WSU Professor Creates Super Computer
2004-04-12 10:23:08
What do you do when you need super computing power and the latest in supercomputers costs more than $24 million to build? (At least that’s what the Department of Energy is paying.) If you’re Dr. Paul Schimpf, associate professor of engineering at WSU Spokane, you simply build your own. Schimpf is doing original research in biomedical engineering and received a $400,000 grant from the National Science Foundation (NSF) to fund his research. His research project requires a lot of heavy duty computing accomplished quickly, so he needs a lot of computer power. His answer was to create a “Beowulf Cluster.” (To read more about Schimpf’s research go to: http://www.terabytetriangle.com/index.php/id=5&article_ID=7)
Thomas Sterling and Don Becker, who worked at Center of Excellence in Space Data and Information Sciences (CESDIS) under the sponsorship of the Earth and Space Sciences (ESS) project, built the first Beowulf machine in the summer of 1994. Their cluster computer consisted of 16 DX4 processors connected by channel bonded Ethernet. Their idea of providing Commodity Off The Shelf (COTS) base systems to satisfy specific computational requirements quickly spread through NASA and into the academic and research communities. Why they chose to name their creation after the hero of an Old English folk epic predating the 8th century has not been explained, but the name, Beowulf, persisted.
Schimpf’s Beowulf cluster is made up of nine separate machines -- eight nodes and a host. Each node contains a single processor, 512 megabytes of memory, a 20-gigabyte hard drive, and runs at 1.7 GHz. The host machine has all that plus a 40-gig hard disk. The whole thing uses the Linux operating system. The eight nodes can be selected and monitored using a KVM switch. The host is connected to the Internet. The whole is a parallel computer that speeds up multitudes of numerical computations by dividing the work among all of the computers, which then report the results back to the host. This means that results can be obtained in less than one eighth of the time that would be required on a single computer.
With all that computer power, Schimpf has a lot of work to do before he can crunch the first number. The over-the-counter machines arrive with no operating system and no Graphical User Interface (GUI). There is no Microsoft product in sight. Each machine must have the Linux Operating System installed. Then the network must be built with software on the host that will spawn and distribute the parallel jobs. Schimpf and his research assistants must optimize the software for the specific research being carried out. The optimization work will be the topic for a Master’s Thesis.
Simultaneously, parallel algorithms for locating the source of biomedical signals will be developed. For this, Schimpf will use simulated data. After all elements are in place, actual clinical data will be used to perfect the computational and computing tools.
The use of Beowulf Clusters and parallel computing is an affordable way for researchers to achieve the computing power they need for various applications. Because the software needed to distribute the tasks often must be specially adapted to the specific problem, parallel computing requires a lot of knowledge. Spokane has a robust parallel computing community, so in addition to his research assistants, Schimpf has a many professional colleagues.
Based at SIRTI on the Riverpoint Campus, Professor Schimpf teaches computer engineering, electrical engineering and computer science courses for WSU Spokane. He also coordinates the Engineering Consortium, a collaborative program between WSU Spokane and the University of Idaho for nontraditional students wishing to complete a bachelor of science in computer engineering.
For more information on the new supercomputer at the Pacific Northwest National Laboratory (PNNL) in the Tri-cities, see http://www.terabytetriangle.com/index.php/id=5&article_ID=43 or http://www.spokesmanreview.com/news-story.asp?date=041702&ID=s1132899 -- Billie Moreland, PhD
|
