Ken Kennedy and the CRPC
Source: Science magazine, March 7, 1997
By Jeffrey Mervis
For computer scientist Ken Kennedy, the search for high-end performance
never ends. As director of the Center for Research on Parallel Computation
at Rice University, one of the National Science Foundation's (NSF's)
showcase science and technology centers, he has direct access to powerful
IBM, Cray, and Intel supercomputers. But he also needs high speed
connections to researchers in other institutions. So last summer,
Kennedy's center joined with two local universities in a successful bid to
hook up to NSF's very high speed Backbone Network Service (vBNS), a network
initially created to link five NSF-supported supercomputing facilities.
The center also is competing for a major role in the next iteration of
NSF's supercomputing program.
These programs are allowing researchers like Kennedy to redefine what it
means to be connected. Their scientific needs have long since exceeded the
capabilities of the Internet, the once-proud federal creation that has
become a victim of its own popularity, and led them to hook onto more
capable networks like vBNS to share and manipulate vast amounts of data.
But even these high-speed networks have their limitations: Access is
costly and limited (see table), and in general, the networks don't connect
with each other. Fortunately for Kennedy and other data-hungry
researchers, two separate new initiative have emerged to help lead the way
toward an even more connected future.
One is a 5-year, $500 million program, called the Next Generation Internet
(NGI), that President Clinton announced with great fanfare last October in
the heat of the election campaign (Science, 18 October 1996, p. 335). The
second is a loose-knit university initiative, called Internet-2, to
upgrade campus networks and develop educational applications that make use
of these improved links.
Although these initiatives and existing high-speed agency networks have
rung up independently, all are integral to creating the next U.S.
information highway. Larry Smarr, director of the NSF-funded National
Center for Supercomputing Applications, likens NGI to the top of a
three-layer cake. Internet-2 provides the foundation for universities to
take advantage of improved networking, he says, while individual agency
programs like vBNS sit in the middle, generating the applications that will
create demand for the high-end systems. "The president has said that NGI
is part of his bridge to the 21st century," says Smarr. "And the middle
layer will produce the success stories to justify the cost of Internet-2."
The White House's direct involvement with NGI is likely to make it the most
visible of these efforts. In the 5 months since Clinton announced it,
federal officials have been beavering away at a plan. They are talking
about giving 100 sites around the United States a connection 100 times
more powerful than what is now available on the Internet and wiring 10
sites with 1000 times the present capacity. The Defense Advanced Research
Projects Agency, with $40 million requested for 1998, will get the biggest
share of NGI money, followed by a proposed $35 million for the Department
of Energy (DOE) and $25 million divided among NASA, NSF, and the National
Institute of Standards and Technology. Despite its name, NGI will serve
more as a showcase for new technology than an early vision of the next
Internet. "It's not the star of a commercial network," says DOE's Dave
Nelson. "It's a test-bed for what can be accomplished with greater
capacity and innovative uses. Then it's up to industry to make available
what the customer wants."
The task of sketching out a new information superhighway will fall to a
presidential panel of 22 university and industry bigwigs, which met for
the first time last week. A group led by Carnegie Mellon University
computer scientist Raj Reddy and Microsoft's Jim Gray is hoping to complete
a report on NGI by June, while two other subcommittees-one to examine
high-performance computing and the other to address information
management-are being formed to develop recommendations before the panel
goes out of business in 2 years. Not surprisingly, most of the people at
the table are already involved in such efforts, including the man running
The panel's official name-the Advisory Committee on High-Performance
Computing and Communications, Information Technology, and the Next
Generation Internet-reflects how much ground it has been asked to cover.
One major challenge is to meet the needs of society as well as research
institutions, its initial focus. "If only the top researchers have a clear
pipe and the rest of society has clogged pipes, there will be a revolt,"
says Smarr, a panel member.
The panel will also focus on building a more robust and useful network than
the old Internet. "In 20 years, the average person will be able to access
a perabyte (10
bytes) of information for $100,"
predicts Reddy. "That's equivalent to all the printed material that's ever
been created. But what will they do with all that information? And will
it work reliably, without rebooting?"
Out of the dirt. While the panel deliberates, select groups of researchers
have already found relief from what Smarr calls the "one-lane dirt road"
and "cyber-sewer" that the Internet has become as a result of its
exponential growth. Help has come in the form of specialized networks set
up in the past few years by the four federal agencies with the largest
stake in high-end computing-NASA, NSF, and the Departments of Defense and
Energy. But while they give government researchers and outside scientists
funded by those agencies high-speed access to facilities within the
network, they typically rely on the clogged Internet to connect to other
networks. The result is enhanced computing for a relative handful of
researchers working on selected projects.
"Before we connected to vBNS, we had a standard Internet connection," says
University of Pennsylvania physicist Robert Hollebeek, co-director of the
NSF-funded National Scaleable Cluster Project at Penn, the University of
Maryland, and the University of Illinois, Chicago. The 2-year-old project,
which hopes to provide researchers at the three institutions with access to
high-end computing from their desktops, was one of 13 projects chosen by
NSF in a first round of competition last summer to expand vBNS. "You can't
do high-volume, high-speed work without it," says Hollebeek. NSF expects
to announce another round of winners next month from a pool of 50
proposals, on its way toward linking 100 institutions.
But even vBNS has its limitations. One is its high cost: Hollebeek's
$350,000 grant pays for the special equipment needed to connect to the
nearest node, which is 65 miles away, and for the monthly long-distance
phone bills. It's also not for everyone: Researchers on the three
campuses must apply for the chance to be hooked up and show a "legitimate
reason" to use the greater bandwidth and speed, he says.
Penn is also participating in Internet-2, which was begun last fall. The
consortium, which has tripled in size since 34 universities founded it, has
pledged to spend about $500 million over the next 3 to 5 years to upgrade
campus networks and to develop applications that make use of broadband
capacity. Members also hope to link to each other through one or more of
the existing high-end networks.
Mike Roberts of the Washington-based EDUCOM, who is directing the project,
says some applications-multiple media, interactivity, and real-time
collaborations, for example-would be useful to all researchers, while
others, such as distance education and lifelong learning, are particularly
important to universities. "We're focusing on the average faculty member
who says [the current network] is too hard to use or too slow for research
and teaching," says Roberts.
The new presidential panel is expected to make recommendations on achieving
a seamless fit among these emerging networks. "We want to help them
achieve a balanced portfolio among seeding new hardware, developing the
appropriate software, and conducting research on new application," says
Kennedy. "We're looking 10 to 15 years down the road." If they succeed,
thousands of researchers may find themselves in the driver's seat as they
head down the next information highway.