NEWS
|
|
Contact: Bruce Conley
Associate Director of College Relations
Phone: (605) 274-5526
Fax: (605) 274-4903
www.augie.edu
May 10, 2007
Augustana Students Build, Use Computer Cluster
SIOUX FALLS - For the last two summers, Augustana students working in the chemistry department have been developing hardware and software for cluster computing.
Collaborating with Dr. Brian Moore, this effort started in 2005 with two students, David Huebner (Fargo, ND) and Collin Taphorn, (Storm Lake, IA) and continued through interim 2006 and summer 2006 with the work of Huebner, Joseph Coppock, (Sioux Falls), Jacquelyn Strey (St. Peter, MN) and Ben Bomstad (Inver Grove Heights, MN)
Huebner and Taphorn initially built from scratch a standalone, rack-mounted cluster with high-speed network switch. The system consists of a head node computer and four slave nodes; each node has two AMD Opteron processors. All the computers run the Linux operating system, with an open-source clustering software package called Clustermatic, developed at the Los Alamos Labs. The students researched different possible cluster computing approaches, chose the hardware, and assembled and tested the system.
During interim 2006, Huebner worked on software so the cluster could be used in Dr. Moore’s simulations of large atomic clusters. Existing molecular dynamics code was modified to incorporate parallel processing, using MPI (Message Passing Interface). With these modifications to the simulation program, Moore and his group have been able to simulate clusters almost ten times larger than before, in nearly the same amount of time, because the program runs on eight processors simultaneously. In testing runs, the speed advantage ratio of the MPI/parallel program as opposed to the non-parallel version was 7.7 out of a theoretically possible 8.
With the new cluster hardware and software, Moore and his students have been able to model clusters in sizes up to 15,000 atoms, confirming hypothesis that larger liquid clusters will tend to always revert to the solid state; this effect is very size-dependent and was the main driving force for assembling the cluster computing capability.
The atomic cluster simulations are very computationally intensive, but with a relatively low memory requirement, so the Clustermatic approach, where the nodes can be run completely “diskless”—the program runs all in RAM—is particularly well suited. The students have also been able to run regular MS Windows computer in the Augustana labs as temporary cluster computers. Each computer boots Linux from a CD and loads the Clustermatic software, then the processor is available for use in the cluster. When the CD is removed and the computer is rebooted, the system has not been changed at all since the programs all run in RAM and not on the hard-drive.
The temporary cluster approach seems especially promising to Moore. “The hardware in our dedicated cluster is working very well for us now, but computer hardware quickly becomes obsolete,” he said. “The work that Huebner has done in developing a way to temporarily use any available computer should serve us very well in years to come, as we can now borrow any idle system to put to use in our work.”
The cluster computing work has been derived from a variety of sources. Huebner has been supported for two summers on a combination of NASA undergraduate grants and ARAF; Taphorn and Bomstad were supported out of College funds, working half-time as teacher assistants; Coppock and Strey received their summer funding from an NPURC (Northern Plains Undergraduate Research Center) grant; Dr. Moore’s summer salary was provided by ARAF and NPURC. The computer hardware was purchased from Augustana College start-up funds for Dr. Moore. The cluster currently resides in the Madsen Center server room. Dan Drenkow of IS and the Augustana computer science department provided advice and help in assembling the cluster, especially Dr. Dan Swets.
“I’m especially proud of the fact that this project developed initially out of an NSS (New Student Seminar) conversation,” said Dr. Moore. “I first met David Huebner in his freshman year in my NSS class. As part of NSS, each student is required to meet one-on-one with the faculty member. As David and I talked during that first meeting, I described to him my research and that I wanted to eventually build a computer cluster to help with my work. He said that he had always been interested in cluster computing and that he had previous experience in programming and computer hardware. The next Spring semester, David sent an application to NASA for summer support which was funded and we were able to start the work. But it all started in a discussion in my office a few weeks after David arrived on campus at Augustana, in his NSS class.”