Internet Architecture: Past and Present

Enterprise Networks & Servers,  Jan 2004  by Breit, Lou

• E-mail
• Print
• Link

The Internet is sometime referred to as a "backbone network," but this description is flawed in that the Internet is actually many individual networks connected to form a mesh. These packet switched networks are interconnected with routers, all of which have multiple paths in case of a link failure.

This design originated in 1991 with the research network known as NSFNET, which was funded by the U.S. National Science Foundation. In this hierarchical model, local service providers connect to regional service providers, which then connect to national and global service providers. The backbone networks between these providers are sometimes bypassed, allowing regional service networks to directly connect to other regional networks, depending on traffic volume and agreements.

All of the networks which make up the Internet are managed and administered by large independent service providers. In addition, there are NSPs (network service providers), ISPs (Internet service providers), and exchange points between all parties. National and global NSPs build the network infrastructure, and then sell bandwidth to regional NSPs. These regional NSPs sell bandwidth to local ISPs, who in turn provide services to end users.

The NFSNET connected sites included universities and research centers, originally connected by 56Kbps links but later upgraded to T1 and T3 connections. The ISP business model was developed by individuals who purchased and resold bandwidth and routing services from NSPs. The ISP then installed dial-up facilities with modems and concentrators, and billed end users for service. While the NFSNET functioned well as a separate entity, there was a need to connect to other government backbones (military, energy, and science networks, etc.). Two interconnection points were created called FIXes (Federal Internet Exchanges), FIX-West and FIX-East. Through these points, different networks agreed to "peer" or exchange routing advertisements and traffic. At this point, a mesh network was created.

By 1993, a decision was made to privatize NSFNET. NAPs or network access points were created, which act as Internet exchanges with open access policies to support all types of traffic. NAPs act in a similar fashion to an airport, in which multiple airlines are leased space and facilities. NSPs are the "airlines" in that they use the switching facilities within a NAP to exchange traffic with other networks.

Originally, NAPs used FDDI and switched Ethernet, but eventually migrated to ATM and Gigabit Ethernet switches to handle increasing traffic volume. These core ATM switches are surrounded by high speed routers, which exchange traffic across ATM PVCs (permanent virtual circuits). Typically, these PVCs are fully meshed since they offer a connection to every other NSP router installed at the NAP; however, an NSP may remove a PVC to block traffic from another NSP. This is sometimes done by a larger NSP to avoid peering with a smaller NSP since there is no equal exchange of traffic, though this practice is discouraged.

NSPs install their own router equipment within the switching facilities of the NAP. In this manner, traffic from ISPs travels over the NSP routers through the NAP and then on to other NSPs within the Internet. It is important to note that the NAP is limited to providing the switching facilities, and has no role in establishing agreements between NSPs to exchange traffic. These "peering" agreements are negotiated between individual NSPs, and define how traffic will be exchanged and routed.

In today's Internet, these exchange points are not the only location where traffic is exchanged between NSPs. In many instances, private peering arrangements (outside of a NAP) are established between NSPs through the use of private connections. These private connections are direct physical circuits intended to bypass the sometimes congested switching facilities within the NAP. One way to establish a private connection is to simply run a cable between routers located at the NAP, or to run a leased line between NSP facilities.

The numerous individually managed networks within the Internet are classified as autonomous systems (AS). These autonomous systems act as both a management and routing domain, are typically operated by an NSP or ISP, and are assigned a unique number by the Internet authorities. Interior routing protocols such as OSPF (Open Shortest Path First) are used to maintain internal routing tables, while exterior routing protocols are used to exchange routing information between distinct autonomous systems.

The exterior routing protocol for the Internet is BGP, or Border Gateway Protocol. BGP operates in border routers which connect AS domains together, sharing route information gathered from internal networks. The routes are advertised as an aggregation of addresses, in a similar manner to how a zip code defines a location or block of addresses. Through the use of route aggregation, IP address space can be used with greater efficiency. ISPs can combine blocks of addresses and advertise all of them with a single network address, while allocating the addresses differently to lower level ISPs.