Red and blue swirls, representing warm and cold currents, collide on a map of the North Atlantic Ocean.
This ocean current simulation, called the Miami Ocean Model, is one of the research projects associated with the University’s Laboratory for Computational Science and Engineering.
Researchers in the basement of the Electrical Engineering and Computer Science Building use this type of graphic to visualize and explain complex simulations.
“(Ocean current simulation) is a very computation-intensive problem, so we were looking at ways to solve it using multiple computers,” said Matt O’Keefe, an assistant professor of electrical engineering and the lab’s assistant director. “That’s where it feeds into the lab’s mission.”
O’Keefe said his group ran its ocean current simulation program on a Cray-T3D, a top-of-the-line supercomputer. The computer required five weeks to create a simulation that covers a two-year period of oceanic change.
University researchers also use the lab’s machines to simulate turbulent convection in the sun and convection in the Earth’s mantle. Convection is a circular current pattern created when warmer fluid rises at the same time that cooler fluid is sinking.
The lab was created at the University last year to see if smaller, off-the-shelf computing equipment can perform functions traditionally done by more expensive, custom-made supercomputers.
Smaller computers can perform supercomputer functions by clustering, or attaching computing equipment in parallel so that more than one piece of equipment can work on a problem simultaneously.
Conventional computers have only one processor, which is responsible for computation and control. But by linking processors together, the computers can complete tasks more quickly.
The lab’s Power Challenge Array was recently expanded to link 124 processors in parallel. Previously, only 12 processors were linked.
Another example of clustering is found in the lab’s TeraByte disk file system. This project connected 37 disk arrays, in which each contained seven data disks and one parity disk, which detects errors in transmissions between computers. The entire file system is connected so that is can perform as one disk.
A sustained data transfer rate of about 500 megabytes per second was achieved about a year ago. This speed was more than 100 times faster than the transfer rate of a disk drive on a conventional home computer.
The lab has a unique relationship with several local computer companies. These companies, including Ciprico Inc., loan the lab their latest computers to test their effectiveness in the demanding applications required by University researchers.
The loan period is typically nine months to one year. Afterward, the old equipment is replaced by more up-to-date equipment.
The PowerWall is a digital display that has 3,200 by 2,400 pixels, or image elements, the largest numbers of pixels of any display in the world. The higher the density of pixels, the greater the resolution of a digital image.
The PowerWall measures 6 feet by 8 feet and is used to display complex computer output, such as fluid dynamics or astronomy simulations. The large display allows researchers to better visualize the lab computation results.
In addition to the lab, University researchers also have the option of doing their high-performance computing at the Minnesota Supercomputer Institute, which provides supercomputing power through a contract with Cray Research, Inc.
Paul Woodward, the lab director and an astronomy professor, said his lab is better equipped to deal with some computer applications, such as fluid dynamics simulations. He said other researchers may find that supercomputer labs fit their needs.
“It used to be true that supercomputers would be 50 to 100 times faster than workstations. What we are seeing now is that they are only about three or four times faster, but they cost on the order of 10 times more. So unless you really have a time constraint, it’s more cost effective to use workstations,” O’Keefe said.
The lab has a three-year, $1.2 million grant from the National Science Foundation. The grant is part of NSF’s Metacenter Regional Alliance Program, which helps fund computer labs across the country for research and community outreach. The lab is also supported by grants from Seagate Technology as well as other computer companies.