Surface temperature overlaid with surface current vectors, as forecast for 16 August 2006, 12:00Z. This data-driven forecast was issued on 15 August. Computations in the two modeling domains are two-way nested using message passing on the Crimson Grid.
Image Courtesy Pierre Lermusiaux
|
Harvard University's campus grid is only two years old, but it's already supporting the research projects of 21 different faculty members in 21 different scientific disciplines. These researchers use the Crimson Grid to investigate voting patterns and earthquakes, simulate cancerous tumors and subatomic particles, and model ocean dynamics and acoustics.
"We were running in a distributed fashion before, but with fewer, older computers," says Pierre Lermusiaux, an oceanography researcher at Harvard. "We moved some of our ocean modeling to the Crimson Grid recently. We are not using the whole grid yet but we have already increased our performance by a factor of ten."
Lermusiaux and collaborators are using the Crimson Grid to aid their participation in two experiments studying ocean dynamics and acoustic propagation in the Monterey Bay region off the coast of California. The Adaptive Sampling and Prediction project is exploring the use of autonomous vehicles to optimally and adaptively sample ocean conditions and use the collected data to more accurately predict future conditions. As part of ASAP, the Harvard group, led by Allan Robinson, models and predicts ocean conditions such as temperature, salinity and currents. These predictions help direct where the autonomous vehicles should sample ocean conditions. Similar predictions are also carried out for the second project, the Persistent Littoral Undersea Surveillance Network, or PLUSNet, which is studying underwater acoustic surveillance.
"We run ocean models for a large-scale region of 150 by 230 kilometers and a nested smaller region of 50 by 70 kilometers," explains Lermusiaux. "The models for the two regions run in parallel on the grid and communicate via message-passing."
The ocean modeling jobs are submitted to the Crimson Grid using Condor, where they may be run on one of a number of campus resources connected to the grid.
"Crimson Grid resources could be single desktop computers or any size cluster," says Joy Sircar, Chief Information and Technology Officer at Harvard University's Division of Engineering and Applied Sciences and Crimson Grid project leader. "The resource owner could be an individual student, faculty, graduate student, faculty lab, department or a whole school."
The Crimson Grid got off the ground in 2004 with funding from Harvard and IBM. For the last two years, Crimson Grid developers have been focused on getting scientists involved and federating clusters from laboratories and departments. Grid usage has grown from 100 jobs per month to 10,000 jobs per month.
"We have seen a remarkable degree of satisfaction in faculty and researchers who have gotten their work done a lot more easily than they may have within the confines of their own small labs," adds Sircar. "People seem to be finding it a lot easier to collaborate between different disciplines."
The Harvard ocean modeling group, which includes Patrick Haley, Pierre Lermusiaux, Wayne Leslie, Oleg Logutov and Allan Robinson, thanks all ASAP and PLUSNet collaborators as well as the Crimson Grid IT team and acknowledges support from the Office of Naval Research.
Learn more at Harvard's ASAP and PLUSnet Web sites.
For more information on the Crimson Grid, contact jsircar@harvard.edu.
—Katie Yurkewicz
e-mail this article
|
