By VINCENT KIERNAN
The Chronicle
Section: Information Technology
Volume 50, Issue 44, Page A29
Tom Goodale needs a really muscular network connection. Forget a typical campus Internet hookup, much less -- horror of horrors! -- dial up.
Mr. Goodale, a research associate at Louisiana State University's Center for Computation & Technology, uses supercomputers to simulate phenomena in astrophysics and relativity, such as collisions between neutron stars. But the data produced by the simulations are so massive -- exceeding a terabyte, or the equivalent of 15,000 music CD's -- that often it is not feasible for Mr. Goodale to retrieve results over existing computer networks.
Instead, he and colleagues resort to copying data from the supercomputer onto tapes or computer hard disks, which are shipped to the researcher.
"Quite often, the most efficient way to get data from one point to another is by FedEx," says Mr. Goodale.
This puts a brake on research because scholars can't take a look at the results and quickly tweak the simulation for another try, says Gabrielle Allen, an associate professor of computer science at Louisiana State. And some scientists are less ambitious than they might otherwise be in their research projects, she says, to avoid the hassles of transporting data hither and yon.
That is changing, however. A group of colleges has banded together to create a national, fiber-optic computer network for academe, called National LambdaRail, which is relying on financial support from colleges, higher-education groups, and state governments. Proponents say the network will provide opportunities for new breakthroughs in research while helping colleges control their networking costs.
"It really is a pivotal time for academia and research," says Harvey B. Newman, a professor of physics at the California Institute of Technology.
But the project faces some hurdles because colleges have little experience in operating long-distance computer networks. And some institutions are deciding, at least for now, that the new network is more powerful than they need or can afford.
How It Will Work
The National LambdaRail organization is creating a web of fiber-optic superhighways across the nation: one east-west route from New York to Seattle via Chicago and Denver; another connecting Jacksonville, Fla., Houston, and Los Angeles; and north-south links from Seattle to Los Angeles, Denver to El Paso, Kansas City, Mo., to Houston, and New York to Jacksonville. Assembling and operating the network is expected to cost $80-million over five years.
Some segments are already working, and the system is scheduled to be complete by next spring. Initially, it will use four wavelengths of light, or "lambdas," as networking cognoscenti call them, after the Greek letter used to designate wavelengths in physics.
The system will not be routinely connected to the Internet. Rather, a participating institution will use a lambda to set up a custom-made, private connection between two locations. That connection would be insulated from hacking and other problems that afflict the Internet.
Institutions that are members of the National LambdaRail organization will have unlimited use of three of the lambdas. The fourth is reserved for the Internet2 advanced-networking consortium, which is contributing a total of $10-million to the National LambdaRail project over five years. Internet2 will use that lambda for research on technologies that eventually could be used in a successor to its Abilene high-speed network. "This is knowledge that we're all going to need within the next five years to keep growing the Internet," says Douglas E. Van Houweling, president of Internet2.
Each lambda will be a yawning maw able to swallow 10 gigabits of data each second. A standard dial-up connection would take more than two weeks to move that much data. Moreover, the fiber-optic lines in National LambdaRail have the capacity for at least 40 different lambdas simultaneously.
Indeed, its capacity is LambdaRail's big selling point. Data from some experiments are expected to soon top one petabyte, or the amount of information that could be contained in 20 million four-drawer filing cabinets. For example, the Large Hadron Collider, a particle accelerator that is expected to become operational at the CERN laboratory in Switzerland in 2007, will generate petabytes of data for each experiment conducted there. Each experiment will have 2,000 researchers scattered around the globe who will want access to that data.
But LambdaRail could also be useful for seemingly more-mundane applications, such as high-quality videoconferencing, which consumes a huge amount of bandwidth.
Buying, Not Leasing
Unlike the networks currently used by academe, the fiber-optic lines in National LambdaRail will not be leased from a telecommunications company. Instead, the National LambdaRail organization will own the lines. The system uses "dark fiber," or fiber-optic lines laid by telecommunications companies in the 1990's dot-com boom that are currently unused. The lines now can be acquired at fire-sale prices (The Chronicle, March 14, 2003).
Those lines are being bought without any help from the federal government. Colleges and state governments are underwriting the project's entire budget. Cisco Systems Inc., which makes equipment used in networks, also is a member and is selling equipment to the project at a sharp discount. The layout of the network reflects which institutions or states have been willing to pony up the $5-million membership fee.
"We're sort of following the money," says Thomas W. West, president and CEO of National LambdaRail.
That determined the path of the network's southern route. Originally, it was to run from Phoenix to Dallas and on to Atlanta, bypassing states such as New Mexico and Louisiana. Then the Southeastern Universities Research Association, a consortium of more than 60 institutions, offered LambdaRail access to dark fiber that AT&T had donated to it, but with a string attached: LambdaRail had to move its southern route farther south. That request dovetailed nicely with decisions by institutions in New Mexico and Texas to join LambdaRail, so the route was revised to stretch from Los Angeles to Jacksonville.
While this strategy for determining the system's layout makes sure that its construction costs are covered, it also bypasses other regions.
The most glaring omission is the upper West and Midwest. In all, 16 to 18 states will have no service from National LambdaRail under its current blueprint, says Tracy Futhey, the vice president for information technology at Duke University who also chairs the Board of Directors of National LambdaRail.
Mr. West says he would like to extend LambdaRail from Chicago to Minneapolis and on to Seattle "if at all possible," but that could cost as much as $25-million. "That's going to take some help from external sources," he says, because institutions in the region are unlikely to have the funds to pay the whole cost.
Institutions in the upper Midwest certainly are interested. The University of Wisconsin and University of Minnesota systems are arranging for their own dark-fiber connections to LambdaRail in Chicago. They and 14 other institutions have formed the Northern Tier Network Consortium to acquire fiber-optic service following railroad tracks, Interstate 94, and Interstate 90. That network could connect participating institutions to the National LambdaRail via Boise, Idaho, Chicago, or Seattle, says Bonita M. Neas, an assistant vice president at North Dakota State University who is the vice chair of the consortium's executive committee.
The federal government played a key role in financing high-speed network links during the Internet's infancy. But Sangtae Kim, director of the National Science Foundation's division of shared cyberinfrastructure, is noncommittal about whether the agency would help bankroll an expansion of National LambdaRail. "It's a question of how it helps drive the science," Mr. Kim says.
Buying a membership in National LambdaRail is not the only cost in getting on the fiber-optic superhighway. Researchers who want to use LambdaRail must have the proper fiber-optic connection in their laboratories or offices. That could be pricey for institutions that have not yet invested in fiber-optic campus networking.
"It's going to be a challenge for some campuses," says Duke's Ms. Futhey.
A college also has to arrange for a link from its own campus network to LambdaRail's nearest connection point. For some, that's not difficult: California institutions, for example, plan to use a statewide optical network operated by the Corporation for Education Network Initiatives in California, a nonprofit corporation serving elementary and secondary schools and colleges in that state.
In some places, colleges or states are establishing regional optical networks that will be hooked into LambdaRail, much as a superhighway ringing a city can intersect with interstate highways running from other parts of the country. For example, nine Florida institutions have formed a nonprofit company to operate their own statewide optical network, called Florida LambdaRail, which will connect to National LambdaRail in Jacksonville. Similarly, Georgia State University has created a nonprofit company, called Southern Light Rail, that could provide connections to the national network for institutions in Alabama, Georgia, Kentucky, South Carolina, and Tennessee. Thirty Texas institutions have formed the Lonestar Education and Research Network, which will develop a statewide fiber-optic network connected to National LambdaRail.
Gubernatorial Endorsement
Perhaps nowhere has LambdaRail had a higher political profile than in Louisiana. That state's governor, Kathleen B. Blanco, explicitly endorsed participation in the system during her annual state-of-the state address in March.
"Investing in technology for our universities will allow us to become one of the select members of the National LambdaRail initiative," she told lawmakers. "Through these efforts, we can distinguish Louisiana as a major player in high-performance computing and network technology and secure tremendous economic development gains in the future."
In March, the state Board of Regents agreed to pay the $5-million cost for joining National LambdaRail, and last month the legislature agreed to underwrite a $25-million statewide fiber-optic network that would connect colleges to one another and to LambdaRail.
"We believe that we are preparing ourselves to be big players in this technology," says E. Joseph Savoie, the state's commissioner of higher education. "We hope it will mean a brighter future for our state."
Researchers and administrators say that a LambdaRail connection will help them win more research grants because the researchers will be well positioned to collaborate with other scholars around the globe.
Gary R. Crane, director of IT initiatives for the Southeastern Universities Research Association, likens the situation to the impact that railroads had on small towns. A lot of Western towns disappeared when bypassed by railroads, he says.
Proponents in Louisiana and elsewhere say that National LambdaRail will promote industry because the LambdaRail organization will impose few limits on how the system can be used. Commercial development of the Internet, they say, was stifled because the National Science Foundation, which financed the network in its early years, forbade commercial use of the network.
However, the network still will not have an "anything goes" policy. National LambdaRail's bylaws dictate that at least half of the system's capacity must be dedicated to network research.
But Will It Work?
Supporters evince confidence that National LambdaRail will work as advertised. "Technologically, it's not risky," says David J. Farber, a professor of computer science and public policy at Carnegie Mellon University who is National LambdaRail's chief scientist.
But other challenges remain, particularly with respect to financial and operating issues. For example, National LambdaRail has not yet decided on a method for determining who gets to use a given lambda at a given time. Nor has it figured out how much to charge researchers who want to acquire sole use of one of the system's 36 lambdas that initially will not be used.
Indeed, the plan for LambdaRail is "a risky proposal," says Daniel J. Updegrove, vice president for information technology at the University of Texas at Austin and chairman of the Lonestar network's Board of Directors. "We don't really know how much it's going to cost to engineer and run this stuff," he says.
Still, many officials think the risk is manageable. "There's always a risk when you do something innovative," says J. Gary Augustson, vice provost for information technology at Pennsylvania State University.
Cost, rather than risk, is a factor in the decisions by some smaller institutions -- particularly those with few large-scale research projects -- not to join National LambdaRail, at least for now. "This is, for some schools, considered a luxury," says Marla J. Meehl, manager of network engineering and telecommunications at the National Center for Atmospheric Research, in Boulder, Colo., which is a member of National LambdaRail.
LambdaRail officials agree. "NLR is not, and should not be, a network that everybody feels they have to participate in," says Duke's Ms. Futhey.
One institution that has opted not to join is the University of Colorado at Colorado Springs. Currently, its campus's connection to the Internet can carry 20 megabits per second. Although that is less than 1 percent of the capacity of one of LambdaRail's lambdas, the connection is sufficient for the college's needs, particularly because a LambdaRail connection would have cost at least $50,000 annually, says Jerry Wilson, director of information technology at the university.
"There wasn't a lot of people saying, 'My God, without this, my research project is going down the tubes,'" he says.
Even a few research-intensive institutions are deciding to pass. One is the University of South Florida, in Tampa, which belongs to the Internet2 consortium but has elected to remain out of LambdaRail, even though Florida's statewide optical network will run right through Tampa.
"How much is enough?" asked a memorandum on LambdaRail prepared this spring by the university's staff. The answer was what the university already has: a 155-megabit connection provided by Internet2's Abilene backbone. Joining LambdaRail would have cost the campus at least $500,000 annually for five years, the report said, recommending that the campus not join.
But many see National LambdaRail as a bargain. "This is really the next way to do things," says Lawrence H. Landweber, a professor emeritus of computer science at the University of Wisconsin at Madison who is a senior adviser at the National Science Foundation. "It's incredibly cheap for what you get."
"Once it starts," says Mr. West, "people will see the value of the investment."
http://chronicle.com
http://chronicle.com/weekly/v50/i44/44a02901.htm
Publish Date:
07-01-2004