Page 2 of 2   « previous page

Resolving turbulence and the internal structure of flames in real-world devices is extremely difficult. Turbulence is intrinsically three-dimensional and spans a broad spectrum of length and time scales. Against this fluid mechanical backdrop, finite-rate chemical processes are modulated by turbulence. Hence, developing the stationary statistics required to validate accurate models of real-world devices requires running extremely large grid point problems - approaching a billion grid points and beyond-for very long durations. And the limitations of most previous HPC systems restricted detailed chemistry simulations to two dimensions and a scale of millions of grid points. Now, leadership-class systems like the Cray XT3 at ORNL can take full advantage of new modeling techniques and allow researchers to extend simulations to three dimensions and approach the parameter space of practical devices. (For example, see Figure 1.)

With these new investigations, the Sandia researchers are studying a variety of phenomena that control the performance of combustion technologies, including flame extinction and re-ignition, flame stabilization, auto-ignition, and soot formation. The knowledge gained from these studies can be used to create new models of combustion phenomena, validate and refine existing combustion and mixing models, and advance the predictive capability of engineering computational fluid dynamics (CFD) approaches to design tomorrow's more fuel-efficient, cleaner-burning combustion devices. Ultimately, the work may also have applications in astrophysics, fluid mixing, and other fields.

"Even highly complex multi-physics interactions, such as 'turbulence-chemistry' interactions in combustion flows, are now accessible through the growth of processor speed, computer memory, storage, and advances in computational algorithms and chemical models," says Dr. Jacqueline Chen, distinguished member of technical staff and head of the Sandia team. "These simulations are enabling scientists to 'see' with unprecedented clarity, and therefore understand the physical phenomena that govern the fuel efficiency of the automobiles we drive, control the stabilization of combustion in aircraft engines, and determine the emission levels of small particulates in exhaust streams that adversely affect human health and our environment."

Designing Better Devices to Help Ailing Hearts at Rice University
The United States sees 800,000 new cases of heart disease every year. At least 50,000 of these cases would require a heart transplant, but only 2,500 donor hearts are available annually. For patients waiting for a transplant, ventricular assist devices (VADs) are currently the only hope of prolonging life. These devices are a few centimeters in diameter, small enough to be implanted, and can offset the insufficient flow rate and pressure head of an ailing heart while the patient awaits a transplant. In some cases, these devices can even lead to a complete recovery. By increasing the heart's oxygenation and "reducing its workload," the heart can actually regenerate.

The development of VADs is a long process requiring understanding of disparate fields of science and technology from fluid dynamics to physiology and cardiac surgery. Historically, the design of VADs grew out of high-tech miniaturized pumps for simple fluids, and was adapted to blood in a long and costly process of iterative design refinement including bench tests on flow patterns and blood damage, animal tests, and finally clinical trials. Until recently, this process was chiefly guided by the intuition and experience of the device engineers and surgeons. However, in the last few years, CFD has emerged as a powerful tool to speed up and rationalize the design and validation process.

A research team funded by the National Science Foundation and led by Dr. Matteo Pasquali, professor of chemical and biomolecular engineering at Rice University, and Dr. Marek Behr, professor of computational analysis of technical systems at RWTH, Aachen University, has been applying new computational techniques and models to study flow patterns and blood damage in VADs. (One of the chief problems in the design of VADs is the minimization of blood damage, or hemolysis that potentially results in multiple organ failure.) The new models combine advanced CFD tools based on finite element methods, mesh partitioning, and iterative linear solvers, with computing advances - such as massively parallel Linux clusters with fast interconnects - and novel models grounded in the biophysics of cells. New multi-scale models developed by Dr. Dhruv Arora, a postdoctoral researcher in chemical and biomolecular engineering, and Oscar Coronado, PhD candidate in chemical and biomolecular engineering at Rice University, represent blood at two scales. At the cell scale, red blood cells are modeled as microscopic droplets that deform and interact with the flow, releasing hemoglobin. At the device level, blood is modeled as a complex fluid, whose behavior depends on the local flow conditions.

The Rice-RWTH team has been partnering with medical researchers to study VADs at various stages of development. Recent simulations of the GYRO VAD at Baylor College of Medicine have shown that the new models predict accurately both the hydrodynamic performance of the pump (pressure head versus flow rate and impeller rotation speed) and the rate at which hemolysis leaks plasma from the blood cells. (See Figure 2.)

New work in collaboration with Houston-based MicroMed Technology and the Texas Heart Institute is focusing on the DeBakey VAD. This VAD is about the size of a human thumb and has already been implanted in more than 300 patients worldwide. The new work focuses on optimizing the pump shape and performance for patients with different needs, for children, and on a new system consisting of two redesigned pumps working in tandem, which could potentially serve as a Total Artificial Heart.

Pump shapes and operating conditions will be tested on the Rice-RWTH virtual models, run on a Linux-based Cray XD1 system. These models require solving the flow equations, which involve the simultaneous solution of several million coupled time-dependent nonlinear equations, and the concurrent tracking of hundreds of "representative" blood cells swimming and deforming in the intense turbulent flow in the small device. (The transit time of a blood cell in the DeBakey VAD is between 100 and 500 milliseconds.) In the not-too-distant future, the team believes researchers may be able to design VADs and even a Total Artificial Heart by combining high-performance computing with models grounded in the physiology and biophysics of blood.

Performing Cutting-Edge Research on Linux-based Systems
As these examples demonstrate, academic researchers can now apply Linux-based HPC systems to even the most demanding cutting-edge problems. As more research programs like these adopt Linux-based clusters, the ability to port Linux to even the most advanced HPC systems will become even more valuable. Fortunately, with advances in both HPC system design and Linux itself, using the academic operating system of choice never has to be a barrier to conducting breakthrough research.

Leading-Edge Research Centers

Pittsburgh Supercomputing Center
The Pittsburgh Supercomputing Center (PSC), located in Pittsburgh, Pennsylvania, is a joint effort of Carnegie Mellon University and the University of Pittsburgh, together with the Westinghouse Electric Company. It was established in 1986 and is supported by several federal agencies, the Commonwealth of Pennsylvania, and private industry.

PSC's mission is fourfold: to enable solutions to important scientific and engineering problems by providing leading-edge computation resources to the national community; to advance computational science, computational techniques, and the National Information Infrastructure; to educate researchers in high-performance techniques and their utility; and to assist the private sector in exploiting high-performance computing for its competitive advantage. PSC is also a leading site in the National Science Foundation's (NSF) Shared Cyberinfrastructure program, providing U.S. academic researchers with support for and access to leadership-class computing infrastructure and research.

PSC recently unveiled its newest and most powerful HPC system, a Cray XT3 system named "Big Ben" after the Pittsburgh Steelers quarterback. Acquired through a $9.7 million grant from the NSF in September 2004, Big Ben - the first Cray XT3 system to ship from Cray - provides 2,090 processors with an overall peak performance of 10 teraflops, or 10 trillion calculations per second. If all 6.5 billion people on earth held a calculator and did one calculation a second, altogether they would still be 1,500 times slower than Big Ben. Big Ben will serve as an integral component of the NSF-supported TeraGrid, the world's largest, most comprehensive cyberinfrastructure for open scientific research. www.psc.edu

National Center for Computational Sciences
The National Center for Computational Sciences (NCCS) at the Oak Ridge National Laboratory was established in 1992 and in 2004 was designated by the Secretary of Energy as the Leadership Computing Facility for the nation, providing a resource 100 times more powerful than current capabilities. The Leadership Computing Facility is building the world's most powerful supercomputer for unclassified scientific research and provides researchers worldwide an unparalleled environment for new discoveries.

Accessible to scientists on a peer-reviewed proposal basis, topics of recent projects using the NCCS's resources range from astrophysics and combustion simulations to fusion, accelerator, and chemistry research. As a designated User Facility, the NCCS delivers leadership-class computing for science and engineering, focusing on a select number of very large grand-challenge scale problems.

The NCCS is a resource provider for the NSF Teragrid, a high-speed backbone for scientific data transmission that facilitates NCCS's state-of-the-art computing and storage capabilities to be used by researchers across the nation.

In 2005 the NCCS National Leadership Class Facility deployed a 25-teraflop Cray XT3 supercomputer, along with an 18-teraflop Cray X1E vector MPP supercomputer. The NCCS plans to expand to a 100-teraflop Cray system at Oak Ridge in 2006, and in 2007 to move to a system with over 250 peak teraflops and up to 100 sustained teraflops on real-world problems. www.nccs.ornl.gov

Rice University Computer and Information Technology Institute
Rice University in Houston ,Texas is one of the leading research universities in the United States. The university is currently deploying a 28-chassis, 672-core Cray XD1 system - the largest Cray XD1 system deployed to date - to power its new research computing system. The acquisition was funded by a $2 million federal grant, one of the largest awarded under the National Science Foundation's Major Research Infrastructure program.

The Rice research system will support an expanding community of researchers in fields as diverse as biotechnology, nanotechnology, psychology, earth sciences, fluid dynamics, and computer science. Designed to support hundreds of users in the future, the system will initially be used for a series of memory-intensive applications sponsored by the National Science Foundation. For example, bioinformatics researchers will use the system to leverage robotic motion-planning algorithms for computer-aided drug design. Earth scientists will model and simulate in great detail the deformation of sediments, soils, and other materials near the earth's surface. Psychologists will use the system to understand the structure and functions of the human brain better and how they impact the development of speech, language, memory, perception, and motor skills. Computer scientists will test new programming tools, compilers, and system software for high-performance computers. www.rice.edu

Page 2 of 2   « previous page