The Internet Reinvented

Tomorrow's Net, driven again by researchers and academia, will include new backbones, new protocols, and new applications.

Daniel P. Dern and Scott Mace

Today's Internet, while faster than ever before and bigger each month than the previous month, is still so slow and otherwise inadequate that an army of expert minds is determined to reinvent it.

Despite broadband backbones in the OC-12 (622 Mbps) range, end-to-end throughput on the public Internet has been measured between LAN-based workstations at as little as 40 Kbps -- the equivalent of what a pair of modems could do and slightly slower than the original ARPANET, a pre-Internet backbone running at 56 Kbps.

For people in the research and academic arenas, this bandwidth and its unpredictable availability are insufficient to support many of the new applications they want and need. In many cases, without adequate networking, they will be forced to buy rather than share very expensive equipment, such as high-voltage electron microscopes. Or they may have to travel to California-Berkeley's Spectro-Microscopy Lab and witness real-time presentation of large data sets from projects such as the University of Michigan's Upper Atmosphere Research Lab.

It isn't simply a question of "more bandwidth, please," either. Today's Internet does not support other features that are as essential to the next generation of networked applications as higher speeds are.

The current Internet delivers what's called best-effort service. Version 4 of TCP/IP, which is what runs on the Internet today, has no provision f or specifying or guaranteeing quality-of-service (QoS) attributes and levels for these attributes. IPv4 also can't reserve bandwidth, assure maximum network latency, or provide adequate security.

Universities and research institutions need these features today. The corporate and consumer worlds have begun to feel the need and see the value for such capabilities in the commercial Internet as well.

Getting to this next-generation Internet requires a new generation of hardware (e.g., switches and routers) and carrier services. It also requires new protocols, new network management tools, and a deeper understanding of the network needs of high-performance applications. It will need major project and program management to deploy and coordinate these changes without disrupting existing Internet service. Ways are needed to make the new capabilities available to users and their applications in a simple, easy-to-specify-and-use fashion -- plus educating and training developers and users.

Such a daunt ing task is beyond the scope of any one vendor, university, or government agency. But it's not beyond the scope of lots of these, working in teams -- and that's what's happening.

Paving the Way

U.S. educational communities have been working together for more than a decade to articulate their networking needs. Several interrelated initiatives are under way that will pave the way -- and begin to construct -- seminal pieces of the next iteration of the Internet. These pieces include:

• The White House's Next-Generation Internet (NGI) initiative
• The National Science Foundation's (NSF) Very High Bandwidth Network Service (VBNS)
• Internet2, an effort of a consortium of universities working with corporate and government partners
• IPng, the next generation of the Internet protocol, aka IPv6

Together, these initiatives are shooting to yield new protocols, new hardware, new software, new knowledge, and new network test-beds demonstrating applications that make use of thei r capabilities.

NGI

NGI is a White House multiagency initiative that was announced in October 1996. Arising from the High-Performance Computing and Communications initiative, NGI is now part of the U.S. government's Large-Scale Networking initiative. The NGI program will be coordinated in the framework of the National Science and Technology Council. High-level strategy will come from the Committee on Computing, Information, and Communications (CCIC), and implementation strategy from the Large-Scale Networking Working Group.

By last fall, researchers were already demonstrating five "precursor applications." NGI includes research into protocols, development, and deployment of high-end test-beds, plus demonstrating applications. It is meeting some of these goals through Internet2 and/or VBNS.

Of all the initiatives, NGI is the one most on the cutting edge, especially its Class 1 networks funded by the Defense Advanced Research Projects Agency (DARPA) and the Department of Defense (DoD) , where the network technologies proper are being tested. Some of these networks aren't even married to TCP/IP. They include:

• Collaborative Advanced Interagency Research Network (CAIRN) offers researchers nationwide a suite of Ascend Gigarouters for experimentation with RSVP, multicast, and other IPv6 protocols.
• The National Transparent Optical Network Consortium (NTONC) plans to build a $40 million prototype network in California to serve as a test-bed and evaluate the performance of advanced optical communications components. These technical advances are expected to lead to a network that can carry 10 to 100 Tb of data per second, a capacity well beyond anything currently planned.
• The Multiwavelength Optical Networking (MONET) network, stretching from Washington, D.C., to New Jersey, aims to figure out how to build a multiwavelength national optical network.
• Advanced Technology Demonstration Network (ATDNet) initially is an OC-48 (2.4 Gbps) network in the Washington, D.C., area. It was created to allow federal agencies to deploy emerging asynchronous transfer mode (ATM) and Synchronous Optical Network (SONET) technologies.
• The Advanced Communications Technology Satellite ATM Internetwork connects several DoD High-Performance Computing centers (a subset known as the Defense Research and Engineering Network [DREN] test-bed) and the Multidimensional Applications Gigabit Internetworking Consortium (MAGIC) and ATDNet gigabit test-beds. Research topics include network signaling, congestion management, ATM and IP multicast, and gateways to non-ATM LANs.

A key NGI goal is to develop and demonstrate two test-beds that are, respectively, 100 and 1000 times faster than today's Internet in terms of end-to-end performance -- meaning about 100 Mbps and 1 Gbps. Network services that NGI will be working on include areas such as transaction security and network management. Much effort is being put into making use of off-the-shelf products and services where they exist, and to making NGI easy for companies to work with. As of last fall, more than 150 Silicon Valley companies were involved as partners.

VBNS

Before NGI, before Internet2, the NSF already was working to provide its constituency with networking beyond what the commercial Internet could deliver. The NSF's answer was to begin a dedicated network, VBNS, to provide next-generation network service to qualified researchers and academic users.

In the spring of 1995, the NSF made a five-year cooperative agreement that was worth up to $50 million with MCI for VBNS. The network, which has been operational since April 1995, links five NSF supercomputer centers, at locations including the Cornell Theory Center and the National Center for Supercomputing Applications (NCSA). Over time, this will spread to about 100 research institutions.

As of last fall, "the VBNS includes 14,000 miles of OC-12 links , and we also offer switched virtual circuits [SVCs]," says Charles Lee, VB NS program manager in MCI Internet engineering at MCI Telecommunications. "So two sites can signal directly into MCI's ATM switches and go directly through the MCI network using Layer 2 end to end without any routers in between.

"It's an incubator for the development of next-generation applications," Lee says. "You must break the chicken-and-egg cycle -- nobody will develop the applications with a network, and vice versa. VBNS is a springboard to the next generations of technology."

MCI is trying to bring native IPv6 to VBNS by March, but this may be subject to shipment of software that supports IPv6.

By June, MCI hopes to bring up reserved-bandwidth services. "Our initial QoS offering adds a reserved-bandwidth service to traditional IP datagram forwarding," Lee says. "This is for applications that need a high bandwidth with assurances of very low loss or delay. It sets up, on a per-application basis, a special path through the Internet, by signaling from the end system using RSVP, and one o f our VBNS routers will translate that into an ATM virtual circuit."

The reserved bandwidth is important, Lee points out, "for when the commercial services can't support it end to end." Similarly, using Protocol-Independent Multicast (PIM) is enabling otherwise infeasible activity such as interconnecting to CAnet, Canada's big research network, to provide IP multicasting.

Although you can currently use the Multicast Backbone (Mbone) for voice and video, "that's generally done by using special routers or workstations to do tunneled IP," Lee says. "The Mbone is basically a low-performance multicast overlay. PIM on the VBNS does it with high performance."

By the end of this year, MCI plans to further enhance VBNS with source-based routing, with more than 100 domestic connections and approximately 20 international connections. And by 2000, VBNS backbone speeds should be up to OC-48.

"We can do very-high-speed wave division multiplexing (WDM) and get very-high bandwidth," says Rick Wilder, senior manager of Internet technology at MCI. "We have some 40-Gbps legs, and in our Reston, Virginia, lab, we have the new Cisco 12000 series routers that are built to accommodate OC-48. But you need to be able to plug it into something, and at present, OC-12 is state of the art in terms of what you can buy in IP routers or ATM switches that's reliable."

(WDM involves packing multiple optical-transmission streams into one fiber by sending each stream on a separate color channel.)

Meanwhile, VBNS-related activity has already borne fruit. MCI developed, and is sharing information on, a monitoring capability that lets the company look at the IP traffic inside cell streams as they flow past at high speeds. This is the first technology transfer MCI can point to from the VBNS effort.

VBNS will also be a part of the NGI efforts, providing a place for testing new applications and trying out cutting-edge network technologies.

Internet2: Academia's Next Stand

After VBNS took flight, un iversities agreed to pool their resources for a new level of internetworking. The result was a project that was dubbed, misleadingly, Internet2.This was misleading in that although it was in pursuit of next-generation Internet technology, it was not intended to replace the existing Internet, nor to build a new network for general users.

UCAID, the University Corporation for Advanced Internet Development, was formed in September 1997 to manage Internet2 and to assist other consortia, such as North Carolina's Gigabit Point of Presence, also known as a GigaPOP (see the figure "The U.S.'s First GigaPOP" ). GigaPOPs will help aggregate traffic from universities, avoiding many of the problems created by the architecture of today's Internet Network Access Points (NAPs).

Nine of UCAID's corporate members -- Advanced Network and Services, Bay Networks, Cisco Systems, Fore Systems, IBM, Newbridge Networks, Nortel, Starburst Communications, and 3Com -- have joined at the partner level. T his means that they have committed to contribute more than $1 million each to Internet2 over the course of the next three to five years.

"When NSF started privatizing, there was a view that our troubles would be over," says Mike Roberts, vice president of Educom, a Washington, D.C., consortium of 600 colleges with interests in information technology for education.

"But we didn't get enough attention in terms of how to arrange for the continuing evolution and for the introduction of new technology into the Internet usefully incorporated. We missed that.

"And we also failed to understand that the private sector, the folks buying up regional nets, the ISPs, would have to concentrate on the bottom line...and that automatically meant that the 'center of gravity' in what they would do to enhance technology would be slower than the research institutions would want."

Internet2, VBNS, and NGI are also interrelated, although Internet2 and VBNS have self-contained missions of their own, independe nt of NGI or each other. At present, VBNS provides backbone network service for Internet2. Internet2 and UCAID are also providing some of the participation by the higher-education arena in the NGI effort. Indeed, Internet2 is seen as fulfilling the first goal of the NGI program, hooking up the 100 top universities and developing next-generation networked applications.

Internet2's plans call for operational testing by the fall of this year, although, using VBNS, some applications have already been demonstrated, including some at meetings that were held in Washington D.C., last fall. Internet2 applications cut across all academic disciplines. Some will be collaborative environments, and others will be digital libraries. Some will facilitate research, and others will enable distance learning.

Internet2 also provides a place to test various policy issues, such as how to cost and charge for bandwidth reservation. It's also a place to experiment with ways to leverage the GigaPOPs, such as with local cac hes and replicating servers, and satellite up- and down-links to improve network efficiency.

Besides the remote instrumentation mentioned earlier, collaboration environments will allow for audio, video, text, and whiteboard discussions in real time. Other applications support new forms of collaboration through 3-D virtual shared presences known as immersive environments. Finally, telemedicine, including remote diagnosis and monitoring, will get a boost from Internet2.

Intensively interactive graphical/multimedia applications are also prime candidates for NGI, involving things like rich scientific visualization, collaborative virtual reality (VR), and 3-D immersive environments, with names such as computer-assisted virtual environment (CAVE); CAVE research network (CAVERN); ImmersaDesk; Narrative, Immersive, Constructivist/Collaborative Environment (NICE); and Tele-Immersion (combining networked VR and video with a lot of computing and data mining).

Although the original Internet2 organizers thought that fewer than two dozen schools would need to be part of the new network, as the word spread, pretty much every major research university in the U.S."wanted in," according to Roberts -- 114 in all.

To participate, each university had to commit up front to half a million dollars to pay for their upgrade of the WAN, to pay $25,000 a year to the central group to cover coordination costs, and to create at least one new application.

Terabits to Come

Unsurprisingly, the research community continues to look ahead to still-faster networks to put to use. Educom's Roberts reports there is discussion of an OC-192 (which is nearly 10 Gbps) network for highly qualified researchers, of which there are between 50 to 100.

"Cutting-edge deployers are already providing 40 to 60 Gbps in a single fiber...and the theoretical limit of a fiber is 100 Tbps," says Craig Partridge of BBN/GTE -- 2000 times more than the current delivered capability and 100 times more than the lab limit.

"That could carry us perhaps for another decade, depending on how things grow...and, of course, we'll also be laying more fiber," Partridge says.

---

Where to Find

CAIRN -- Collaborative Advanced Interagency Research Network
Internet: http://www.cairn.net

Internet2
Internet: http://www.internet2.edu

Multiwavelength Optical Networking (MONET)
Internet: http://www.bell-labs.com/project/MONET/mon_pro.html

Next-Generation Internet (NGI) Initiative
Internet: http://www.ngi.gov

NTONC -- National Transparent Optical Network Consortium
Internet: http://www-phys.llnl.gov/H_Div/photonics/NTONC.html

UCAID (University Corporation for Advanced Internet Development)
Internet: http://www.ucaid.edu

VBNS
Internet: http://www.vbns.net

---

The Proposed NGI Time Line

The Proposed NGI Time Line
Initiative Deliverables	Expected Date
100+-site high-performance test-bed providing OC-3 (155 Mbps) connections over an OC-12 (644 Mbps) infr astructure	1999*
Federal, academic, and industry partnerships conducting applications/networking research on the 100x test-bed	1999
10+-site ultrahigh-performance test-bed providing OC-48 connections (2.5 Gbps)	2000
Tested models for NGI protocols, management tools, QoS provisions, security, and advanced services	2000
100+ high-value applications testing and benefiting from a high-performance test-bed (e.g., remote, real-time, collaborative NGI network control of selected laboratories)	2000
Consortium conducting networking/applications research on the 100x test-bed	2001
Ensure interoperability across multiple carriers among 100+ lead users tested for reliable data communication	2001
Tbps packet switching demonstrated	2002
10+ advanced applications testing and benefiting from ultra-high-performance test-bed	2002
Key: * All dates are fiscal years starting the prior October 1.

---

VBNS Network Backbone

illustration_link (30 Kbytes)

The Very High Bandwidth Network Service (VBNS) system already has 14,000 miles of OC-12 links.

---

The U.S.'s First GigaPOP

illustration_link (39 Kbytes)

North Carolina's Gigabit Point of Presence (GigaPOP) aggregates a variety of traffic onto a single high-bandwidth link.

---

Daniel P. Dern writes about the Internet and other technologies. You can contact him at the following: ddern@world.std.com or http://www.dern.com . Scott Mace ( smace@byte.com ) is a BYTE senior editor located in San Mateo, California.