King Fahd University of Petroleum and Minerals

University Speeds Research with HPC Cluster, Sees Faster Performance, Lower Cost

King Fahd University of Petroleum & Minerals commissioned a center for high-performance computing (HPC) with a 1,024-core cluster running the Windows® HPC Server 2008 operating system and Linux in a dual-boot environment. Researchers use the system to solve complex problems more quickly. The university has found that the Windows HPC Server 2008–based HPC environment is easier to use and manage and is cost-effective.

 

Business Needs

King Fahd University of Petroleum & Minerals (KFUPM) was officially established by royal decree in 1963, with the mission of preparing professionals for scientific leadership in Saudi Arabia and producing research that contributes to the knowledge and sustainable development of the region. The university’s research ranges from nanosciences, petroleum, and seismic studies to advanced fluid dynamics and parallel processing. 

The university has always sought opportunities to further promote scientific research. “KFUPM has always been at the forefront of the adoption and creation of the latest tools and techniques for research,” says Dr. Khalid S.Al-Sultan, Rector of KFUPM. “In keeping with the advances in technology, we decided to provide a scalable, high-performance computing cluster to our faculty and students to perform advanced research.”

The majority of this research requires computers for help with complex computations and analysis. Until recently, KFUPM researchers used desktop computers, workstations, and small clusters to support their work. “Relying on workstations limited our researchers’ ability to use realistic test cases in their research,” says Syed Tariq Magrabi, HPC Project Lead and Technical Manager for KFUPM.

In 2007, KFUPM set out to establish a high-performance computing (HPC) center to push the limits of scientific research. The university wanted to ensure that its HPC center would be useful to as many researchers as possible. Therefore, it needed multiple operating systems to support diverse applications.

Solution

After a careful evaluation, KFUPM decided to roll out a 1,024-core cluster running in a dual-boot environment with Red Hat Linux 5.2 and the Windows® HPC Server 2008 operating system. “Some applications run only on Linux while others run only on Windows,” says Dr. Sadiq M. Sait, Chief Information Officer for KFUPM. “By choosing to run in a dual-boot environment, we’re able to satisfy our researchers’ requirements and give them the full power of 1,024 cores.”

I

We’re giving our researchers a premium high-end computing environment where they now can run larger and more realistic test cases and importantly, we are also focusing on getting an enhanced ROI through increased usage. Dr. Sadiq M. Sait Chief Information Officer, King Fahd University of Petroleum & Minerals

t took the university approximately one month to install, test, and roll out Windows HPC Server 2008, as compared to two months for other HPC deployments. “We could really minimize the time required for installation and testing, thanks to our choice of Windows HPC Server 2008, with its simplified setup and deployment tools, built-in diagnostics, integrated job scheduler, the Microsoft® Message Passing Interface (MPI), and the other tools available for the Windows environment,” says Farhan Khan, HPC Technical Team Member for KFUPM. “We’re already familiar with Windows, so it was quite easy to set up, with no formal training necessary. Also, we had access to support from Microsoft if we had questions.”

KFUPM researchers enjoy the Windows HPC Server 2008 Job Scheduler, a feature that they use to submit jobs right from their desktop computers. The IT team finds that integration with KFUPM’s existing Active Directory® service structure makes it easy to manage security and user authentication for the cluster. The IT team also takes advantage of native monitoring and administration tools, such as the Heat Map feature, which provides a real-time, interactive view of the overall health and status of all 128 nodes.

As of April 2009, KFUPM researchers are using Windows HPC Server 2008 to run several applications, such as fluid dynamics simulation software from Fluent. The university also uses the Microsoft MPI Software Development Kit (SDK) to develop parallel programs. KFUPM plans to run many other applications on Windows HPC Server 2008 in the near future, including MathWorks MATLAB, Mathematica from Wolfram Research, and ANSYS products. “We see a lot of opportunity for applications to run on Windows HPC Server 2008 to support our researchers,” says Magrabi.

Benefits

Using Windows HPC Server 2008, KFUPM benefits from shorter installation and testing times, integration with its existing environment, straightforward management, and numerous software application options. Says Dr.Sadiq, “We’re giving our researchers the intuitive user experience they need to run larger, more realistic test cases and accelerate the speed with which they accomplish great things in their fields.”

Optimal performance

KFUPM has tested both sides of its cluster and discovered that Windows HPC Server 2008 performs well. “Initial trends indicate that, in our environment, Windows HPC Server 2008 performs to satisfaction,” comments Magrabi. “That means our researchers get results more quickly.”

Intuitive use

KFUPM researchers have been able to easily make use of their Windows HPC Server 2008-based environment. “We’re supporting our large base of Windows users without requiring a major shift in learning for them,” explains Dr. Sadiq. “We’ve found that Windows HPC Server 2008 is straightforward, and therefore we’re encouraging our graduate students to use it as a development environment to create custom applications to solve problems, especially in nanosciences and medical physics. We didn’t expect to see such effortless adoption.”

Streamlined management

Because a small team handles both sides of the cluster, KFUPM particularly appreciates the minimal learning curve associated with Windows HPC Server 2008. “Most of the management tools, including diagnostics, are integrated into the cluster manager,” says Magrabi. “It’s easier to manage and monitor our Windows HPC Server 2008 cluster because there’s little need for scripting to troubleshoot jobs—all the information we need is right there to see.”