Research net set to fly

The network research community is building an $80 million nationwide optical network that advocates say is the most ambitious and promising test bed since the creation of Arpanet - the Internet's precursor - in the late 1960s.

The network research community is building an $80 million nationwide optical network that advocates say is the most ambitious and promising test bed since the creation of Arpanet - the Internet's precursor - in the late 1960s.

Called the National LambdaRail (NLR), the privately funded network will link universities in a dozen U.S. cities with 40 optical wavelengths, each operating at 10G bit/sec. Cisco is providing the multiplexers, switches and routers for the NLR, and Level 3 is supplying the dark fiber.

The first NLR link - from Chicago to Pittsburgh - was lit in November. Additional cities from Washington, D.C., to Seattle are scheduled to be connected by May (see map below).

Developments from the NLR could make their way into commercial offerings in as little as 18 to 24 months, participants say. Among the likely products are next-generation protocols, network management tools, rapid provisioning capabilities and security techniques.

NLR is "a giant step forward,'' says Ron Johnson, vice president of computing and communications in the Office of the Provost at the University of Washington, which is a founding member of the NLR. "The U.S. has lost its lead in Internet innovation. That's likely to be coming from Asia and Europe in terms of high-end network deployments and high-end applications. That's not good for our research community, and it's not good for the U.S. economy.''

Taking back the lead

The NLR is an attempt by the U.S. research community to seize back the lead on Internet development.

More than a decade has passed since U.S. academic, corporate and government researchers had access to a truly experimental, nationwide high-speed network.

In the late 1980s and early 1990s, the U.S. research community had access to the National Science Foundation's NSFnet, which linked five supercomputer centers at 45M bit/sec. However, NSFnet was used primarily for high-end scientific applications such as weather modeling and aerospace simulations rather than network traffic analysis and protocol testing.

The NLR, on the other hand, will reserve half its bandwidth for network-oriented research, while the other half will be available for scientific applications. That's why proponents compare it to the Arpanet, which was a military-funded network research project. The NLR can support experimentation at the optical, switching, routing, middleware and application layers of the network.

Disruptive, next-generation Internet technologies are "not going to be discovered in some little lab but in connecting to a real-world network and seeing what happens,'' Johnson says. "That's where the innovation came from with Arpanet and NSFnet. NLR gives us a tool to work on those sorts of applications.''

The NLR demonstrates "a level of collaboration that hasn't existed on a national scale aimed at new technologies and network research since the early days of the Arpanet,'' agrees Tracy Futhey, vice president for IT and CIO at Duke University and chair of the NLR board. "It's a really exciting project.''

The network project is an outgrowth of several years of talk within U.S. academic circles about the need for a new infrastructure for network experimentation and development. That's because network researchers can't run many experiments on the Internet2, a production network that links 200-plus U.S. universities with one 10G bit/sec link. College students swapping music and video files consume much of Internet2 bandwidth.

"With Internet2, we have a very big highway - 10G bit/sec - but it's one lane,'' explains NLR CEO Tom West. "The problem for network researchers is the freshman downloading Napster or the astronomer who is eating up the bandwidth . . . NLR is 40 lanes vs. one lane.''

The NLR is a five-year project, and half the funds for the project have been committed. Universities are financing the NLR, which has received no federal funding. In contrast, Arpanet and NSFnet were U.S. government projects.

Duke University is spending $1 million this year to participate in the NLR and to upgrade its regional and campus networks to support 10G bit/sec. Duke has committed to participate in the NLR for the next five years.

"Duke is making a multimillion dollar commitment that optical networking and high speeds are something that we have to have from the desktop on out,'' Futhey says. "This gives us the opportunity to have our faculty access supercomputing centers and other academic institutions across the nation.''

The NLR is being built using optical dense wavelength division multiplexing (DWDM) technology. It can support 40 simultaneous light wavelengths, but only four wavelengths are being lit during the initial network rollout. The other wavelengths will be added as additional funds are committed to the project.

On top of the optical DWDM network, the NLR is deploying a switched Ethernet network and a routed IP network. Proponents say the NLR is the first wide-area use of 10G Ethernet, which is common in corporate LANs.

Thanks to the plunge in dark fiber prices over the last three years, the NLR is buying its own network footprint and will provision the network itself. This will be the first time the university community has owned a network test bed.

"When I talk to service providers they ask why we aren't buying managed services,'' West says. "It's a control issue. It's recognizing that we have multiple [network] layer needs. We do believe [ownership] has cost benefits.''

Putting it to use

The NLR will be used for a wide range of network research from optics to traffic measurement and application development.

"Researchers developing new optic technology need breakable waves that they can [run experiments] on. They need dark fiber,'' West explains. "Another camp of researchers . . . wants to be able to go into the network operations center and gather data and measure the data. New techniques will come out of that research to improve the flow of information and security.''

Research on the NLR is scheduled to begin next summer. Although the initial experiments haven't been chosen yet, NLR members anticipate running scientific applications that require processing and storing massive amounts of data in a distributed fashion over high-speed network connections.

"A lot of research in the science space is very similar to the modeling and simulation done by the oil industry or banks,'' says Dave Richardson, manager of network engineering and special projects for the University of Washington's Computing and Communications Division.

"There will be some pretty clear wins for enterprises that will come out of that type of work," he adds.

Corporate network managers likely will benefit from lessons learned from the NLR about operating 10G Ethernet on a WAN. That service is not yet commercially available, and Cisco is tweaking its products to support this aspect of the NLR.

"We're looking at the ability to create point-to-point or point-to-multipoint switched Ethernet virtual private networks between different parts of the network depending on what the researchers need,'' Richardson says. "I could see that fitting very well into a corporate network that has large campuses in multiple locations.''

The two network companies participating in the NLR so far - Cisco and Level 3 - predict that the experimental test bed will spawn new offerings for corporate network managers during the next two to eight years.

Cisco officials won't quantify the company's investment in the NLR, but says it is "substantial.''

The company plans to use the test bed for its own research as well as giving grants to university researchers that take advantage of the backbone.

The NLR uses shipping Cisco products, but Cisco plans to test early beta versions of new products on the backbone.

Level 3 bid on the project because it wanted a glimpse into the Internet's future.

No sideline sitters

"We didn't want to be sitting on the sidelines while advancements in network research were happening,'' says Geoff Jordan, vice president of research and education for Level 3. "We didn't want to wait for the [research] to trickle through a standards body or six years later have a hardware vendor tell us what the latest and greatest is.''

Whether the NLR will be as influential as the Arpanet or NSFnet remains to be seen.

"For the last 10 years, most of the high-speed experimental networks have been developed in support of application use as opposed to network research," says Paul Barford, assistant professor of computer science at the University of Wisconsin at Madison and an NLR board member. "NLR is timely and essential to have for the Internet to grow and scale as it has for the last 30 years.''

Editors' Picks
Join the discussion
Be the first to comment on this article. Our Commenting Policies