RESEARCH FOCUS: I-WAY Software Tools and Applications

Ian Foster, Argonne National Laboratory; Carl Kesselman, California Institute of Technology; Dennis Gannon, Indiana University

The Information Wide Area Year (I-WAY) project was conceived in early 1995 as a large-scale testbed for innovative high-performance and geographically distributed applications. A multi-institutional team led by researchers at Argonne National Laboratory and the Electronic Visualization Laboratory at the University of Illinois at Chicago worked on an accelerated schedule to produce a national-scale Asynchronous Transfer Mode (ATM) network connecting supercomputers, mass storage systems, and advanced visualization devices at 17 different sites within North America. This network was deployed at the SUPERCOMPUTING '95 conference in San Diego in December and was used by more than 60 application groups for experiments in high-performance computing, collaborative design, and the coupling of remote supercomputers and databases into local environments.

The I-WAY Software Environment

A central part of the I-WAY experiment was the design, implementation, and evaluation of a software environment called I-Soft, providing uniform authentication, resource reservation, process creation, and communication functions across I-WAY resources. This environment, developed by a team led by Ian Foster from Argonne, allows users to log on to a single I-WAY Point of Presence (I-POP) workstation, request resources on computers distributed across the nation, and initiate parallel programs using standard tools. I-POP machines are located at each I-WAY site and provide a uniform management environment.

Much of the I-Soft work was built upon CRPC technology, specifically, on the facilities provided by the Nexus runtime system. Nexus has been developed over the past two years at Argonne and Caltech and provides protocol selection and interoperability mechanisms that are essential in heterogeneous environments such as the I-WAY. Nexus was at the core of almost every tool used on the I-WAY. For example, to provide a portable, high-performance implementation of the standard Message Passing Interface, Argonne researchers Jonathan Geisler and Steven Tuecke adapted the Argonne MPICH implementation of MPI to use Nexus. Similarly, Mike Papka used Nexus to develop a specialized communication library called CAVEcomm for connecting multiple immersive virtual reality environments to supercomputers. Even the I-WAY scheduler used Nexus mechanisms--Northwestern University student Warren Smith wrote this using nPerl, a Nexus-based communication library for the popular Perl scripting language.

I-WAY Applications

Some applications focused on connecting a single supercomputer, database, or scientific instrument to an immersive virtual reality environment (CAVE(tm) or ImmersaDesk(tm)). Others sought to couple multiple supercomputers in order to solve extremely large problems. A third set coupled multiple virtual environments so that users at different locations could interact with each other and with supercomputer simulations.

One interesting application was developed by Craig Lee of the Aerospace Corporation and Carl Kesselman of Caltech. They devised a system that uses infrared and visual image data obtained in real time from weather satellites to detect cloud locations. Raw image data was obtained at a satellite downlink located at the Aerospace Corporation in El Segundo, California. This data was sent over the I-WAY to an IBM SP-2 parallel computer at Argonne. Processing on the supercomputer yielded information about the position and altitude of clouds, which was then geolocated and rendered stereoscopically on the ImmersaDesk in San Diego. Supercomputing participants used the interactive capabilities of the ImmersaDesk to fly around and through clouds detected by satellites only minutes earlier. This complex and heterogeneous application was implemented by using the Compositional C++ language, developed by Kesselman and K. Mani Chandy at Caltech, and was executed on the I-WAY by using Nexus mechanisms.

Dennis Gannon of CRPC affiliate site Indiana University was involved in the development of another application that simulated and visualized a hypothetical collision between the Andromeda and Milky Way galaxies. Andromeda is the nearest great spiral galaxy to the Milky Way--our home galaxy--and is on a collision course with us. The Indiana team, led by Kate Keahey, Shelby Yang, and Peter Beckman, worked with Mike Norman and Greg Bryan of the National Center for Supercomputing Applications, Lars Hernquist and John Dubinsky of the University of California at Santa Cruz, and Joel Welling and Ralph Roskies of the Pittsburgh Supercomputer Center to develop a simulation of this collision that was functionally decomposed into four components. Each component was developed by a different member institution of the Grand Challenge Cosmology Consortium, a National Science Foundation-funded Grand Challenge application group. These four components were then distributed over the I-WAY, each operating on a different supercomputer. Two of the components involved a distributed N-body simulation of the stars, the third was a hydrodynamic simulation of the interstellar gas dynamics, and the fourth was a distributed visualization system that ran on the ImmersaDesk at the conference site.

A Promising Future

The successful deployment of the I-WAY network and the more than 60 applications at SUPERCOMPUTING '95 provided a dramatic demonstration of the feasibility and promise of high-performance distributed computing. It provided tantalizing glimpses of what can be achieved when application scientists have access to high-performance networking, computing, and graphics. Moreover, it generated valuable insights into the hard technical problems that must be addressed to ensure progress in this area. I-WAY participants are currently analyzing the lessons learned from this experience and are developing plans for next- generation I-WAY systems.

For more information, see and .

Table of Contents