SCIENCE & ENGINEERING NEWS
Ithaca, N.Y. — Computer programs that can adapt to changing conditions – both in the virtual worlds they are creating and the hardware on which they are running – will be developed under a $5 million project funded as part of the $90 million Information Technology Research initiative of the National Science Foundation (NSF), announced by the White House Wednesday (September 13).
The program, called the Adaptive Software Project, will be conducted by researchers at Cornell University, the Mississippi State University, the College of William and Mary, Ohio State University and Clark-Atlanta University.
Cornell will be the lead institution for the project and Keshav Pingali, Cornell professor of computer science, will be principal investigator, leading an interdisciplinary team of computer scientists, engineers and physicists.
“In building a computer simulation you need to make decisions as you detect new conditions,” Pingali says. “We are going to study the mechanisms of evolution and steal ideas from nature. The biological model is a useful one: When conditions change it acts as a stimulus, and the response is to evolve.”
Adaptive programs, he says, will be able to adjust to changing conditions in the simulation, and also to changes in the hardware on which they are running, accessing more or fewer processors as they are needed and available, and correcting for problems such as the failure of one or more processors in a parallel processing computer. In other words, Pingali says, the software will behave like a living organism, adapting “either to improve results or to improve its own survival.”
The project will develop two simulations as test beds:
Cornell Theory Center’s Computational Materials Institute (CMI) will develop a simulation of the propagation of cracks in structures, such as those that form in aging aircraft, under the leadership of Tony Ingraffea, the Dwight C. Baum Professor of Engineering at Cornell. The CMI has a world reputation for its work on fracture simulation, and in the past three years alone, has won a Grand Challenge award ($1.8 million), a Research Infrastructure award ($1.5 million), and a Knowledge and Distributed Intelligence award ($1.8 million) from the NSF.
The Engineering Research Center at Mississippi State will simulate the behavior of “multi-phase” fluids in which a chemical reaction is taking place, such as the mixtures of liquids and gases that are found in combustion chambers, under the leadership of Professor Bharat Soni.
William and Mary scientists will contribute software for “mesh generation,” a step in the solution of complex equations of motion. Ohio State will develop software to create graphic displays as the output of simulations.
These simulations will run on The Cornell Theory Center’s (CTC) Velocity and Velocity + clusters, supercomputers made up of large numbers of Intel Pentium III processors running in parallel.
Traditionally in such simulations, programmers must set up the conditions of the simulated world and the methods the computer will use in advance. They must also specify the hardware on which the programs will run, setting aside, for example, a certain number of parallel processors and a certain amount of memory.
Adaptive software will be able to choose different mathematical models depending on how the simulation develops, based both on the conditions of the simulation and the availability of processors and memory.
In crack propagation, for example, the Cornell researchers are able to do very sophisticated simulations only at the scale of the object itself, “but all the physics is occurring in a very small volume around the crack tip,” Ingraffea explains. The new software will be able to shift down through intermediate scales to the atomic scale, choosing the best method to use at each scale, based on the conditions it senses at the surrounding levels. “This ITR award demonstrates once again the innovative research being done in CTC’s Computational Materials Institute,” says CTC director Thomas F. Coleman.
Cornell team members, in addtion to Pingali and Ingraffea include Paul Chew and Paul Stodghill, research associates in the Department of Computer Science; Steve Vavasis, associate professor of computer science; physicists Thomas Arias, associate professor, and James Sethna, professor, and Gerd Heber and Chris Myers, CTC research associates.