High
Performance Computing Made Easy with Windows HPC Server
2008
By
Joe Lee
|
|
|
Introduction
In the old days, High Performance Computing (HPC) implies extremely
expensive hardware and software installations, intensive training
to get accustomed to unfamiliar tools and working environment,
heavy administration costs for daily operations, only relevant
to complex, large scale and long-running number crunching applications,
and so on (please refer to the Network Computing article “Supercomputing
Made Possible on Campus” for details). However, as technologies
have advanced rapidly in recent years, HPC becomes viable and
practical through the combination of affordable hardware boxes
and Microsoft software products. HPC is no longer a term for discussion
but has become a fundamental enabler of innovation in projects
ranging from scientific to business to social sciences. This article
addresses a simple HPC environment that is built on top of commonly
used servers and the Microsoft Windows HPC Server 2008 software.
Microsoft Windows HPC Server 2008
Microsoft Windows HPC Server 2008 features fast deployment, simple
administration, easy application development, highly scalable,
and cost-effective, enabling researchers to be more focus on data
analysis and hence, more productive. With this, researchers can
now have yet another choice of high performance computational
needs in addition to the traditional HPC technology. Many successful
stories have been made available for reference in Microsoft’s
HPC website. For details, please visit the following link: http://www.microsoft.com/hpc/en/us/case-studies.aspx .
Obviously, it is straightforward to build a Windows HPC Server
2008 cluster by using the Windows Deployment Services. Once the
head node (or a cluster of head nodes for high availability) has
been installed, the set up of compute nodes can be very quickly
in parallel on all nodes. Subsequent configuration and testing
are simple as integrated tools are available. Throughout the whole
installation process, administrators require no assistance or
training from the vendors due to the familiarity of the Windows
infrastructure and interface. When the whole cluster is up and
running, administrators simply use the graphical user interface
to schedule jobs and manage resources. They can perform proactive
adjustments by checking the status of the compute nodes with built-in
monitoring tools. As enhanced reporting tools have been built
into the system, there is no need to manually pull together statistics
to produce reports.
As the Windows HPC cluster uses the existing corporate infrastructure
and Microsoft Active Directory for security, account management
and operations management, the environment can be easily maintained
by administrators. The integrated environment has sufficient security
features to authorize cluster use, safeguard data, and allow remote
access, facilitating researchers to continue working on their
projects outside of regular working hours and sharing results
no matter where they are. Therefore, precious resources can be
more accessible, better collaboration and information sharing
is achievable, and shorter project timeline can be foreseen.
The Windows HPC cluster provides a more friendly and recognizable
computing environment for end users. Working with such a familiar
HPC environment has distinct advantages, including saving the
effort to master sophisticated command lines to harness the power
of HPC, predictability of how the tools look, feel and work, troubleshooting
problems through ample resources and comfortably addressing the
problems to technical support. Researchers and application developers
can use Visual Studio (which supports the Microsoft Message Passing
Interface) to easily develop, debug, and run parallel programs
on the HPC cluster. Besides, there are a wide variety of third-party
applications for selection to meet project requirements. Furthermore,
users can specify the resources required for running a job, such
as amount of memory, number of cores, sockets, nodes, and software
licenses. As a result, the HPC releases researchers from management
of the system as well as search of user-friendly tools to tackle
problems, making them possible to concentrate on data analysis,
experiment with more data sets, and produce high quality output.
CSC Potential Applications
The Computing Services Centre (CSC) is planning to start out with
a simple HPC setup of 8 to 16 nodes as a proof-of-concept for
this technology. The application in mind is a simple data warehouse
for collected system performance data. If the setup is successful,
the technology can be introduced to departments, adding one more
dimension to the choices for HPC. The potential uses will include
scientific and business applications requiring high computing
power such as simulations.
Conclusion
The Microsoft claims that the Windows HPC Server is fast to set
up out of the box, easy to use, and with high-throughput capacity.
It is equipped with a comprehensive set of tools that are easy
to deploy, manage, and integrate with existing infrastructures,
increasing productivity on the one hand while reducing complexity
on the other. While minimal training is required for experienced
administrators to build and operate the cluster, end users feel
at ease working in the familiar Windows environment. If this proved
to be true, researchers can focus on their research work, resulting
in greater productivity. This will be one of our moves to support
research requirements which are important for University advancement.