Photo by Stephanie Cartier
Researchers casually refer to them as a bunch of "screaming boxes," but the potential within these boxes is generating a buzz of excitement in the scientific community. The "boxes" are actually 58 high-performance computers connected in a cluster that is being used to build powerful models of disease and infection.
At a recent introductory presentation and reception, Dr. Betz Halloran and system analyst programmer Shufu Xu unveiled some of the capabilities of the Linux computer cluster they have dubbed Psyche. The Public Health Sciences Division team, which includes Halloran's research partner, Dr. Ira Longini, uses the Psyche cluster to run thousands of simulations of flu pandemics in the United States, evaluating various intervention strategies.
"One action on one computer may take you a thousand hours, but if you have a thousand computer servers and let them do the same thing at the same time, it will take only about an hour," Xu said. "A cluster gives you faster turnaround in scientific research. The simulation of an influenza pandemic in the United States took 120 hours to finish using the cluster, but on one personal computer it would be virtually impossible."
After running for five days straight, the cluster churned out a model of a pandemic flu, using statistics to determine how the disease would work its way across the nation. It's a powerful model that enables the researchers to assist government leaders in preparations for such an event.
"One of the things we work on is simulation of large populations for pandemic-flu planning — so right now we're working closely with the White House and with Department of Health and Human Services to help with pandemic planning. We are able to run the entire U.S. simulator on this — 280 million people — with each person represented in the computer, and go through the pandemic, which we weren't able to do before."
The cluster also has the potential to show how infections occur between people. "Besides the national-flu simulator, we're building a social-network model to simulate HIV transmission," Xu said. The simulator can currently model how about 3,000 people interact, but collaborators expect to soon be able to increase the population to represent a real geographic area in Africa. Understanding how a disease is spread in various locations could be key in prevention and treatment.
The team, which hopes to increase funds for the cluster, is planning to work on the design and analysis of vaccine studies, which is one of Halloran's areas of interest. "We're able to do not only large-scale statistical analyses, but also simulations of vaccine studies where we're trying to generate data and develop new methods for analysis," Halloran said.
Halloran, Longini and their computers are currently supported through the Center's general fund, which is sustained in part by private donations.