Advances in networking technology and computational infrastructure make it possible to construct large-scale high-performance distributed computing environments, or computational grids, that provide dependable, consistent, and pervasive access to high-end computational resources. These environments have the potential to change fundamentally the way we think about computing, as our ability to compute will no longer be limited to the resources we currently have on hand. For example, the ability to integrate TFLOP/s computing resources on demand will allow us to integrate sophisticated analysis, image processing, and real-time control into scientific instruments such as microscopes, telescopes, and MRI machines. Or, we can call upon multiple supercomputers to address ultra-large scientific simulations, perform computationally intensive analyses of distributed PetaByte datasets, or simply steer a remote simulation from our desktop.
Computational grid applications place new demands on both networks and computational infrastructure. For networks, these applications introduce a need for high peak performance, performance guarantees, and high-speed interfaces to supercomputers. For supercomputers, grid applications introduce a need for on-demand scheduling, rather than the batch scheduling traditionally used in supercomputer systems. In addition, sophisticated software is required to provide users with a uniform view of geographically distributed resources.
Scientists at Argonne National Laboratory are working in collaboration with researchers at other laboratories and universities to create a large-scale prototype of such future computational grids. This prototype, called the Globus Ubiquitous Supercomputing Testbed, or GUSTO, comprises supercomputers and other computer systems at national laboratories and other institutions across the United States and in Europe. These systems are connected by a variety of high-speed networks, including ESnet and vBNS. Advanced management software developed in the Globus research project is being deployed on these systems, with the goal of providing a uniform set of services for authentication, scheduling, information, and so forth. (For more information on Globus, see http://www.globus.org.) The GUSTO testbed was unveiled at the SC'97 conference, held in San Jose in November 1997. At the conference, ten different application groups used computers at 15 different sites in the U.S., Germany, and Sweden, totaling 3600 processors and 2 TFLOP/s of compute power. Computers at Argonne, NERSC, and LANL were included in the testbed.