely monitor, manage, and repair itself, they faced a choice between Linux and Windows® HPC Server 2008. They chose Windows. Benchmark testing has confirmed that choice, showing Windows HPC Server to be 26 percent faster than Linux.