The structure and makeup of the Internet has adapted as the needs of its community have changed. Today's Internet serves the largest and most diverse community of network users in the computing world. A brief chronology and summary of significant components are provided in this chapter to set the stage for understanding the challenges of interfacing the Internet and the steps to build scalable internetworks.
The Internet started as an experiment in the late 1960s by the Advanced Research Projects Agency (ARPA, now called DARPA) of the U.S. Department of Defense. DARPA experimented with the connection of computer networks by giving grants to multiple universities and private companies to get them involved in the research.
In December 1969, the experimental network went online with the connection of a four-node network connected via 56 Kbps circuits. This new technology proved to be highly reliable and led to the creation of two similar military networks, MILNET in the U.S. and MINET in Europe. Thousands of hosts and users subsequently connected their private networks (universities and government) to the ARPANET, thus creating the initial "ARPA Internet." ARPANET had an Acceptable Use Policy (AUP), which prohibited the use of the Internet for commercial use. ARPANET was decommissioned in 1989.
By 1985, the ARPANET was heavily used and congested. In response, the National Science Foundation (NSF) initiated phase one development of the NSFNET. The NSFNET was composed of multiple regional networks and peer networks (such as the NASA Science Network) connected to a major backbone that constituted the core of the overall NSFNET.
In its earliest form, in 1986, the NSFNET created a three-tiered network architecture. The architecture connected campuses and research organizations to regional networks, which in turn connected to a main backbone linking six nationally funded super-computer centers. The original links were 56 Kbps.
The links were upgraded in 1988 to faster T1 (1.544 Mbps) links as a result of the NSFNET 1987 competitive solicitation for a faster network service, awarded to Merit Network, Inc. and its partners MCI, IBM, and the state of Michigan. The NSFNET T1 backbone connected a total of 13 sites that included Merit, BARRNET, MIDnet, Westnet, NorthWestNet, SESQUINET, SURANet, NCAR (National Center of Atmospheric Research), and five NSF supercomputer centers.
In 1990, Merit, IBM, and MCI started a new organization known as Advanced Network and Services (ANS). Merit Network's Internet engineering group provided a policy routing database and routing consultation and management services for the NSFNET, whereas ANS operated the backbone routers and a Network Operation Center (NOC).
By 1991, data traffic had increased tremendously, which necessitated upgrading the NSFNET's backbone network service to T3 (45 Mbps) links. Figure 1-1 illustrates the original NSFNET with respect to the location of its core and regional backbones.
As late as the early 1990s, the NSFNET was still reserved for research and educational applications, and government agency backbones were reserved for mission-oriented purposes. But new pressures were being felt by these and other emerging networks. Different agencies needed to interconnect with one another. Commerical and general-purpose interests were clamoring for network access, and Internet service providers (ISPs) were emerging to accommodate those interests, defining an entirely new industry in the process. Networks in places other than the U.S. had developed, along with interest in international connections. As the various new and existing entities pursued their goals, the complexity of connections and infrastructure grew.
Government agency networks interconnected at Federal Internet eXchange (FIX) points on both the east and west coasts. Commercial network organizations had formed the Commercial Internet eXchange (CIX) association, which built an interconnect point on the west coast. At the same time, ISPs around the world, particularly in Europe and Asia, had developed substantial infrastructures and connectivity. To begin sorting out the growing complexity, Sprint was appointed by NSFNET to be the International Connections Manager (ICM)--to provide connectivity between the backbone services in the U.S. and European and Asian networks. NSFNET was decommissioned in April 1995.
The decommissioning of NSFNET had to be done in specific stages to ensure continuous connectivity to institutions and government agencies that used to be connected to the regional networks. Today's Internet structure is a move from a core network (NSFNET) to a more distributed architecture operated by commercial providers such as Sprint, MCI, BBN, and others connected via major network exchange points. Figure 1-2 illustrates the general form of the Internet today.
The contemporary Internet is a collection of providers that have connection points called POP (point of presence) over multiple regions. Its collection of POPs and the way its POPs are interconnected form a provider's network. Customers are connected to providers via the POPs. Customers of providers can be providers themselves. Providers that have POPs throughout the U.S. are called national providers.
Providers that cover specific regions (regional providers) connect themselves to other providers at one or multiple points. To enable customers of one provider to reach customers of another provider, Network Access Points (NAPs) are defined as interconnection points. The term ISP is usually used when referring to anyone who provides service, whether directly to end users or to other providers. The term NSP (network service provider) is usually restricted to providers who have NSF funding to manage the Network Access Points, such as Sprint, Ameritech, and MFS. The term NSP, however, is also used more loosely to refer to any provider that connects to all the NAPs.
NSFNET has supported data and research on networking needs since 1986. NSFNET also supported the goals of the High Performance Computing and Communications (HPCC) Program, which promoted leading-edge research and science programs. The National Research and Education Network (NREN) Program, which is a subdivision of the HPCC Program, called for Gigabit-per-second networking for research and education to be in place by the mid 1990s. All these needs, in addition to the April 1995 expiration deadline of the Cooperative Agreement for NSFNET Backbone Network Services, lead NSFNET to solicit for NSFNET services. This process is generally referred to as solicitation.
The first NSF solicitation, in 1987, lead to the NSFNET backbone upgrade to T3 links by the end of 1993. In 1992, NSF wanted to develop a follow-up solicitation that would accommodate and promote the role of commercial service providers and that would lay down the structure of a new and robust Internet model. At the same time, NSF would step back from the actual operation of the network and focus on research aspects and initiatives. The final NSF solicitation (NSF 93-52) was issued in May 1993.
The final solicitation included four separate projects for which proposals were invited:
The solicitation for this project was to invite proposals from companies to implement and manage a specific number of NAPs where the vBNS and other appropriate networks may interconnect. These NAPs should enable regional networks, network service providers, and the U.S. research and education community to connect and exchange traffic with one another. They also should provide for the interconnection of networks in an environment that is not subject to the NSF Acceptable Use Policy. (This policy was put in place to restrict the use of the Internet for research and education.) Thus, general usage, including commercial usage, can go through the NAPs also.
The NAP is defined as a high-speed network or switch to which a number of routers can be connected for the purpose of traffic exchange. NAPs must operate at speeds of at least 100 Mbps and must be able to be upgraded as required by demand and usage. The NAP could be as simple as an FDDI switch (100 Mbps) or an ATM switch (155 Mbps) passing traffic from one provider to the other.
The concept of the NAP is built on the FIX (Federal Internet eXchange) and the CIX (Commercial Internet eXchange), which are built around FDDI rings with attached Internet networks operating at speeds of up to 45 Mbps.
The traffic on the NAP should not be restricted to that which is in support of research and education. Networks connected to the NAP are permitted to exchange traffic without violating the use policies of any other networks interconnected to the NAP.
There are four NSF-awarded NAPs:
The NSFNET backbone service was physically connected to the Sprint NAP on September 13, 1994. It was physically connected to the PacBell NAP and Ameritech NAP in mid-October 1994 and early January 1995, respectively. The NSFNET backbone service was upgraded to the collocated FDDI offered by MFS on March 22, 1995.
Additional NAPs are being created around the world as providers keep finding the need to interconnect.
Networks attaching to NAPs must operate at speeds commensurate with the speeds of attached networks (1.5 Mbps or greater) and must be upgradable as required by demand, usage, and program goals. NAPs must be able to switch both IP and CLNP (ConnectionLess Networking Protocol). The requirements to switch CLNP packets and to implement IDRP-based (InterDomain Routing Protocol, ISO OSI Exterior Gateway Protocol) procedures may be waived depending on the overall level of service and the U.S. government's desire to foster the use of ISO OSI protocols.
A NAP manager should be appointed to each NAP with duties that include the following:
The current physical configuration of today's NAPs is a mixture of FDDI/ATM switches with different access methods, ranging from DS3 for dedicated and FR/ATM/SMDS for switched. Figure 1-3 shows a possible configuration, based on some contemporary NAPs. The routers could be managed either by the NSP or the NAP manager. Different configurations, fees, and policies are set by the NAP manager. Connections from different LATA (Local Access and Transport Area) are provided by Inter eXchange Carriers (IXC).
Due to the decommissioning of the NSFNET backbone, federal regional networks faced the problem of transitioning to the new infrastructure where they have to be connected to new NSPs. The Federal Networking Council (FNC) Engineering and Planning Group (FEPG) was responsible for making a recommendation on how to transition to the new NAP-NSP operational environment with minimal disruption to users, specifically in federal agency communications with the U.S. academic and research communities.
Existing Federal Internet eXchanges (FIX West and FIX East) were to be connected to the major NSPs (MCInet, Sprintlink, ANS). The FIX West backbone formerly was maintained at NASA Ames. Now it is connected to the major NSPs, and route servers were installed to peer with the federal agencies. The FIX East backbone formerly was maintained at SURA (College Park, MD). Now it is connected to the major NSPs and is also bridged to the MAE-East facility (Tyson's Corner, VA) of MFS.
The CIX (pronounced Kix) is a nonprofit trade association of Public Data Internetwork Service Providers. The association promotes and encourages the development of the public data communications internetworking services industry in both national and international markets. The CIX provides a neutral forum to exchange ideas, information, and experimental projects among suppliers of internetworking services. Some benefits CIX provides its members include:
With increasing ISP connectivity to NAPs, the CIX becomes essential in the coordination of legislative issues between members. In fact, the role of the CIX for physical connectivity is not as important as its role in coordination between parties. With the existence of a number of other high bandwidth connection points such as the NAPs, the CIX plays a minor role in the connectivity game. ISPs who still rely on the CIX as their only physical connection to the Internet are still way behind.
On July 13, 1994, the CIX board voted to block traffic from ISPs who are not CIX members. CIX membership costs approximately $7,500 annually.
Although NAP connectivity is primarily something ISPs have to worry about, the level of redundancy and diversity of NAP connections affects traffic patterns and trajectories in the whole Internet. As such, the delays or speed of access caused by ISPs' interconnectivity affect the performance of everyone's Internet access. As you will see in the rest of this book, speed of access to the NAPs and the distance an ISP or a customer is from the NAP affects routing behaviors and traffic trajectories.
Another project for which NSF solicited services is the Route Arbiter (RA) project, which is charged with providing equitable treatment of the various network service providers with regard to routing administration. The RA will provide for a common database of route information to promote stability and manageability of networks.
Multiple providers connecting to the NAP have created a scalability issue because each provider will have to peer with all other providers to exchange routing and policy information. The RA project was developed to reduce the full peering mesh between all providers. Instead of peering among each other, providers will peer with a central system called a route server. The route server will maintain a database of all information needed for providers to set their routing policies. Figure 1-4 shows the physical connectivity and logical peering between a route server and various service providers.
The following are the major tasks of the RA per the NSFNET proposal:
Today, the RA project is a joint effort of Merit Network, Inc., the University of Southern California Information Sciences Institute (ISI), Cisco Systems, as a subcontractor to ISI, and the University of Michigan ROC, as a subcontractor to Merit.
The RA service is comprised of four projects:
As you have already seen, the main parts of the Route Arbiter concept are the route server and the RADB. The practical and administrative goals of the RADB apply mainly to service providers connecting to the NAP. Configuring the correct information in the RADB is essential in setting the required routing policies, as explained in Appendix A, "RIPE-181." As a customer of a provider, you may never have to configure such language. What is important, though, is not the language itself but rather understanding the reasoning behind the policies being set. As you will see in this book, policies are the basis of routing behaviors and architectures.
On the other hand, the concept of a route server and peering with centralized routers is not restricted to providers and NAPs, and could be implemented in any architecture that needs it. As part of the implementation section of this book, the route server concept will come up as a means of creating a one-to-many relationship between peers.
The very high-speed Backbone Network Service (vBNS) project was created to provide a specialized backbone service for the high-performance computing users of the major government-supported SuperComputer Centers (SCCs) and for the research community. The vBNS will continue the tradition that NSFNET has provided in this field. The vBNS will be connected to the NSFNET- specified NAPs. On April 24, 1995, MCI and NSF announced the launch of the vBNS.
MCI duties include the following:
The five-year, $50-million agreement between MCI and NSFNET will tie together NSF's five major high performance communication centers:
The vBNS has been called the R & D lab for the 21st century. The use of advanced switching and fiber optic transmission technologies, Asynchronous Transfer Mode (ATM), and Synchronous Optical Netwok (SONET) will enable very high-speed, high-capacity voice and video signals to be integrated.
The NSF is already in the process of authorizing use of the vBNS for "meritorious" high-bandwidth applications, such as using super-computer modeling at NCAR to understand how and where icing occurs on aircraft. Other applications at NCSA consist of building computational models to simulate the workings of biological membranes and how cholesterol inserts into membranes.
The vBNS will be accessible to select application sites through four NAPs in New York, San Francisco, Chicago, and Washington, D.C. Figure 1-5 shows the geographical relationships between the centers and NAPs. The vBNS is mainly composed of OC3 /T3 (OC12 is in the process of being deployed) links connected via high-end systems, such as Cisco routers and Cisco ATM switches.
The vBNS is a specialized network that emerged due to continuing needs for high-speed connections between members of the research and development community, one of the main charters of the NSFNET. Although the vBNS does not have any bearing on global routing behavior, the preceding brief overview is meant to give the reader background on how NSFNET covered all its bases before being decommissioned in 1995.
As part of the NSFNET solicitation for transitioning to the new Internet architecture, NSF requested that regional networks (also called mid-level networks) start transitioning their connections from the NSFNET backbones to other providers.
Regional networks have been a part of NSFNET since its creation and have played a major role in the network connectivity of the research and education community. Regional network providers (RNPs) connect a broad base of client/member organizations (such as universities), providing them with multiple networking services and with Inter Regional Connectivity (IRC).
The anticipated duties of the Regional network providers per the NSF 93-52 program solicitation follow:
In the process of moving the regionals from the NSFNET to new ISP connections, NSF suggested that the regional networks be connected either directly to the NAPs or to providers connected to the NAPs. During the transition, NSF supported, for one year, connection fees that would decrease and eventually cease (after the first term of the NAP Manager/RA Cooperative Agreement, which shall be no more than four years.)
Table 1-1 lists some of the old NSFNET regional providers and their new respective providers under the current Internet environment. As you can see, most of the regional providers have shifted to either MCInet or Sprintlink. Moving the regional providers to the new Internet architecture in time for the April 1995 deadline was one of the major milestones that NSFNET had to achieve.
|Old Regional Network||New Internet Provider|
|Cornell Theory Ctr.||MCInet|
In addition to the four main projects relating to architectural aspects of the new Internet, NSF recognized that information services would be a critical component in the even more widespread, freewheeling network. As a result, a solicitation for one or more Network Information Services (NIS) managers for the NSFNET was proposed. This solicitation invites proposals for the following:
At the time of the solicitation, the domestic, non-military portion of the Internet included the NSFNET and other federally sponsored networks such as the NASA Science Internet (NSI) and Energy Sciences Network (ESnet). All these networks, as well as some other networks of the Internet, were related to the National Research and Education Network (NREN), which was defined in the President's fiscal 1992 budget. The NSF solicitation for Database Services, Information Services, and Registration services were needed to help the evolution of the NSFNET and the development of the NREN.
At the time of the proposal, certain network information services were being offered by a variety of providers; some of these services included the following:
Under the new solicitation, NIS managers should provide services to end users and to campus and mid-level network service providers, and should coordinate with mid-level and other network organizations, such as with Merit, Inc.
In response to NSF's solicitation for NIS managers, in January 1993 the InterNIC  was established as a collaborative project among AT&T, General Atomics, and Network Solutions, Inc. It was to be supported by three five-year cooperative agreements with the NSF. During the second year performance review, funding by the NSF to General Atomics stopped. AT&T was awarded the Database and Directory Services, and Network Solutions was awarded the Registration Services and the NIC Support Services.
The NIS manager will act in accordance to RFC 1174, which states the following:
The Internet System has employed a central Internet Assigned Numbers Authority (IANA) for the allocation and assignment of various numeric identifiers needed for the operation of the Internet. The IANA function is performed by the University of Southern California's Information Sciences Institute. The IANA has the discretionary authority to delegate portions of this responsibility and, with respect to numeric network and autonomous system identifiers, has lodged this responsibility with an Internet Registry (IR).
The NIS manager will either become the IR or a delegate registry authorized by the IR. The Internet registration services to be provided will include:
Today, NSI is providing assistance in registering networks, domains, AS numbers, and other entities to the Internet community via telephone, electronic mail, and U.S. postal mail. RS will work closely with domain administrators, network coordinators, ISPs, and other various users to register Internet domains, Autonomous System numbers, and networks.
The RS will provide databases and information servers such as WHOIS registry for domains, networks, AS numbers, and their associated Point Of Contacts (POCs). The RS also offers Gopher and Wide Area Information Server (WAIS) interfaces for retrieving information.
The documents distributed by the InterNIC registration services include templates, network information, and policies to request network numbers and register domain name servers.
Some of the templates include:
The implementation of this service should utilize distributed database and other advanced technologies. The NIS manager could coordinate this role with respect to other organizations that have created and maintained relevant directories and databases. AT&T is providing the following services under the NSF agreement.
The original solicitation for "Information Services" was granted to General Atomics in 1993 and taken away in February 1995. At that time, Network Solutions, Inc. took over the proposal, and it was renamed NIC Support Services.
The goal of this service is to provide a forum for the research and education community, Network Information Centers (NICs) staff, and the academic Internet community, within which the responsibilities, duties, and functions of the InterNIC may be defined. As of now, this service is divided into two components:
Other Internet Registries (IR) were created outside the U.S.; these registries perform functions similar to those performed by the InterNIC in the U.S.
Created in 1989, RIPE is a collaborative organization that consists of European Internet service providers. It aims to provide the necessary administration and coordination to enable the operation of the European Internet.
APNIC  is the IR for the Asia Pacific rim. It provides the IP registration and domain name services for that region. Created in 1993, APNIC started as a 10-month pilot project with the goal of providing Internet Registry functions and Routing Register functions (the RR function has not materialized to date). The pilot proved to be successful, and the APNIC is now in full operation serving as an IR.
Other Internet Registers are listed on the InternetNIC home page.
With the creation of a new breed of ISPs that want to interconnect with one another, offering the required connectivity while maintaining flexibility and control has become more challenging. Each provider has a set of rules, or policies, that describe what to accept and what to advertise to all other neighboring providers. Example policies include determining route filtering from a particular ISP and choosing the preferred path to a specific destination. The potential for the various policies from interconnected providers to conflict with and contradict one another is enormous.
To address these challenges, a neutral Routing Registry (RR) for each global domain had to be created. Each RR will maintain a database of routing policies created and updated by each service provider. The collection of these different databases is known as the Internetworking Routing Registries (IRR).
The role of the RR is not to determine policies, but rather to act as a repository for routing information and to perform consistency checking on the registered information with the other RRs. This should provide a globally consistent view of all policies used by providers all over the world.
Autonomous Systems (ASs) use exterior gateway protocols such as BGP to work with one another. In complex environments, there should be a formal way of describing and communicating policies between different ASs. Maintaining a huge database containing all registered policies for the whole world is cumbersome and difficult to maintain. This is why a more distributed approach was created. Each RR will maintain its own database and will have to coordinate extensively to achieve consistency between the different databases. Some of the different IRR databases in existence today are:
Each of the preceding registries serves a limited number of customers except for the Routing Arbiter Database (RADB), which handles all requests not serviced by other registries. As mentioned earlier, the RADB is part of the Routing Arbiter (RA) project, which is a collaboration between Merit and ISI with subcontracts to Cisco Systems and the University of Michigan ROC.
The decommissioning of the NSFNET in April 1995 marked the beginning of a new era. The Internet today is a playground for hundreds and thousands of providers competing for market share. For many businesses and organizations, connecting their networks to the global Internet is no longer a luxury but a requirement for staying competitive.
The structure of the contemporary Internet has implications for service providers and their customers in terms of speed of access, reliability, and cost of use. Some of the questions organizations that want to connect to the Internet should ask are: Are providers--whether established or relatively new to the business--well-versed with routing behaviors and architectures? For that matter, how much do customers of providers need to know and do with respect to routing architecture? Do we really know what constitutes a stable network? Is the bandwidth of our access line all we need to worry about to have the "fastest" Internet connection? The next chapter is intended to help ISPs and their customers evaluate these questions in a basic way. Later chapters get into details of routing architecture.
Interdomain routing is fairly new to everybody and is evolving every day. The rest of this book builds upon this chapter's overview of the structure of the Internet in explaining and demonstrating current routing practices.
Q- Are there other NAPs besides the four NSF-awarded NAPs?
A- Yes. As connectivity needs keep growing, more NAPs are being created. Many exchange points are spread over North America, Europe, Asia/Pacific, South America, Africa, and the Middle East.
Q- If I am a customer of a provider, do I have to connect to a NAP?
A- No. NAPs are mainly for interconnection between providers. If you are a customer of a provider, your connection will be to the provider only. But how your provider is connected to one or more NAP can affect the quality of your connection.
Q- Is the function of the route server at the NAP to switch traffic between providers?
A- No. The route server keeps a database of routing policies used by providers. Providers use the NAP physical media to exchange traffic directly between one another.
Q- Do all providers that connect to a NAP have to peer with the route server?
A- Although this is the recommended procedure, in some situations, major providers end up peering directly with each other, while smaller providers are required to peer with the route server.
Q- What is the difference between IRs and IRRs?
A- Internet Registries (IRs) such as the InterNIC are responsible for registration services such as IP address assignment. Internet Routing Registries are responsible for maintaining databases of routing policies for service providers.
Q- How are database services different from the Route Arbiter Database?
A- Database services are part of the Network Information Services. These databases include communication documents such as RFCs. The RADB is a database of routing policies.
 RFC 1786; Representation of IP Routing Policies in a Routing Registry (Ripe-81++)