Applications:
U.S. SCIENTIFIC COMPUTING ENTERS NEW ERA WITH Grid3
Scientists of the Open Science Grid Consortium, who met at a workshop at
Harvard University on Sept. 9, announced the success of a nine-month trial
operation of a U.S. data Grid for particle physics experiments and other
scientific applications. The prototype Grid, called Grid3, uses the Internet
to combine the computational resources of 26 universities and national
laboratories across the United States to serve the computing needs of more
than 10 research groups in particle physics, astrophysics, bioinformatics and
computer science.
"This is a breakthrough for scientific Grid computing," said Paul Avery,
professor of physics from the University of Florida, and director of the
National Science Foundation's International Virtual Data Grid Laboratory.
"Grid3 is a simple Grid, but it does what a computer Grid is supposed to do
for scientific research. It combines computing power from multiple sources and
transports data to and from offsite locations so that individual scientists
can solve scientific problems at their desktop computers."
Grids are becoming critical to data storage and analysis in a range of
data-intensive sciences. For example, when they begin operating in 2007, the
world's largest-ever particle physics experiments—the ATLAS and CMS
experiments at the Large Hadron Collider, a particle accelerator now under
construction at CERN, the European Particle Physics Laboratory in Geneva,
Switzerland—will depend on Grid technologies for storage, transport and
analysis of unprecedented volumes of data by collaborators at universities and
laboratories across the globe. Each LHC experiment numbers over 2,000
collaborators.
To ensure that ATLAS and CMS can meet their computational needs when
scientific operations begin, the experiments must meet a series of "data
challenges," preoperational exercises of computational and data capacity
tested with simulations of successively larger percentages of the actual data
production and analysis that will occur at LHC experiments.
"For US-ATLAS and US-CMS, the United States teams in the LHC collaborations,
Grid3 has been key to meeting the milestones of our data challenges," said
Boston University physicist James Shank, executive program manager for
US-ATLAS computing. "Grid3 has run stably and with little effort during its
nine months of operation. Grid3 has also provided unique opportunities for
collaboration between the experiments through cooperative use of each others'
computing resources. Both CMS and ATLAS have altered the way they allocate
CPUs, for example, in order to accommodate each others' priorities and
schedules."
While Grid3's simple functionality has supported the experiments' data
challenges thus far, scientists said it must be significantly enhanced to meet
actual analysis needs for LHC data-taking when the experiments begin
operations in 2007. Toward that goal, U.S. researchers and computer scientists
have formed the Open Science Grid Consortium to develop Grid3 into a
production infrastructure that will operate at a larger scale with a broader
base of partners and resources from more organizations, offering more
sophisticated services.
"We have a lot of work to do to take Grid3 to the scale and capabilities of
the Open Science Grid," said Ruth Pordes, associate head of the Computing
Division at the Department of Energy's Fermilab and coordinator of the
Department of Energy's Particle Physics Data Grid project. "But the success of
Grid3 and the enthusiasm for Open Science Grid give us confidence that we can
make it work. US-ATLAS and US-CMS have committed their ongoing computing
programs to the support of the Open Science Grid, not only to benefit their
own experiments but to promote the open use of a shared Grid infrastructure by
the broader science community."
The core technologies for Grid3 as well as for the European Grids are based on
the Virtual Data Toolkit, which includes the NSF middleware initiative
software distributions.
"Achieving a common middleware base across the Grids is the result of a lot of
good collaboration and hard work," said computer science professor Miron
Livny, leader of the Condor Project at the University of Wisconsin.
The Open Science Grid will "federate" with other Grids now under development
around the world to create a truly global data Grid for science. For ATLAS and
CMS, the Open Science Grid will "interoperate" with the LHC Computing Grid in
Europe, and others, to provide a global Grid for LHC experimenters around the
world.
"The demonstrated interoperability of Grid3 and the European LHC Computing
Grid for the movement of data and distribution of computation gives us growing
confidence in the applicability and usability of our infrastructure for the
future of global Grid computing," said computer scientist Ian Baird, head of
Grid deployment at CERN, where rapid development of Grid capability is also
underway.
University of Chicago physicist Rob Gardner, an ATLAS collaborator, cited the
cooperation among Grid3, the LHC Computing Grid and NorduGrid, a
Scandinavian-developed research Grid.
"The ATLAS Collaboration has used the combined resources of the international
LHC Computing Grid, the NorduGrid Project and Grid3 for its data challenge
simulations," said Gardner, who, with Pordes, serves as Grid3 coordinator. "It
is especially gratifying that the groups developing the various Grid
infrastructures have cooperated so that our global experiment community can
run science jobs efficiently across the continents."
Besides the LHC particle physics experiments, Grid3 has also served
computational needs of other researchers. Astrophysicists from the Sloan
Digital Sky Survey, a project to create the world's most comprehensive map of
the sky, used Grid3 to perform data analysis. Collaborators of the proposed
BTeV experiment at Fermilab also used Grid3 for particle event simulations.
Experimenters of LIGO, the Laser Interferometer Gravitational Wave
Observatory, performed an analysis searching for continuous gravitational wave
signals. Experimenters performed two biology applications, a biomolecular
analysis and a protein sequence analysis, on Grid3. Computer scientists have
also used Grid3 for research projects.
Computer scientist Ian Foster, of the Department of Energy's Argonne National
Laboratory and the University of Chicago, is leader of the Globus Alliance, an
organization devoted to the development of common worldwide standards and open
source software for global Grid computing.
"We are especially pleased that scientists from a number of different fields
use Grid3 and show that many areas of research can share the same computing
resources. The mix of applications from experimental science and computer
science research running on Grid3 shows that our underlying Grid protocols are
increasingly robust and flexible," Foster said.
To use Grid3, a scientist must belong to one of the member organizations that
provide computing capacity. Each user registers through an authentication
system to receive a "Grid certificate," which works like a passport to
identify users. US-CMS scientist Rob Harris of Fermilab said he had used Grid3
for particle event simulations.
"Grid3 provides us convenient access to the intensive computing required for
our complex detector simulations," Harris said.
Each Grid3 member site presents a common interface to users.
"The use of common standards in Grid development is an important factor in
determining future compatibility and interoperability among the world's data
Grids," said Vicky White, head of Fermilab's Computing Division. "Grid3 shares
software and infrastructure with other national Grid infrastructures in the
U.S. and Europe. The Open Science Grid will continue to work toward common
protocols that will allow truly ubiquitous access to the data as well as the
computers."
Some 30 scientists from member universities and laboratories collaborated on
the development of Grid3. The National Science Foundation, the Department of
Energy's Office of Science and the member universities provided funding for
the project.
More information about Grid3 is available at www.ivdgl.org/Grid2003/.
Grid3 Member Organizations include:
- Argonne National Laboratory.
- Boston University.
- Brookhaven National Laboratory.
- Caltech.
- Fermi National Accelerator Laboratory.
- Hampton University.
- Harvard University.
- Indiana University.
- Thomas Jefferson National Facility.
- Johns Hopkins University.
- Kyungpook National University/KISTI.
- Lawrence Berkeley National Laboratory.
- Stanford Linear Accelerator.
- University at Buffalo.
- University of California-San Diego.
- University of Chicago.
- University of Florida.
- University of Michigan.
- University of New Mexico.
- University of Oklahoma.
- University of Southern California.
- University of Texas-Arlington.
- University of Texas, Brownsville.
- University of Wisconsin.
- University of Wisconsin-Milwaukee.
- Vanderbilt University.
|