Airplanes fly over Kader Frendi's home in Madison frequently, and the roaring of the craft's approach to the airport serves as a reminder of research he is conducting at The University of Alabama in Huntsville.

Frendi is searching for ways to reduce noise associated with the takeoff and landing of aircraft. He has been investigating the science of noise generation and noise reduction during most of his academic career. He joined UAH's mechanical and aerospace engineering department five years ago. Computational fluid dynamics (CFD) uses computers to solve complex equations that model a problem of external or internal airflow over vehicle shapes.

"One of the big concerns in the aeronautical field is the noise level generated by the takeoff and landing approaches of airplanes," Frendi said. "Most people don't know it, but during takeoff and landing the noise from the airframe is almost equal to the noise made by the airplane's propulsion system." Most of the noise during the plane's approach to the runway is generated by the landing gear, the wing and the moving parts of the wing.

However, a limiting issue in CFD research is computing power. The computations get pretty complex very quickly when at each grid point you need to compute five variables (sometimes more). In addition, the computation needs to be carried out for a certain length of time to get all of the flow statistics. This complication arises because of the turbulent nature of the flow around such objects.

Such research in the past would take months or years. Today, Frendi will get the results within one week. These advances have been helped by more powerful computers, but even today's fast PCs still lack the power necessary for CFD research.

So, Frendi is using the concept of cluster computing to solve that problem. He has created the CFD Laboratory at UAH by linking 64 computers together. "This 64-node cluster will allow us to find the noise sources and analyze the data much more quickly," he said.

One of the challenges in cluster computing is dividing the computations among the various machines. Each of the 64 machines will handle anywhere from 500,000 to 1 million points apiece. "We will have to balance the load among the machines," Frendi said. "If one node doesn't have an equitable share, then it slows that one computer down. If you slow down one computer, you slow down all 64."

CFD researchers are ambitious, he said, and the science will continue to evolve rapidly. As a matter of fact, Frendi was quick to point out that even though the university just opened the 64-node lab, he is already looking at expanding UAH's CFD Lab and doubling its capacity to 128 computers. "We've designed the lab with growth in mind. CFD research has a lot of applications, and the best way to take advantage of those opportunities is to build the university's CFD capabilities."