IT Surfs The Ocean's WavesIT Surfs The Ocean's Waves
Linux cluster gives researchers a hardy platform for wave simulation.
July 3, 2002
Thanks to some heavy-duty computing, researchers at Johns Hopkins University in Baltimore are breaking new ground in oceanography. The project, which is focused on how the Atlantic Ocean affects its surrounding climate, began in earnest 18 months ago when the university chose to run its data collection and analysis applications on a cluster of Dell PowerEdge servers using Red Hat Linux 6.2.
"Simulating the North Atlantic is a computationally complex job," says Thomas Haine, assistant professor of earth and planetary sciences, and this research couldn't be done without the right IT platform.
The application running on Dell servers helps researchers study the effect of the ocean on climate.The ocean's ability to absorb gases, carbon dioxide in particular, can have a profound effect on global warming and climate patterns across land masses nearest the ocean, Haine says. To simulate the temperature, salinity, and pressure of the North Atlantic, Haine relies on actual data collected from satellites and ships and sent to the university via the Internet or posted to Web sites. Haine's Dell cluster consists of eight Dell 2450 PowerEdge servers. Each server has two Pentium III CPUs, 18 Gbytes of hard drive, and 1 Gbyte of memory. Data is stored on each server within the cluster, and the cluster is interconnected via two networks. The cluster uses Dell Fast Ethernet switches to connect with the university's network. Over these switches, researchers move files and perform administrative functions. The cluster uses Myricom Inc.'s Myrinet packet-communication and switching interfaces to connect the servers together within the cluster. Using both connections to split up the workload brings speed and efficiency to the cluster. Johns Hopkins' staff considered systems from Silicon Graphics Inc. and Sun Microsystems but ultimately chose Dell and Linux because of price. An expensive supercomputer was out of the question. "It was pretty clear that performance for dollar would be optimized if we went with a network of clustered PC servers," Haine says. "The cluster was about $50,000. We could have done this with Unix but wouldn't have gotten as much for our money. I suspect we would have been looking at twice that from Sun and SGI." The project reached a milestone last week when the university submitted a research paper for publication in the Journal Of Physical Oceanography.
About the Author(s)
You May Also Like