The Rocket Propulsion Lab (RPL) at Pusan National University specializes in aerospace engineering research. To meet the need for increased computing power, RPL added to its environment a 260-node cluster running the Windows HPC Server 2008 operating system. Having the easy-to-use cluster is helping RPL make significant gains in researcher productivity and job-scheduling efficiency, thus supporting advances in South Korea’s aerospace efforts.Business Needs
Pusan National University is renowned for its aerospace research. Three of the university’s engineering labs are part of the South Korean National Space Lab. In 2007, the university’s Rocket Propulsion Lab (RPL) began to conduct advanced research on next-generation rocket-propulsion systems that are environmentally friendly. This effort toward advancing the South Korean aerospace industry through new research strategies has garnered significant attention from the world of academia.
But making strides in the aerospace field requires advanced computing capabilities. Prior to 2007, RPL used parallel computing in the form of clusters comprising traditional PCs, first running Red Hat Linux with a GNU Fortran compiler, and then migrating to the
||Professor Jeong-Yeol Choi
BK21 Aerospace Program Director, Pusan National University
Windows XP operating system and Compaq Visual Fortran 6.0 compiler. “But we needed higher resolution to study the internal combustion stages of jet engine simulations and to get accurate research results. We needed to move from studying two-dimensional perspectives to three-dimensional perspectives,” says Professor Jeong-Yeol Choi, BK21 Aerospace Program Director at the Pusan National University Rocket Propulsion Lab. “The problem is that it takes enormous amounts of computing power to get detailed research results. In aerospace research, as with computational fluid dynamics research, the need for increased resolution—and therefore additional computing power—is always growing.”Solution
In 2007, Pusan National University’s RPL made its first foray into the world of server-based, high-performance computing (HPC) when it deployed a 192-core cluster running Debian GNU/Linux, with an Intel Fortran 10.0 compiler. “Our lab started with Linux because that’s what was generally available at the time, but there were issues in using the system,” explains Professor Choi. “Students who were not familiar with UNIX took a long time to understand Linux, which slowed down their research.”
In 2009, having determined that it needed even more HPC power for its aerospace research, RPL took steps in a different direction. Rather than adding to its Linux cluster, the lab adopted a 32-node, 288-core cluster running the Windows HPC Server 2008 operating system. The lab also upgraded to the Intel Fortran 11.0 compiler and anticipates adding 32 nodes and reaching a total of 1024 cores in 2011.Benefits
Ease of Use
RPL has noted the ease with which anyone in the lab can use Windows HPC Server 2008. “Researchers who are new to the lab are already familiar with the Windows environment, which is a major advantage because they can immediately be productive when they make the jump to parallel computing,” says Professor Choi.
Researchers have responded positively to the Windows HPC Server 2008 environment. “Linux requires installation of an individualized scheduler and job queue, which means that users have to understand all sections of the system to operate it,” says Jae-Ryul Shin, Ph.D., Post-Doctoral Research Associate at the Pusan National University Rocket Propulsion Lab. “But with Windows HPC Server 2008, we have a one-click solution that is very convenient, so users can concentrate on engineering issues.” High Efficiency Through Better Control Methods
RPL experiences higher levels of efficiency with its Windows Server–based cluster than with the Linux cluster because it has greater control over, and can better interpret, its computation work. “The Linux system’s job queue does not allow for the customized selection of computation nodes,” explains Soo-Hee Won, Research Associate at Pusan National University Rocket Propulsion Lab. Specifically, conflicts with computations using Open Multi-Processing (OpenMP) and the Message Passing Interface (MPI) on the Linux cluster have caused increased time and costs for researchers. “We’ve found that, running Windows HPC Server 2008, a specified node can handle a specified program, thereby avoiding conflicts and more effectively controlling computational resources.”
Additionally, RPL believes that the Windows HPC Server 2008 control panel and HPC Cluster Manager offer advantages with regard to corrections and supplementations. “We’ve also found that the Windows HPC Job Scheduler is more detailed than the Ganglia monitoring system for Linux. The information provided by the Job Control Tool helps us better prepare directions for computational analysis program corrections and supplementation,” says Won.Unified Solution
RPL now has an operating system that meets its needs in terms of the changing requirements for parallel computing operation and control methods. “With Linux, the job scheduler and disk control tools are individually selected, packaged, and delivered to us, even when expert suppliers assist us with the process,” says Professor Choi. “But with Windows HPC Server 2008, we get a unified control environment for computations, which makes the system better organized and more trustworthy.”
For more information about other Microsoft customer successes, please visit: