At the highest level, today's Internet consists of multiple national and regional Internet Service Providers (ISP) and interconnection points where the ISPs meet and exchange traffic. This infrastructure is similar to that of the old NSFNET, which consisted of a three-tier structure:
On the NSFNET, regional networks would aggregate their traffic and "hand it off" to the NSFNET backbone. The regional networks comprised multiple local business and campus networks. Although there were many regional and local networks, there was only one backbone network.
As mentioned in Section 2, the NSFNET has been decommissioned. In its place are multiple nationwide networks, which are similar to the NSFNET backbone network. Regional networks still aggregate their traffic and hand it off to the nationwide backbone network to which they are connected. Interexchange points (IXP) are located around the country where traffic is exchanged between national and regional ISPs. Peering agreements are used between the ISPs connected at an IXP to determine how traffic is routed. These service providers and interexchange centers are the main components of the U.S. Internet. This section will describe different elements of the Internet architecture and the different routing protocols used on today's Internet.
3. 1 INTERNET SERVICE PROVIDERS
ISPs are classified according to their network and customer base. The network classification refers to whether or not the ISP owns or leases its network. An ISP that does not own or lease its network is referred to as a reseller. The customer base classification refers to an ISP's type of customers, national or regional. A particular ISP may have national and regional customers, but generally it has more of one type than another. There are three types of ISPs:
The following sections provide further detail for each type of ISP.
3. 1 .1 National Service Providers
The first category of ISPs is NSP, which provide national backbone service. This type of service provider owns or leases its own backbone network and has a nationwide customer base. Additionally, NSPs are generally connected to all the major IXPs and have peering agreements with other major NSPs at these exchange points. Traffic originating with a customer on an NSP that is destined for a customer on another NSP is transferred from the originating NSP's network to the terminating NSP's network at an IXP. The NSPs network infrastructure consists of routers (network layer) and switches (data link layer) that are owned by the NSP. The following are examples of NSPs:
Of the NSPs, MCI and Sprint are the only two that own their entire network. Other NSPs may own small parts of their networks, but most of their networks consist of circuits leased from the PN providers. Most of these circuits are leased from the large Interexchange Carriers (IEC)1. However, some circuits are also leased from the Local Exchange Carriers (LEC) (e.g., Bell Atlantic), Competitive Access Providers (CAP) (e.g., Metropolitan Fiber Systems), and smaller IECs (e.g., LDDS).
Exhibit 3-1 depicts a representative backbone network for UUNet, one of the NSPs mentioned above. As shown in the exhibit, UUNet (like most NSPs) has redundant connectivity between each switching node on its backbone network.
NSPs rarely sell directly to small consumers (e.g., small businesses and residential customers) because of the added "customer handholding" required by smaller, less experienced users. Instead, NSPs sell their services to large businesses and resellers. Resellers in-turn resell Internet service to small business and residential customers. It is important to note that not all NSPs resell their networks, e.g. PSINet.
The architecture of an NSP's network may be separated into access and transport. Access refers to the customer's connection to the NSP, whereas transport refers to the backbone of the NSP's network. Customers connect to NSPs via leased and dial-up lines. Typical leased lines are 56-Kbps or T1 and usually terminate at an NSP's point of presence (POP).
For dial-up customers, the NSP usually has digital and/or analog modem banks terminating from its POP into the local central office using T1s. Because NSPs have national presence and reach, once a customer's traffic reaches an NSP's POP, it has essentially reached the Internet. The typical backbone of an NSP comprises routers and switches connected by T1, T3, or even OC-level circuits. These circuits may be leased from one or more IEC. One NSP, PSINet, leases backbone circuits from five different IECs.
The NSP market has not escaped the notice of existing PN providers anxious to get involved in the growth of the Internet. In the short term, PN providers have chosen to partner with NSP providers for Internet backbone transport instead of developing NSP expertise in-house. For example, GTE recently announced a partnership with UUNet to provide Internet access under the GTE name to customers in 46 U.S. states2. Cross PN-NSP service agreements also exist between Pacific Bell and America On-Line (which owns ANS), and AT&T and BBN. The recently announced merger between UUNet and MFS may be a harbinger of future mergers between NSPs and PN providers. PN providers own the data links necessary to run an NSP and have the marketing savvy to sell Internet service to business and residential customers. NSPs, on the other hand, have the in-house technical expertise to manage the switches, routers, and interconnection arrangements necessary to make the NSP backbone work.
Other future developments in the NSP market will include service differentiation to target selected customer markets. For example, MCI and BBN have announced services that provide a higher quality of service to business customers who subscribe to their NSP. BBN provides priority treatment to business customers through Internet Protocol version 6 (IPv6) priority service protocols. MCI provides a separate network for its business subscribers' Internet traffic. This separate network includes locally hosted mirror sites from popular Web sites on other NSP networks and in the future will include IPv6 priority treatment.
3. 1 .2 Regional Service Providers
The second category of ISPs are the RSPs. These service providers are similar to the NSPs in that they own or lease their backbone network but are much smaller in scale. Their networks encompass a single region and usually have a regional customer base. RSPs have peering agreements with NSPs to transfer traffic over the Internet.
RSPs either connect directly to the NSP or connect to an IXP where they transfer traffic to the NSP network. NorthWestNet is an example of an RSP that connects directly to an NSP. NorthWestNet, which provides service to customers in Washington, Oregon, and Idaho, has direct connections to both MCI and Sprint's NSP networks. Erols is an example of a network with a direct connection to an IXP. Erols, which provides service to customers in the metropolitan Washington, DC area, is connected to the Metropolitan Area Ethernet-East (MAE-EAST) IXP where it can transfer traffic to most of the larger NSPs and several smaller RSPs.
RSP service is an attractive option for residential and small business customers. Because of the small customer base, RSPs can offer more "hands-on" assistance in the form of customer training and help desk operators trained to assist less knowledgeable users.
Like NSP networks, the RSP's network architecture may be separated into access and transport portions, though with different meanings. In the RSP scenario, access refers not only to the customer connecting to the RSP but also the RSP connecting, if at all, to the Internet. Transport refers to the backbone of the RSP's network. As in the NSP scenario, customers connect to RSPs via leased and dial-up lines. Typical leased lines are 56-Kbps or T1 and usually terminate at an RSP's POP. For dial-up customers, the RSP usually has digital and/or analog modem banks terminating from its POP into the local central office using T1s.
An RSP's backbone is typically restricted to a region, as opposed to NSPs who have a national presence and whose backbone spans the entire United States. Transport on an RSP's network, or backbone, comprises T1 and T3 circuits that connect their POPs and customers in a particular region. These circuits are leased from LECs, CAPs, and IECs. As noted above, RSPs' customers are primarily small business and residential subscribers. In the coming years, new companies will enter this market. Most notable are the Internet service offerings from the IECs and the Region Bell Operating Companies (RBOC). This increased competition may cause some consolidation of the RSP market when smaller RSPs go out of business or are bought out by larger firms. The remaining RSPs will survive by targeting market niches, such as high volume residential users or businesses new to the Internet.
The exhibits below show two example RSP network backbones. Exhibit 3-2 shows NorthWestNet's backbone and Exhibit 3-3 shows CERFnet's backbone. Note that NorthWestNet has redundant connections to Sprint and MCI to transfer traffic, whereas CERFnet connects directly to NAPs to share traffic.
3. 1 .3 Resellers
Resellers are another member of the Internet provider family. Resellers purchase service from NSPs or RSPs and resell this service to small business and residential customers. Resellers are differentiated from RSP because resellers do not own or lease a network infrastructure. Instead resellers typically operate out of a single site with a modem bank for customer access and a T1 connection to transfer traffic to the NSP/RSP network.
There are approximately 1,400 Internet resellers in the United States, most of which base their business on monthly subscriptions to Internet service. As the Internet market matures, monthly Internet service is becoming a commodity. This trend has been furthered by the entry of the RBOCs and IECs into the residential Internet service market. Typically, unlimited access is provided on a monthly basis for a flat-rate fee or a combination of flat-rate and usage-based pricing.
Source: NorthWestNet, Inc.
Because Internet service has significant economies of scale, the market favors the larger providers who can spread their fixed costs over a larger customer base. Because of this, many experts predict that the number of Internet resellers will decrease dramatically in the next few years. The Yankee Group predicts that there will only by 200 resellers left in business by 2000.
The remaining resellers may survive by looking for market niches. For example, instead of simply providing monthly Internet subscriptions, resellers are already starting to provide value-added services such as Web page hosting, Web page development, security management, and electronic commerce consulting. In these
areas, a reseller may be able to provide better service to small businesses than a larger NSP or RSP company.
3. 2 INTEREXCHANGE POINTS
With the dissolution of the NSFNET backbone, the NSF was concerned with maintaining connectivity between the commercial ISP networks and the research and education community. To address this issue, the NSF sponsored three primary and one secondary NAPs. Without the sponsorship of a core set of exchange points, the NSF feared that the commercial backbone providers would likely setup a hodgepodge of bilateral connect points potentially resulting in routing chaos.
Under the NSF model, each NAP operator provides the exchange facility while the ISP that connects to the NAP establishes the exchange agreements, also known as peering agreements, with the other ISPs connecting to the same NAP. The purpose of a peering agreement is to ensure that traffic from one ISP can reach all the customers on another ISP by exchanging routing information of the two ISPs.
Today there are many more IXP centers on the Internet other than the original four sponsored by NSF. The term NAP is applied only to the NSF sponsored IXPs, whereas all IXPs provide the same functionality, a common place for ISPs to exchange data. Various cities and organizations have used different names for the exchange point, e.g., NAP, MAE, CIX, Federal Internet Exchange (FIX). Exhibit 3-4 presents a snapshot of several of the larger IXPs in the United States.
It is important to note that an IXP does not have to serve the national ISPs. There are metropolitan exchange points (MXP) used today, which are similar in structure to the NAPs, but service only local and regional traffic. This means that traffic originating and terminating in a single region would not traverse any of the national ISPs' backbones, thus removing some of the burden on these networks. The remainder of this section describes the structure of an IXP and details the different types of peering agreements used by the ISPs at an IXP.
Exhibit 3-4
Selected Major IXP Locations
[click here to view exhibit 3-4]
3. 2 .1 IXP Functionality and Architecture
The large, national-scope IXPs, such as the NAPs or MAEs, interconnect numerous national ISPs and may exchange data requiring large amounts of bandwidth. The smaller regional or metropolitan IXPs will have fewer interconnects and require much less bandwidth. The IXP structure is similar regardless of the size of the IXP or the technical architecture used to exchange the traffic.
IXP facilities generally consist of a high-speed LAN or metropolitan area network (MAN) architecture capable of interconnecting various wide area network (WAN) technologies. ISPs connect to the IXP LAN via either a high-speed router or an asynchronous transfer mode (ATM) switch capable of connecting to the IXP architecture. Each of the connecting ISPs must negotiate bilateral or multilateral peering agreements with other ISPs interconnecting at the IXP. The Routing Arbiter administers the traffic routing resulting from these peering agreements. This traffic routing and addressing information is provided to each ISP's router by a route server within the IXP LAN. Incoming packets are routed to the high-speed LAN ring where the route server indicates the possible routes available to the packet. The most common NAP architecture is a Fiber Distributed Data Interface (FDDI) dual ring backbone LAN running at 100 Mbps. Routers for each ISP are homed to the dual ring bus in the various access configurations discussed below:
1. The ISP provides and manages its own router collocated at the IXP facility. The ISP would have dedicated access to this router via its own dedicated line (typically a T1 or T3). This option may not be available at all IXPs because of space limitations.
2. The ISP leases an IXP provided router located at the IXP. The ISP has dedicated access to the IXP router via its own dedicated line.
3. The ISP leases the dedicated connection and the router from the IXP.
4. The ISP leases switched access service to the IXP facility from the IXP or another provider. Switched access may include ATM, Switched Multimegabit Data Service (SMDS), and frame relay.
These access configurations are shown in Exhibit 3-5. For each of the access configurations, all equipment is located in a single facility.
Exhibit 3-5
Typical National-scope IXP Configurations
[click here to view exhibit 3-5]
Other IXP architectures that have been used include SMDS and ATM networks. Lower bandwidth solutions such as SMDS may be more commonplace in regional or metropolitan IXPs.
All IXPs are privately owned and administered by IECs, Incumbent Local Exchange Carriers (ILEC), Competitive Local Exchange Carriers (CLEC), or ISPs. The four NSF-sponsored NAPs are owned by Sprint, MFS, Pacific Bell, and Ameritech. Regional and metropolitan IXPs may also be owned by ISPs, e.g., the SMDS Washington Area Bypass (SWAB) is operated by PSINet. The IXPs normally charge flat interconnection fees and usage based fees to the interconnecting ISPs.
Large IECs, ILECs, and CLECs can provide network management for their IXPs from their PN network management centers. Most IXP operators will ensure reliability of service and mean time to repair, and provide maintenance for collocated equipment. The dual ring FDDI buses used in many large IXPs are also very robust to a single line fiber cut. A single dedicated connection from the ISP network to the IXP router will pose the greatest vulnerability in the IXP architecture. Redundant connections to the IXP should be used by regional ISPs that do not have presence at multiple IXPs.
3. 2 .2 IXP Peering Agreements
The policies for data exchange at an IXP are set forth by the parties involved. Just because an ISP connects to a particular IXP does not guarantee that that ISP can exchange traffic with every other ISP connected to that exchange point. Agreements that specify how traffic is carried and transferred, and how billing is handled have to be established and maintained between the ISPs on an IXP. Any ISP can connect to an IXP as long as the ISP agrees to the predefined policies. Currently, there are three different types of exchange policies:
A bilateral agreement is between only two ISPs at an exchange center. A multilateral agreement is between many ISPs at an exchange center. A multi-party bilateral agreement is between a small ISP and a large ISP to carry the small ISP's traffic to other ISPs. The more IXPs a single ISP connects to the better the performance and reliability of the ISP's service. Each IXP has its own procedures for establishing peering agreements among the IXP-attached ISPs.
A peering agreement is defined as the advertising of routes via a routing protocol for customers of the IXP participants. Specifically, the ISP is obligated to advertise all its customer's routes to all other participating ISPs and accept routes from the customer's routes advertised by the ISP. ISPs are required to peer with the IXP's route server which facilitates the routing exchange between the ISPs routers. The route server gathers the routing information from each ISP's router, processes the information based on the ISP's routing policy requirements, and passes the processed routing information to each of the IXP-attached ISPs. Currently, ISI handles the work done on the routing management system, while Merit implements and maintains the route servers and route server databases.
3. 2 .3 National-scope IXP Architecture Example
Pacific Bell's NAP, located in San Francisco, California, is fairly typical of national-scope IXPs. PacBell's NAP is an ATM/FDDI hybrid LAN, whereas other national-scope IXPs may be straight FDDI design or an FDDI/Ethernet hybrid. PacBell's use of ATM makes it one of the fastest IXPs, capable of up to 139 Mbps for OC-3 access. PacBell's FasTrakSM ATM Cell Relay Service offering is being rolled out in phases, first utilizing Permanent Virtual Circuits (PVC) and in the future, Switched Virtual Circuits (SVC). As the ATM technology matures and becomes more of an industry and user standard, PacBell and other IXP operators will migrate to fully switched ATM IXP backbones.
SF NAP consists of ATM switching sites in the San Francisco area connected by OC-3 Synchronous Optical Network (SONET) links. Participants can access the NAP network using an ADC Kentrox ADSU and a Cisco 7000 or 7010 router. Access speeds reach 36.8 Mbps for DS-3 access and 139 Mbps for OC-3 access.
In addition to the ATM network, the NAP includes an interconnected FDDI dual-ring LAN. The FDDI LAN provides service to customers that require bandwidth less than 30 Mbps. The FDDI LAN was added when PacBell tests indicated that the ATM network was dropping cells at speeds between 20 Mbps and 30 Mbps. ISPs provide or lease dedicated T1 or T3 connections to PacBell DSUs and Cisco 7000 routers connected to the FDDI backbone. Exhibit 3-6 depicts PacBell's San Francisco NAP ATM/FDDI hybrid network architecture.
3. 2 .3 .1 Routing
Each participating ISP must negotiate bilateral peering agreements with other ISPs before connecting with PacBell's San Francisco NAP. Routing on the FDDI ring is accomplished via the route server database maintained by the Routing Arbiter. On request, PacBell will provide NAP clients with a PVC to the Routing Arbiter route server database to receive and provide routing updates. Routing among peered ISPs may also be accomplished by direct PVC connections between the ISPs at the NAP, without regards to the route server database.
Exhibit 3-6
PacBell San Francisco NAP ATM/FDDI Hybrid Architecture
[click here to view exhibit 3-6]
3. 2 .3 .2 National ISP Clients
The San Francisco NAP interconnects numerous national and regional ISPs. National ISPs include ANS, MCI, and Sprint.
3. 2 .4 Metropolitan IXP Architecture Example
PSI, Inc. manages a metropolitan IXP in the Washington, DC area. PSI established the SWAB as an alternative IXP to the MAE-EAST NAP. SWAB operates nearly identically to the national-scope IXPs, requiring participating ISPs to negotiate peering agreements. Unlike the NAPs, the SWAB network is not facilities-based. Instead, each interconnecting ISP subscribes to Bell Atlantic's SMDS service over which the TCP/IP is routed.
Each participating ISP must subscribe to Bell Atlantic's SMDS service at a specified access class (speed). SMDS may be accessed at up to 34 Mbps, making it a lower bandwidth solution than FDDI or ATM. The ISP must supply its own dedicated access (either T1 or T3) to the SMDS service. To route TCP/IP over SMDS, the ISP must also provide an SMDS capable CSU/DSU and an IP router that supports SMDS encapsulation at the SWAB interface.
SWAB provides broadcast capabilities by use of SMDS address groups. The SWAB participants can have their SMDS address included in the SWAB SMDS address group for broadcast purposes.
3. 2 .4 .1 Routing
The functionality of the Routing Arbiter's route server database is provided using SMDS address screening. Address screening is used to filter out SMDS addresses from the SMDS connection, analogous to how the SS7 network can screen calls from a voice line. An ISP's screen accepts packets from peered ISPs, while refusing packets from other ISPs. Each ISP must request that Bell Atlantic screen SMDS addresses from their SWAB interface.
3. 2 .4 .2 National ISP Clients
Currently, PSINet and UUNet are the only national ISPs interconnected at the SWAB.
3. 3 INTERNET ROUTING PROTOCOLS
The Internet, as previously described, is a collection of networks that allows communications between research institutions, universities, and many other organizations worldwide. These networks are connected by routers. A router is connected to two or more networks, appearing to each of these networks as a connected host. Forwarding an IP datagram generally requires the router to choose the address of the next router in the path or, for the final hop, the destination host. This choice, called routing, depends on a routing database located within the router. The routing database is also known as a routing table or forwarding table. The routing database should be maintained dynamically to reflect the current topology of the Internet. A router normally accomplishes this by participating in distributed routing and least-cost routing algorithms with other routers.
Routers within the Internet are organized hierarchically. Some routers are used to move information through one particular group of networks under the same administrative authority and control, known as an autonomous system (AS). Routers used for this purpose are called interior routers and they use a variety of Interior Gateway Protocols (IGP). Routers that move information between ASs are called exterior routers and they use Exterior Gateway Protocols (EGP).
There is no standard protocol for either IGP or EGP. However, there are three protocols that are used by the ISPs and at the IXPs on the Internet. Generally, ISPs use the Routing Information Protocol (RIP) or the Open Shortest Path First (OSPF) protocol.
Most IXPs use the Border Gateway Protocol Version 4 (BGP4) as their routing protocol. All three protocols are dynamic in that the routers interact with adjacent routers to learn which networks each router is currently connected. The IGP protocols, RIP and OSPF, are detailed in Section 3.3.1 and 3.3.2, respectively. BGP4 is presented in Section 3.3.3.
3. 3 .1 Routing Information Protocol
RIP was developed by the Xerox Corporation in the early 1980s for use in Xerox Network Systems (XNS) networks. RIP is a dynamic protocol that continually updates its routing table based on information received from its adjacent routers. RIP is a distance-vector protocol, meaning that each router maintains a table of distances (hop counts) from itself to each other router in the system. These routing tables are updated based on RIP messages from adjacent routers.
RIP performs five basic operations:
On execution, RIP determines which of the routers interfaces are up and sends a request packet out on each interface. The purpose of this request packet is to ask each of its adjacent routers for their entire routing table.
A request received operation occurs when a router receives a packet from one of its adjacent routers asking for all or part of the router's routing table. The router will process the request and reply by sending the requested data.
A response received operation occurs when a router receives a response to its request for all or part of its adjacent routers' routing table. When a response is received, the router must validate the response and update its routing table.
Regular routing updates occur every 30 seconds. A router sends either all or part of its routing table to all of its adjacent routers. This ensures that each router on the network consistently has an accurate routing table. Finally, a triggered update occurs when a router notices that one of its routes has changed. The router sends all routes from its routing table which are affected by the changed route, which may or may not be the entire table.
Although RIP appears to be a very simple protocol, it does have serious limitations. First, as shown by the series of operations in RIP, the protocol propagates either all or part of a router's routing table every 30 seconds in addition to any triggered updates. Subsequently, the protocol is very slow to stabilize when network failures or routing errors occur.
Second, RIP limits the number of hops between any two hosts on the network to 16. This means that hosts that are more than 15 hops apart within a single AS will not be able to communicate with one another. As a result, RIP is not well suited for large internetworks and works best in small environments.
Finally, when faced with multiple routes between a router and a network, RIP always chooses the path with smallest number of hops. This choice does not consider other cost factors such as line speed and line utilization, which are important when choosing a path between two nodes. Although RIP is still a very popular protocol, many companies are moving toward its replacement, OSPF.
3. 3 .2 Open Shortest Path First
OSPF was developed by the Internet Engineering Task Force (IETF) as a replacement for RIP. OSPF is designed to overcome the limitations of RIP and is supported by all major routing vendors. OSPF uses IP and its own protocol and the transport layer, not UDP or TCP. OSPF is a dynamic link-state protocol, unlike RIP, which is a distance-vector protocol. In a link-state protocol, a router does not exchange distances with its neighbors. Instead, each router tests the status of its links with its neighbors and sends this information to each adjacent router. Routers using OSPF are able to build an entire routing table based on the link-state information received from each of its neighbors.
In contrast to RIP, OSPF does not make its routing decisions based on the number of hops to a destination. Instead, OSPF assigns a dimensionless cost to each of interfaces of the router. This cost is not based on hop count, but on throughput, round trip time, reliability, etc. When the router is faced with multiple paths for a particular route, the routing decision is made using this cost. If two routes exist with the same cost, OSPF distributes the traffic equally among the routes. Additionally, OSPF allows multiple routes to a destination based on the IP type of service, e.g., Telnet, FTP, SMTP. This means that a router can chose the best route for outgoing packets based on the type of traffic contained within the packet.
As described in Section 3.3.1, RIP is not well suited for larger internetworks because of its functionality. OSPF however, is designed for larger networks and stabilizes much faster when network failures or routing errors occur. OSPF also does not impose limitations on the number of hops between any two hosts because it does not use this metric when making routing decisions. Although RIP is still very popular, OSPF will ultimately replace RIP as the Internet grows.
3. 3 .3 Border Gateway Protocol Version 4
The primary routing protocol used on the Internet is BGP4. This protocol is used on Internet core (high level) routers to dynamically learn network reachability, respond to outages, and avoid routing loops in interconnected networks. Although RIP and OSPF are IGPs, BGP4 is an EGP used to pass traffic between different autonomous systems. BGP4 uses the TCP protocol to communicate routing information with its BGP4 peers.
Routers using BGP4 classify traffic as either local traffic or transit traffic. Local traffic is traffic that either originates or terminates in the router's AS. All other traffic is classified as transit traffic. The goal of BGP4 is to reduce the amount of transit traffic on the Internet.
The BGP4 system exchanges network reachability information with other BGP4 systems. This information includes the full path of autonomous systems that traffic must transit to reach the destination. The network reachability information is used by the router to construct a graph of AS connectivity. Once constructed, routing loops can be removed from the AS connectivity graph and routing policy decisions can be enforced.
BGP4 peers initially exchange their full routing tables. From then on incremental updates are sent as the routing tables change. BGP4 assigns a version number to the routing table and all adjacent routers will have the same version number for their routing tables. This version number changes whenever the routing table is updated as a result of routing information changes. To ensure that the each adjacent router is alive, keepalive3 packets are sent between BGP4 peers whereas notification packets are sent in response to errors or other special conditions.
After a router using BGP4 receives routing updates, the protocol decides which paths to choose to reach a specific destination. Like RIP, BGP4 is a distance-vector protocol that allows only a single path to a destination. However, BGP4 does not impose a limit on the number of hops between two hosts and stabilizes quickly after network failures or routing errors occur. The decision process is based on different factors, including next hop, path length, route origin, local preference. BGP4 always propagates the best path to its adjacent routers. Currently, BGP4 is used by most IXPs on the Internet but is not defined as the standard EGP.
3. 4 INTERNET ACCESS
The last (and in some ways the most vulnerable) component of the Internet architecture is the link between the service provider and customer. This access connection is typically a single dedicated or switched line over PN facilities.
Because access is provided over a single PN line, the connection is vulnerable to outages. This situation is identical to the "last mile" vulnerability of the PN architecture. Most other parts of the Internet architecture can use redundant links to route around outages. However, the access link is typically a single point of failure for an end user's connection to the Internet.
Internet access can be divided into two broad categories: business access and residential access. These categories are described separately below.
3. 4 .1 Business Access
Large and medium-size businesses use dedicated lines to connect their enterprise LAN/WAN to the Internet. These lines are either bundled with the ISP's service or leased separately by the company. In either case, the connection travels over PN facilities.
Most large businesses use T1 (1.544 Mbps) or higher connection speeds. Medium-size businesses use T1 or fractional T1 speeds (i.e., 128 Kbps to 768 Kbps) depending on their traffic requirements. Small businesses (10 to 50 employee sites) may be able to get by with a 56 Kbps leased line or a 128 Kbps Integrated Services Digital Network (ISDN) connection.
Leased line connections are available from ILECs and in metropolitan areas from CLECs. Today, CLEC companies include CAPs (e.g., Metropolitan Fiber Systems, Teleport Communications Group) and in many cases, IECs (e.g., LDDS, AT&T, MCIMetro). As legislation opens the local exchange to increased competition, leased lines may be available from utility companies, cable companies, or other providers.
3. 4 .2 Residential Access
Residential access connects a single user's computer to an ISP, reseller, or on-line provider. Most residential access is through modem connections over a LEC analog telephone lines. However, ISDN is gaining popularity with residential users as ISDN equipment and service prices drop. Both ISDN and analog modem connections use PN switched connections. The characteristics of analog modem and ISDN connections are described in Exhibit 3-7 below.
The bandwidth allocation for ISDN and analog modems is symmetric, meaning that there is an equal amount of inbound and outbound bandwidth. Unfortunately, many traffic applications are asymmetric, whereby the user receives far more inbound traffic than he or she generates. Examples of asymmetric applications include video-on-demand (small request to access a movie results in many gigabits of high resolution video) and Internet access (small request to access a Web page results in many megabits of text and images from the Web page).
Analog Modem and ISDN Characteristics
Characteristics | Analog Modem | ISDN |
Speed | 2.4 to 33.6 Kbps | 64 to 128 Kbps |
Equipment Cost | $100 to $150 | $300 to $400 |
Representative Service Cost4 | Flat Rate Monthly ($40/month) | Monthly Plus Usage ($100/month + $0.02/minute) |
ILECs, cable companies, and direct satellite companies are testing and deploying several asymmetric access technologies (see Exhibit 3-8 below). These technologies have up to 30 Mbps of inbound bandwidth and up to 2 Mbps of outbound bandwidth.
Asymmetric Internet Access Characteristics
Characteristics | Direct Broadcast Satellite | ADSL | Cable Modems |
Service Provider | DirecTV Satellite | ILECs | Cable Companies |
Inbound Speed | 400 Kbps | 1.544 to 6 Mbps | 10 to 30 Mbps |
Outbound Speed | 28.8 Kbps (over analog phone lines) | 16 to 512 Kbps | 768 Kbps to 2 Mbps |
Equipment Cost | $1,700 | $1,000 | $500 |
Service Cost | $40/month | $60 to $100/month | $40/month |
Status | Deployed | In trial | In trial |
The direct broadcast satellite offering is the only one of the three that is currently in widespread distribution. Direct broadcast satellite allows a user to receive inbound traffic over a 1-meter satellite dish and transmit outbound traffic over a standard analog modem line.
Asymmetric Digital Subscriber Line (ADSL) is a technology developed by the RBOCs to provide high bandwidth asymmetric connections over standard copper twisted pair wire. ADSL was originally developed exclusively for the home entertainment market (e.g., video-on-demand, interactive cable). However, as residential Internet access has grown in popularity, the LECs have added Internet access to their ADSL marketing efforts. ADSL is popular with LECs because copper cable is the basis for almost every residential phone installation. ADSL has a head start over its rival technologies because of the widespread deployment of copper wire (which reaches 98 percent of U.S. homes compared to 60 percent for cable). However, ADSL does have several drawbacks:
* Installation costs are high to upgrade existing copper cable to carry ADSL signals.
* Subscribers must be within 10,000 feet of the central office to reliably receive ADSL signals.
* Strong local AM stations can interfere with ADSL signals.
* The bandwidth available for communication is far less than the bandwidth available over cable modems.
Cable modems have the highest inbound and outbound bandwidth, but also have the most obstacles to widespread deployment. Cable modems depend on a two-way communication path between the cable operator and the subscriber. Almost every cable installation is designed to provide only a one-way path for video. To facilitate Internet access over cable plant, cable operators must upgrade their coaxial cable networks to two-way operation. Once upgraded, cable operators may have additional problems with the reliability of their plant, e.g., cable wires are installed only several inches below ground level and are highly susceptible to outages due to unintentional cable cuts. Once these issues are addressed, cable modems may easily fill a niche in the new market of Internet-enabled television (i.e., WebTV). Currently, access for these devices is provided using analog modems over dial-up lines.