T his issue of the ORNL Review highlights many important computational science activities at Oak Ridge National Laboratory. The articles describe both the computational infrastructure that is in place to support breakthrough computational science, as well as significant accomplishments in science made through computation.
Some articles illustrate that computation is becoming an equal partner with theory and experimentation in the advancement of science.
Rich Sincovec, former director of ORNL’s Computer Science and Mathematics Division, starts off the issue with a discussion of the future of high-performance computing.
Ken Kliewer describes the Center for Computational Sciences, which is the focal point for computational science at ORNL, as well as one of the leading computational centers in the world. The future direction of computing is toward widely distributed networked computing capabilities. The connection between Oak Ridge and Sandia national laboratories to link two powerful computers allows scientists to address major problems such as predicting global climage change, as described by Kliewer. Bill Shelton and Malcolm Stocks show how this connection will help scientists better understand the magnetic structure of magnetic alloys.
Algorithms, tools, and software to facilitate the use of benchmarking and performance evaluation of high-performance computers are described by Ed D’Azevedo et al. This article, as well as a number of other articles in this issue, stresses the importance of interdisciplinary teams of mathematicians, computer scientists, and computational scientists working together to advance science through computation. The successful scenario has the application needs driving the algorithmic, tool, and software activities. As a consequence of such collaborations, numerous examples exist that support the thesis that each order of magnitude increase in computing power is accompanied by an order of magnitude increase in efficiency as a result of improved algorithms. This synergism between increased computational power and improved algorithms is fundamentally important because extra computing power is quickly consumed in an attack on problems previously considered intractable.
Tools and software components that enable a broader spectrum of users to more easily use advanced computing resources are especially important. They enable scientists to spend their time gaining scientific insight from the computations rather than bogged down in the details of using the computing infrastructure. Parallel virtual machine (PVM) software originally developed at ORNL has had a significant impact on heterogeneous network computing, and its broad availability has virtually guaranteed portability of applications based on PVM. Tools that emulate shared memory, such as the Distributed Object Network Input/Output (DONIO) Library, not only make distributed memory computers easier to use but often lead to better performance. Software components such as ScaLAPACK, QMRPACK, and sparse linear algebra facilitate the development of applications for high-performance computers, as described in this issue.
Large computations typically produce large output data sets that require visualization to understand the results. As the volume of data generated by computations, sensors, or other means continues to grow, new approaches to understanding the data have become mandatory. Ray Flanery et al. discuss the role of visualization along with future directions that promise to enhance interactions between humans and computers. Ross Toedte and Dianne Wooten write about ORNL’s scientific visualization services and successful use of visualizations to glean scientific insights.
James Arthur Kohl discusses the use of ORNL-developed CUMULVS to simulate and influence the outcome of scientific experiments. Computer simulations of materials are discussed in Shelton and Stocks’ article. Osman Yasar explores the use of a parallelized computer code to design and analyze more efficient, pollution-minimized engines. Bill Butler et al. examine the potential of giant magnetoresistance for improving electronic data storage systems. Srdan Simunovic et al. discuss car crash modeling and analyze the performance of lightweight materials to accelerate their use in automotive applications. B. Radhakrishnan et al. show that computer simulations shed light on material microstructure. Laura Toran et al. describe the use of high-performance computing to understand and facilitate cleanup or containment strategies for contaminated groundwater sites. The existence of a new “hexatic” phase between solid and liquid is shown through computations by Mark Mostoller et al. Edge dislocations in silicon are explored by Ted Kaplan et al. Ed Uberbacher describes the important role that computation has had in the Human Genome Project and its expected significant role in the future.
Another article on our scientific computing covers DOE 2000 efforts at ORNL in which electron microscopes are operated remotely for research projects and electronic notebooks are used by scientists across the nation to record, share, and search information on instrument use and experimental results. The final article by Kimberly Barnes et al. focuses on ORNL research in adapting computer technologies to allow geographically dispersed decision makers to respond quickly in a collaborative environment to better manage disasters. Thomas Zacharia, director, ORNL’s Computer Science and Mathematics Division
Next article | Contents | Search | Mail | Review Home Page | ORNL Home Page