Internet2 Architecture and GigaPoPs

In this section, I want to give you a basic conceptual overview of the Internet2 architecture, and then describe how that architecture addresses some of our requirements.

I'm also going to briefly describe the types of applications that justify Internet2, and how the Internet2 architecture will support the demands of those applications. We'll first talk about the technical requirements for a new generation of applications that the current Internet simply can't support. Then we'll take a look at the emerging high-performance networks that can support these applications. And finally, we'll talk about Internet2 engineering and the gigaPoP architecture.

In a nutshell, the impetus for Internet2 is both current and evolving technical requirements. First of all, the next Internet needs to be able to provide more bandwidth. We need more bandwidth not only to support exotic, high-end research applications, such as the use of virtual reality in medical operations, but we also need it to support the real-time audio and video necessary for distributed learning.

Second, we must be able to support differentiated Quality of Service. We'd like to be able to provide high bandwidth and low delay and jitter to every user, but that's simply not financially possible. We therefore need to establish a sort of technological triage. For example, a student working on a class project does not need the same Quality of Service that we must be able to guarantee a surgeon in the middle of a heart transplant operation.

Third, we must be able to support multicasting. Multicasting provides the ability to broadcast from one source on a network to many receivers in real-time. This capability is obviously essential for distributed learning.

Fourth, we need to be able to support many more nodes than on the current Internet. When the present telephone system was originally designed, they looked at the world's population and then predicted what the population would be in the twenty-second century. What they didn't anticipate was the fact that everyone now has a pager, a cellular telephone, an answering machine and perhaps several other devices. As a result, we're running out of telephone numbers. The same thing is basically happening to IP, or Internet Protocol -- the language or protocol underlying the current Internet. There simply aren't enough addresses to go around.

Finally, we need to provide both increased security and reliable authentication. Security establishes which users can have access to a system or an application, and authentication is the ability to make certain that a user is who they say they are.

All of these are components of the networking technologies that Internet2 should provide.

I'd now like to turn to the application technologies we'd like to improve. These can be divided roughly into five categories: database access, real-time audio and video, real-time and delayed collaboration, distributed computing, and tele-immersion.

Traditional databases involve accessing a few thousand or a few million or perhaps even a few billion different items. New applications are emerging, however, which require databases that are a million times -- or a million million times -- larger than those in use today. One example is satellite data. We anticipate that in a few years, orbiting satellites will beam down more information in a single day than in all of recorded history through today. Similarly, laboratories, like the Thomas Jefferson National Accelerator Facility in Virginia, generate millions upon millions of bytes of scientific information every day. We need to be able to support databases capable of processing this volume of data.

We also need to support real-time audio and video. Real-time and delayed collaboration will allow folks who are physically separated by hundreds or even thousands of miles to work together towards a common objective. For example, climatologists in widely separated laboratories might work together to improve our understanding of global warming. Or, a group made up of architects and architectural students located at sites across the country might collaborate on the layout of architectural spaces. Or faculty and students in remote universities might gather across the network for an advanced graduate seminar on metaphysics.

The ability to link distributed computing resources also offers some exciting opportunities. One demonstration at the recent Internet2 meeting showed how distributed computers could analyze data from remote diagnostic machines. Another project featured a research environment in which scientists at the University of California at San Diego and at Cornell University jointly analyzed biological specimens through a high-powered electron microscope linked to a supercomputer.

Finally, tele-immersion creates 3D audio and video environments that surround the user. These high speed, real-time audio and video environments enable "virtual reality" applications that can range from simple classroom instruction to the practice of surgical techniques.

We'd also like for the next Internet to do a better job of supporting multimedia. This graphic illustrates the bandwidth and storage requirements of some common networked media. For example, text read at 120 words per minute is equal to a digital transmission rate of about 96 bits per second. That data rate can be supported on the Plain Old Telephone System, or POTS, and the storage medium is typically a floppy disk. High quality uncompressed audio signals, on the other hand, require significantly more bandwidth, since we're talking about transmissions of more than a million bits per second. A T1 line or a high-performance LAN is required for this type of transmission, and the typical storage medium is a CD.

Let's look at the bandwidth and storage requirements of some different types of video signals. Relatively low quality video signals, such as H.263, can be supported with the Plain Old Telephone System and can be stored on a few floppy disks or a Zip drive. High quality video signals like MPEG2, however, require a T3 or high-performance LAN connection and are typically distributed on a Digital Video Disk.

The point of looking at all these numbers is to demonstrate that if we want to support high quality audio and video, we need a higher-performance network.

Currently there are three high-performance networks in operation in this country, and two more under construction. The vBNS, or very-high-performance Backbone Network Service, is funded by the National Science Foundation to support research. The other two networks currently in operation are the NASA Research and Education Network, and the Energy Sciences Network. These two networks are funded by NASA and the Department of Energy respectively to support each agency's research mission.

The two networks under development are the NGI and Internet2. The NGI, or Next Generation Internet, is a governmental initiative designed to support the development of new network technologies and applications, including Internet2. The Internet2 project was initiated by the nation's research universities and is focused on providing high-performance networking for research and education.

Let's now turn to the actual architecture of Internet2. In order to understand the architectural concept behind Internet2, you first need to understand the gigaPoP concept which has been adopted by the Internet2 project. The gigaPoP serves as a regional aggregation point for traffic from research and education partners, and from individual campuses. As you can see in this graphic, a gigaPoP can be connected to other gigaPoPs both through existing high-performance networks, such as the vBNS, as well as through traditional Internet Service Providers, or ISPs. GigaPoPs are different from existing exchange points, however, because gigaPoPs also can connect to each other through new and emerging networks.

What's important to note about this architecture is that gigaPoPs do not replace the current Internet. Instead they allow the end-user a broader range of Wide Area Network providers. High-performance network connections are currently very expensive, but the gigaPoP architecture allows multiple institutions to share the cost.

From an engineering perspective, the infrastructure of Internet2 can be broken down into a number of discrete elements. The first, of course, is the advanced applications which justify the construction of Internet2.

The second includes campus and related facilities. For example, my campus is host to the National Crash Analysis Center, which uses computer simulations to crash automotive vehicles and then makes the results available throughout the world.

The third element is made up of individual campus networks, and if the economics of Internet2 follow those of Internet 1, campus networks will represent more than 90% of the cost of this initiative.

The fourth is the connection between campus networks and a regional gigaPoP. The gigaPoP itself is far more sophisticated and complex than our current Internet exchange points. It not only must provide differentiated Quality of Service, it also must provide an interconnect between traditional Internet services and the new Internet2 services.

The gigaPoPs themselves then must be interconnected.

And finally, we must develop the software protocols that will allow all of these different elements to work together smoothly.

In order to meet these requirements, six general principles were established early in the design of Internet2. The first was that whenever possible, facilities and equipment would be bought rather than built, to help speed deployment. The second was that Internet2 architecture should be open rather than closed. The current Internet is an example of the advantages of an open architecture that encourages competition. Third, reliability would be provided through redundancy rather than reliance upon near-perfect performance. Fourth, the engineering design would emphasize simplicity rather than complexity. Fifth, the Internet2 project would be a production network, not an experimental network. In other words, the technology will be a few steps ahead of what's available in commercial networks and may experience some hiccups from time-to-time, but it will be stable enough to support the real work of educators and researchers. Finally, Internet2 is designed to provide services to end-users at universities, not to provide an interconnect between commercial providers.

The original Internet2 engineering timeline was quite aggressive. One goal is that by the turn of the century, most research universities and gigaPoPs will be connected and campus accessibility will be widespread. Furthermore, interconnectivity between gigaPoPs will have expanded beyond the vBNS, and limited Quality of Service will be available.

The requirements for Internet2-capable campus networks remain unchanged from the Monterey Futures Group workshop of 1995. They include unshared 10 Mbps connections to each desktop, a 500 Mbps or faster backbone, a full-function Network Operations Center (including instrumentation, reporting, monitoring, and trouble-ticket sharing) and limited Quality of Service.

We anticipate that over the next few years the bandwidth of the campus to gigaPoP connection will rise from DS3, or 45 Mbps, to OC3, or 155 Mbps. The connection between individual campuses and gigaPoPs is based on more than just technology, however. Links between newly-developed Network Operations Centers will be enhanced as protocols are developed and trust is established. The architecture of Internet2 places substantial responsibility on individual campuses to not only support their own end-users, but also to participate collaboratively in the management of the nationwide network. The alternative to the collaborative management model, which was actively considered during the initial design phase, was to create a national entity that would centrally manage all of Internet2 above the campus level.

The routing responsibilities of the gigaPoP present both a technical and an administrative challenge. For example, in addition to ATM connections to individual Internet2 campuses and commercial network providers, a gigaPoP must also be able to connect to the vBNS and other agency networks, and to regional or state networks. In addition, a gigaPoP must be able to support the IP routing elements necessary for connecting both to campuses with non-ATM connections and to urban area networks and independent Internet service providers. To draw a simple analogy, a gigaPoP must be not only multi-lingual but also multi-cultural to host a party enjoyed by everyone.

Once gigaPoPs are operational, we anticipate that the typical gigaPoP will have five to twelve institutional members. It will provide its members with the facilities and application-specific functionality they need. It will provide connectivity to specific high-performance networks, such as to a local library consortium.

GigaPoPs also will be able to interconnect routed and switched networks. They will serve as domain boundaries and will assist in the sorting of information transfer, including management of Quality of Service.

Finally, gigaPoPs will play a key role as the common element between campuses with different standards and procedures. This is a critical role, as we create a virtual Network Operations Center capable of supporting the management functions of instrumentation, reporting, monitoring and trouble-ticketing.

Connections directly from one gigaPoP to another will be fast -- very fast -- and they will provide transit only between Internet2 members. These connections soon will be operating at speeds in excess of 500 Mbps, which will make them seem like virtual direct connections to the end-user. We anticipate that these inter-gigaPoP connections will follow multiple paths and will make use of multiple vendors.

The requirements for inter-gigaPoP connectivity include reliability, bandwidth capacity, selectable Quality of Service, and adequate management tools.

A high performance network is a marvel of many components. This graphic shows the interdependent relationship between network engineering and network applications. New applications provide the motivation for advancing engineering, and in turn, new engineering capabilities enable the development of new applications. For example, Quality of Service enables real-time audio and video, which in turn motivates more and better Quality of Service, and so on.

The success of the Internet2 project depends on meeting both advanced engineering and networking challenges. The Internet2 project also depends upon building an organization based on effective collaboration between network operations centers and the people and organizations who support them. Other sections in this seminar focus on the applications that will be made possible through meeting these joint technical and human challenges.

About CREN

© CREN, 1999

Contact Us