The National Science Foundation hopes to do for Grid computing what it
did for the Internet.
The U.S. government agency was an early backer of the Internet, with the
establishment of the NSFnet network in 1985. Now the agency hopes its
$12.1 million Middleware
Initiative will do the same for Grid computing.
"Much as the NSFnet network in the mid-1980s and early 1990s laid the
groundwork for the dramatic success of the Internet, we expect this new
NSF program to lay foundations for middleware infrastructure and spur
adoption of the advanced services that will define the networks and
distributed systems of tomorrow," said Alan Blatecky, NSF middleware
program director.
The first fruits of the NSF effort, a free comprehensive package of Grid
middleware, was released in May.
The NSF initiative consists of two teams: GRIDS, the Grid Research
Integration Deployment and Support Center, and EDIT, the Enterprise and Desktop
Integration Technologies consortium.
The GRIDS team consists of the Information Sciences Institute at the
University of Southern California, the University of Chicago, the
National Center for Supercomputing Applications at the University of
Illinois at Urbana-Champaign, the San Diego Supercomputer Center at the
University of California at San Diego, and the University of Wisconsin
at Madison, and the EDIT team consists of Internet2, EDUCAUSE, and
the Southeastern Universities Research Association.
The GRIDS Center has contributed core software to the initial NSF
Middleware Initiative release, NMI-R1, the
center said in its July 8 newsletter.
The Globus Toolkit, Condor-G and
Network Weather Service (NWS) combine to form a suite of Grid
applications that are packaged together for easy installation,
configuration and use, the center said. NMI-R1 is expected to become the
standard distribution for these popular tools, upon which applications
will be built by the NSF-backed TeraGrid, the International Virtual Data
Grid Laboratory (IvDGL), the Grid Physics Network (GriPhyN), the Network
for Earthquake Engineering and Simulation (NEES) and other large-scale,
distributed projects.
"But the scalability of GRIDS software means that users at all levels
can benefit - you don't need access to a supercomputer," the center
said. "Today's desktop PC is more than the equal of a 1992
supercomputer.The availability of such affordable computing power can
let scientists and engineers completely reconceptualize their research,
taking advantage of distributed systems for resource sharing,
collaboration and data management."
Built on the Internet and the World Wide Web, the Grid is a new class of
infrastructure that provides scalable, secure, high-performance
mechanisms for discovering and negotiating access to remote resources,
the GRIDS newsletter said.
"Scientists are now sharing data and
instrumentation on an unprecedented scale, and other geographically
distributed groups are beginning to work together in ways that were
previously impossible," the center said.
Grids rely on Internet-based
middleware - including NMI-R1 components like the Globus Toolkit,
Condor-G and NWS - that provides standard protocols for access to
on-line resources.
The GRIDS contributions to NMI-R1 are all open source, open architecture
software that run on Red Hat Linux 7.2 or Solaris 8.0, use Grid Security
Infrastructure (GSI), based on Public Key Infrastructure (PKI), and
together manage complementary requirements for sharing distributed
resources, the center said.
The Globus Toolkit is a community-based set of services and software
libraries that supports Grids and Grid applications, the newsletter
said. The toolkit includes software for security, information
infrastructure, resource management, data management, communication,
fault detection and portability. Each component defines protocols and
application programming interfaces (APIs), while providing open-source
reference implementations in C and (for client-side APIs) in Java. Its
components can be used separately or together to develop Grid
applications.
Condor-G is a highly distributed batch system for job scheduling and
resource management in multi-domain environments, the center said.
Optimized to work with the Globus Toolkit's inter-domain protocols,
Condor-G contributes its own intra-domain resource and job management
methods to harness widely distributed resources as if they all belong to
a single domain. The combined result is a full-featured front-end for
computational Grids, letting the user manage thousands of jobs running
at distributed sites. It provides job monitoring, logging, notification,
policy enforcement, fault tolerance and credential management.
NWS monitors and dynamically forecasts performance of network and
computational resources, using a distributed set of performance sensors
(network monitors, CPU monitors) for instantaneous readings, the center
said. The ability of its numerical models to predict conditions is
analogous to weather forecasting, hence the name. When used with the
Globus Toolkit and Condor-G, it lets dynamic schedulers provide
statistical Quality-of-Service readings. NWS forecasts end-to-end TCP/IP
performance (bandwidth and latency), available CPU percentage and
available non-paged memory, automatically identifying the best technique
to forecast any given resource.
NMI-R1 also includes a tool called KX.509 from the University of
Michigan. It allows Kerberos sites to interact with Grids by converting
a user's credentials from Kerberos to PEM, the format used by the Grid
Security Infrastructure (GSI).
"NMI-enabled Grid environments certainly provide high performance, but
that doesn't mean they require high-performance computers," the
newsletter said. "Although GRIDS software was developed for
high-performance computing, it will work just as well using commodity
desktop PCs. For that matter, today's supercomputers in fact consist of
many such off-the-shelf PCs - albeit numbering in the thousands - that
are configured in clusters that use Grid software to work in concert."
NSF's latest such system is known as the TeraGrid, and it will be
located at four separate sites (two each in Illinois and California)
connected by a 40 gigabit-per-second network, the center said.
The center urged potential Grid users to get started with the NMI-R1
release now. "You might be surprised how straightforward it is to
install, configure and run your own Grid," the newsletter said.