Text Size Default Text SizeDefault Text Size Large Text SizeLarge Text Size Largest Text SizeLargest Text Size Print Print this Page

2010 News Releases

Longwood "cluster" computer expands research possibilities for students and faculty

October 21, 2010

Dr. John Graham next to the cluster computer Dr. John Graham, associate professor of computer science, stands next to the cluster computer built and maintained by computer science undergraduates.

Longwood University's computer science program built and maintains an incredibly fast computer - actually, 32 computers in one - that is used by students and faculty for research. It was built by undergraduates using leftover computer parts.

The computer, called a cluster computer or simply a cluster, can transfer one billion bits of information per second among each of its 32 processors. It has resulted in two research articles in journals and a presentation by a student at a national research conference.

"For a school of Longwood's size, and a department our size, having a computer like this is very unusual," said Dr. John Graham, associate professor of computer science. "Cluster computers are used to solve certain problems that require exceptional number-crunching power because a cluster can run multiple processes in parallel to each other, all working on parts of the same problem."

Two grants totaling $15,000 from the Cook-Cole College of Arts and Sciences and some hardware from Information and Instructional Technology Services (IITS) contributed to the project, known as the Longwood University Cluster System (LUCS). Work on the cluster began in summer 2008 after the lab for computer science majors moved to the basement of the Hardy House.

"The guts of the computer were built by undergraduates Daniel Honey and Jeffrey Ravenshorst, and another undergraduate, Kathryn Tipton, also helped," Graham said. "When they were working on it, Daniel and Jeffrey would often spend the night in the computer lab, order pizza, and sleep on the couch. It was originally built with a $5,000 grant approved by Dr. Chuck Ross (arts and sciences dean), and a recent grant of $10,000 from Dr. Ross enabled us to upgrade the processors. We just finished upgrading, which took us from 32 to 128 processors, a four-fold performance increase."

The cluster is seven feet tall and consists of 32 boxes called "pizza boxes," due to their appearance, stacked on top of each other in a tower-like structure, which Graham calls a "rack." Each box has a motherboard and four processors, which came from discarded computers, and all 128 processors are networked together.

"Each pizza box is as fast as any computer you can buy at Wal-Mart," Graham said. "Any one of these boxes can see what any other box is doing. When I flip a switch, called a KVM switch, I can look at the status of all 32 boxes. I can do 32 things at one time. We took the guts out of desktop computers that were being replaced, and we put them in the pizza boxes and networked them all together. In the upgrade, we replaced all of guts - the motherboard, disk drives, memory, everything - with new guts."

Each pizza box is completely self-contained, with its own processor, disk, network interface and human interface. Each of the current motherboards has gigabit network speed and two gigabytes of memory. The cluster has a 160 gigabyte disk drive, more than five terabytes of storage, 64 gigabytes of memory, and two servers.

"Cluster computers solve complex problems that require billions of calculations in a second and are far beyond the capability of ordinary desktop computers," he said. "The machines typically cost tens of millions of dollars, so over the past few years, universities have learned how to construct their own 'supercomputers' by linking inexpensive personal computers and writing software that allows these computers to tackle extraordinary problems. These high-performance parallel computing clusters of inexpensive PC hardware are known as 'Beowulf' clusters, derived from a computer built by NASA in 1994. Our type of cluster is the most common one, called an MPI-Cluster because it uses some type of message passing interface, which allows computers to communicate with each other."

IITS contributed a server and a 48-port Gigabit network switch. "The recent upgrade should sustain the project from a hardware perspective for several years," Graham said.

As a direct result of research done in the lab, two articles by Graham have appeared in scholarly journals. "Comparing Parallel Programming Models" appeared in the Journal of Computing Sciences in Colleges in June 2008. "Integration of Parallel Programming into Traditional Computer Science Curricula" appeared in Inroads, published by the Association for Computing Machinery, in December 2007.

Also, a presentation by Daniel Honey, "Measuring Time Complexity of PRAM Implementations," at the National Conference on Undergraduate Research in April 2010 at the University of Montana, was based on research using the cluster. Honey, a computer science major who graduated in May 2010 and is a graduate student at the College of William & Mary, conducted his research under Graham.

"The signature program of our college is undergraduate scholarly work mentored by our faculty members," said Dr. Chuck Ross, dean of the Cook-Cole College of Arts and Sciences. "Because John's computer lab fits this vision perfectly, I have been happy to support it. The original grant was from the Undergraduate Scholarship Fund, and the recent grant was from the Dean's Fund for Scholarship Excellence."

Dr. Phillip Poplin, associate professor of mathematics, used the cluster during the spring semester to run a simulation. A student with whom he had worked in fall 2009, Richard Hayden, had written a simulation of the game "SameGame" and tested different algorithms to determine a good solving algorithm for the game that could be solved by humans, not by a computer. Hayden simulated the problem for one specific board size with one color scheme.

"I was able to use the cluster to gather simulation data for many different board sizes along with many different color schemes," Poplin said. "The cluster allowed me to collect a large amount of data - hundreds of thousands of data points - in approximately one month instead of two to three years if I had only one computer working on the problem. I have a couple more projects that I hope to use the cluster for in the near future."

The cluster also has been used by Dr. Consuelo Alvarez, associate professor of biology, while directing an internship last spring for Daniel Honey. "The cluster was used for sequence analysis and alignment in his work, which was titled 'Using Bioinformatics Tools on the Longwood University Cluster,'" Alvarez said.

Graham's research specialty is integrating parallel programming into computer science education. "The days of your computer having only one processor in it are over," he said. "If your computer has multiple processors, you have to know how to write programs for multiple processors, which is different."

Graham is proud of the computer science lab. Students and faculty who use the lab, which is kept locked, are given a log-in and an access code. As soon as he arrived at Longwood in 2005, he built the former computer science lab in Barlow Hall, in a room next to the registrar's office, which had to be vacated when Barlow was renovated. He had built a similar high-performance computing lab when he taught previously at Coastal Carolina University.

"This is an all-purpose computer lab for upper-level computer science students and anyone who needs computing power above and beyond a desktop computer," Graham said. "Some classes are taught here, as is the lab for the Intro to Computer Science course. We have a robotics lab, a hardware lab, an experimental lab, and a small library here. We're given surplus computers, and we upgrade them. We scavenge pieces, take machines apart, and make them better, bigger and faster. The students stockpile parts so that when one part breaks, they can replace it. We're the ultimate recyclers.

"To use the lab or the computer, you don't have to come here. You can log in from wherever you are on campus. We also have wireless connection, so you can connect wirelessly to this network, which was built and is maintained by the computer science program. Any faculty member is free to use the cluster. This lab is a resource as a research facility to explore big problems."

Every semester Graham has one or sometimes two research assistants in the computer lab who earn three credits in an internship/directed study. This semester his assistants are Amanda Buckley and Kevin Strickland, both senior computer science majors. Past assistants include Kevin Mooney (a 2008 graduate), Jeffrey Ravenshorst ('09) and Daniel Honey.

"These assistants pursue research topics of interest to them, help me on my research projects, help tutor first- and second-year students, and help keep the lab running, which enables them to learn systems administration, in other words, how to take care of this environment," Graham said.

Graham is an editor of Incite, the research journal of the Cook-Cole College of Arts and Sciences, and coaches the men's and women's rugby teams. His daughter, Dr. Sarah Graham Porter, is assistant professor of chemistry at Longwood, and Graham Hall on campus is named for his late grandfather, Samuel L. Graham III, Longwood's business manager from 1929 to 1955.