Globus Toolkit enables Grid computing
Grid collaborators visualized two colliding black holes. Shown are the so-called event horizons-whatever falls in, even light, cannot escape again. Visualization by Argonne research partner Max Planck Institute for Gravitational Physics, Germany.
Argonne technology is bringing closer the day when the Internet can let people share computing, storage, data, programs and other resources as easily as the electric power grid allows people and energy companies to share electricity.
The "Grid" will allow researchers at many facilities to integrate instruments, displays and computational and information resources over a variety of computer platforms to attack increasingly complex challenges faced by scientists, educators, industry and even consumers.
The Grid is more than just a collection of resources. It also includes a set of protocols, services and tools for enabling Grid computing--the Globus ToolkitTM. The toolkit, which was awarded an R&D 100 Award from R&D magazine in 2002, is one of the most widely used systems for Grid computing today. Its components enable the secure, scalable and coordinated use of resources in dynamic, multi-institutional "virtual organizations." The toolkit was developed by the Globus ProjectTM, centered at Argonne's Mathematics and Computer Science Division (MCS), the University of Chicago's Distributed Systems Laboratory and the University of Southern California's Information Sciences Institute.
Growing industrial interest
Industry interest in grids and the Globus Toolkit is growing rapidly. Both Microsoft and IBM are now providing funds to support distributed computing based on these technologies. Entropia is integrating its commercial software with that of the Globus Toolkit, and Platform Computing Inc. is collaborating with the Globus Project to provide a commercially supported version. Nine other companies worldwide have adopted the toolkit as their defacto standard Grid technology platform.
"We certainly welcome this support," said Ian Foster, associate director of MCS Division and professor of computer science at the University of Chicago. "The Globus Project is staunchly committed to open-architecture software, and industry backing will contribute to the public knowledge base of Grid computing. The potential benefit to users is enormous."
Foster and his MCS colleague Steven Tuecke lead Globus Project activities at Argonne and have spearheaded efforts to engage industry in Globus Toolkit development and applications.
Globus technologies are being applied in a wide range of leading-edge activities, including GriPhyN, a physics network to explore applications requiring quadrillions of operations per second; NEESGrid, a national virtual laboratory for earthquake engineering; and the Cactus astrophysics computing portal.
Several national collaboratories supported by the new DOE Scientific Discovery through Advanced Computing program will also use Globus tools:
- Collaboratory for Multichemical Sciences
- Earth Systems Grid II
- National Fusion Collaboratory, and
- Particle Physics Data Grid.
Creating a virtual supercomputer
In a recent major test of Grid computing, the Globus Toolkit harnessed the power of multiple supercomputers with different operating systems, more than quadrupling the system's computing efficiency and earning a prestigious Gordon Bell prize for a team of scientists from Argonne, the University of Chicago, Northern Illinois University and the Max Planck Institute for Gravitational Physics in Germany.
The team created a "virtual supercomputer" to simulate the evolution of gravitational waves according to Einstein's theory of general relativity. The experiments were the largest-ever simulations involving Einstein's general relativity equations and modeled the gravitational effects of the collision of black holes. The supercomputer comprised 512 processors from three SGI Origin2000 machines at the National Center for Supercomputing Applications in Illinois and a 1,024-processor IBM SP2 at the San Diego Supercomputing Center in California.
The research team used three software systems: Argonne's Globus Toolkit; MPICH-G2, a Grid-enabled version of the standard Message Passing Interface developed at Northern Illinois University and Argonne; and the Cactus computational toolkit for scientists and engineers developed by the Max Planck Institute for Gravitational Physics.
Researchers tested various configurations. After a baseline run, researchers modified the software and achieved 63 percent efficiency of 1,500 central processing units (CPUs), meaning the virtual supercomputer spent almost two-thirds of its time crunching numbers and only one-third waiting for data coming from other machines. This was 14 percent more efficient than the baseline. The team also ran the experiments using only 1,140 CPUs and boosted efficiency to 88 percent.
The data these experiments generate are of interest to astrophysicists looking for gravitational signals from celestial events such as the collision and merger of black holes. The results are also significant for computer scientists as they try to improve Grid computing efficiency, said Foster.
"The experiments underscore the potential of linking multiple supercomputers with different operating systems for large-scale simulations across a computational grid," Foster said. "We can merge all their computing power to focus on the most challenging problems of science."
Funding for this work was provided by DOE's Office of Science, the National Science Foundation, NASA, Defense Advanced Research Projects Agency, IBM and Microsoft.