NCSA NEWS

When I-Soft was introduced in 1995 as the software environment for a networking experiment called I-WAY, it was cranky and incomplete. Its software libraries brought some uniformity to the idiosyncratic computing systems they were connecting but not enough to create the transparent environment for which the researchers were striving. Ian Foster, the computer scientist from Argonne National Laboratory who led the development of I-Soft, remembers manually configuring systems as the networks and applications threw unforeseen complications at the fledgling system. The software was a long way from its goal of making distributed computing as simple as placing a long-distance telephone call.

Three years later, and now metamorphosed into Globus, this innovative software still requires operator assistance, but its new automated capabilities are beginning to push the bounds of computing. A case in point is the recent Synthetic Forces Express (SF Express) simulation between the California Institute of Technology (Caltech), the Jet Propulsion Laboratory, and the Naval Command, Control and Ocean Surveillance Center, Naval Research and Development division, funded by the Defense Advanced Research Projects Agency (DARPA).

On March 16, 1998, the SF Express team simulated more than 100,000 vehicles (such as tanks and trucks) in an hour-and-a-half desert battle between opposing "red and blue" forces. The skirmish was the largest of its kind -- consuming 1,386 processors on 13 high-performance computers at nine different sites spanning seven time zones. With Globus's new resource management capabilities, the team was able to initiate the run from a single workstation using a set of coordinated start-up scripts.

"We thought we had to automate resource management to do metacomputing on the scale we envisioned because [an automated system] is less error prone, requires less human time, and is a stepping stone for making this kind of computing more practical," says Paul Messina, director of the Center for Advanced Computing Research (CACR) at Caltech and chief architect of the National Partnership for Advanced Computational Infrastructure (NPACI). "Globus is that stepping stone."

Globus provides a software infrastructure that enables applications to treat distributed, heterogeneous computing resources as a single virtual machine. Foster, a member of the Alliance, has been building these and other new core capabilities with his long-time colleague Carl Kesselman, from the University of Southern California's Information Sciences Institute (ISI), an NPACI partner. The SF Express team was aware of Globus's capabilities, particularly its new resource management features, because the software libraries and utilities are key components of the computational grids being prototyped by both the Alliance and NPACI.

"This DARPA-funded project really benefited from NPACI and Alliance technologies and computing resources," says Messina.