Wide Area Networking

Home - Network - Lec1 and 2 - Lec3 - Lec3b - Lec4 - Lec5 and 6 - Lec7 - Lec9 - Lec10 - Lec11 - Lec12



Wide area networks (WANs) are comprised of routers, routing protocols, and transmission facilities. Properly constructed, WANs enable LANs to be networked together, regardless of how much geographic distance separates them.
Designing, building, and administering WANs requires a very different set of skills than does client/server or LAN administration.

Understanding WAN Technologies
Wide area networks (WANs) and their component technologies continue to increase in importance. Not very many years ago, about the only need most companies had for WANs was to internetwork two or more work locations. This is still a valuable and important use of WAN technologies, but other possible uses are rapidly appearing. For example, a company with only a single work location might need a robust connection to the Internet to support marketing, customer care, and many other functions. Alternatively, outsourcing certain operations or functions, as well as collaborative efforts with other companies, can necessitate the internetworking of privately owned LANs.

Unfortunately, WANs are very different from LANs.
Most LAN technologies adhere tightly to industry standards.
WANs are composite structures built from many different technologies-some standard, some highly proprietary. Many of the competing technologies also differ radically in features, performance, and cost. The most difficult part of building a WAN is matching the appropriate technologies in such a way as to satisfy the underlying business requirements. This requires a deep understanding of each aspect of every WAN component.

The wide area network's technology base includes the following:

Each of these technology categories expands into a surprisingly wide array of choices of individual technologies. Within each technology category lies additional variation in terms of manufacturers, models, and configurations. Before selecting vendors and specific products, you should examine each technology for potential performance capabilities relative to your expected WAN traffic load and performance requirements. Although an exhaustive review of each manufacturer's product line is outside the scope of this book, a survey of each technology can provide a solid foundation for your own evaluation of actual products.

Transmission Facilities
Transmission facilities used to construct the WAN present the richest array of options for the network planner. These facilities cover a wide range of sizes, varieties, and costs.

Note: The DS-3 specification offers 44.736Mbps of bandwidth. When discussed, this bandwidth is frequently referred to simply as 45Mbps.

These facilities also vary greatly in the manner in which they provide connections. There are two primary types of facilities: These two types encompass all versions of facilities, although technological innovation may be blurring their boundaries somewhat. Some would also include a third type of facility, called cell switched, but this is so closely akin to packet switched that there's little difference between them.

Circuit-Switched Facilities
Circuit switching is a communications method that creates a switched, dedicated path between two end stations. A good example of a circuit-switched network is the telephone system. A telephone is hard-wired to a central office telecommunications switch that is owned and operated by the local exchange carrier (LEC). There are many LECs and even more telecommunications switches in the world, yet any telephone can establish a connection to any other telephone through a series of intermediary central office switches. That connection is a physical circuit and is dedicated to that session for the duration of the communications session. After the telephones terminate their sessions, the physical circuit through the switched telecommunications infrastructure is torn down. The resources are then freed up for the next call.

The creation of a dedicated physical circuit through switches is the essence of circuit switching. Every unit of transmission, regardless of whether it's a cell, a frame, or anything else that may be constructed, takes the same physical path through the network infrastructure. This concept may be applied in several different formats. Three examples of circuit-switched transmission facilities include: leased lines, ISDN, and Switched 56

Leased Lines
The leased line is the most robust and flexible of the circuit-switched transmission facilities. These circuits are called leased lines because they're leased from telecommunications carriers for a monthly fee.

In North America, the dominant system for providing digital leased line service is known as the T-Carrier system. The T-Carrier enables 1.544Mbps of bandwidth to be channelized into 24 separate transmission facilities over two pairs of wire. Each channel is 64Kbps wide and can be further channelized into even smaller facilities, such as 9.6Kbps. The 1.544Mbps facility is known as the T-1. A higher capacity facility also exists within the T-Carrier system. This is the 44.736Mbps T-3 facility.

Note: Leased lines are frequently called dedicated or private lines because their is reserved for only the company that's leasing them.

Integrated Services Digital Network (ISDN)
ISDN is a "dial-on-demand" form of digital circuit-switched technology that can transport voice and data simultaneously over the same physical connection. ISDN can be ordered in either Basic Rate (BRI) or Primary Rate (PRI) Interfaces.

The BRI offers 144Kbps in a format known as 2B+D. The 2B refers to two 64Kbps B channels that can be bonded together to form one logical connection at 128Kbps. The D channel is a 16Kbps control channel used for call setup, takedown, and other control functions.

The PRI is, typically, delivered over a T-1 facility at a gross transmission rate of 1.544Mbps. This is usually channelized into 23 64Kbps B channels and one 64Kbps D channel. Higher-rate H channels of either 384, 1536, or 1920kbps can be used instead of, or in combination with, the B and D channels. Although ISDN is technically a circuit-switched facility, it can support circuit-switched, packet-switched, and even semipermanent connections.

Switched 56
Another dial-on-demand circuit-switched variant is Switched 56. Switched 56 offers 56Kbps of bandwidth between any two points that subscribe to this service. As with any dial-on-demand service, no circuit exists until a call is placed. Then the circuit is constructed between the origination and requested destination points. The actual path taken through the switched communications infrastructure is invisible, and immaterial, to the end users. This circuit is torn down when the session is terminated.

The nondedicated nature of Switched 56 makes it an affordable alternative to leased lines. You pay based on usage rather than for the luxury of having bandwidth reserved for you, regardless of whether it's being used. Balanced against affordability is performance. Switched 56 circuits must set up calls to requested destinations. This takes time. Therefore, establishing a communications session can be done much more quickly over a 56Kbps leased line than it can over a Switched 56. After the call is established, performance should be comparable.

Switched 56 is a mature and declining technology. It once offered a combination of lower cost than leased lines but much higher performance than modems and POTS lines. Today, advances in signaling techniques have enabled modems to close the performance gap. Switched 56 still offers a slight improvement over the so-called 56Kbps modems (despite what their name says, they cannot provide and sustain that transmission rate), but not much. Today, Switched 56 is probably best suited as an emergency contingency to leased lines.

Packet-Switched Facilities
Packet-switching facilities feature an internal packet format that's used to encapsulate data to be transported. Unlike circuit-switched facilities, packet-switched facilities do not provide a dedicated connection between two locations. Instead, the premises access facility interconnects with the telecommunications carrier's switched infrastructure. Packets are forwarded in a connectionless manner through this commercial packet-switched network (PSN). The lack of an easily defined path between any two locations has led to the overuse of the cloud symbol as the ubiquitous, but amorphous, network. Two examples of packet-switched networks are the old but familiar X.25 and its more up-to-date cousin, Frame Relay, both discussed in the following sections.

Frame Relay

Frame Relay WANs are built by provisioning a point-to-point private line from the work location to the nearest central office that provides this service. At the central office, this private line terminates in a Frame Relay switch that's either fully or partially meshed with the other Frame Relay switches that compose the carrier's Frame Relay commercial infrastructure. Much like the central office voice switches that compose the Public Switched Telephone Network (PSTN), the Frame Relay switches remain invisible to the user community and its applications.

Frame Relay's primary benefit is that it can reduce the cost of networking locations that are geographically dispersed by minimizing the length of premises access facilities. These circuits are commercially available at 1.544Mbps, with CIRs used to create logical sub-rate connections to multiple locations.

Balanced against this minimization of access facilities cost for point-to-point leased lines is a reduction in performance. Frame Relay introduces a significant amount of overhead in terms of framing and protocol, which is added to the overheads of the point-to-point leased line. The rule of thumb that guides engineering the DLCI and CIRs on a Frame Relay connection is to subscribe a maximum of 1.024Mbps of the 1.544Mbps of available bandwidth. This guarantees that each DLCI receives its committed information rate and that a margin of extra bandwidth is available for temporarily bursting beyond this rate.

Cell-Switched Facilities
A close relative to packet switching is cell switching. The difference between a packet and a cell is the length of the structure. A packet is a variable-length data structure, whereas a cell is a fixed-length data structure. The most familiar cell-switched technology is Asynchronous Transfer Mode (ATM). Although, technically speaking, ATM is currently a circuit-switched technology, it's best categorized independently. ATM was designed to take advantage of the higher-speed transmission facilities such as T-3 and the SONET architectures.

Asynchronous Transfer Mode (ATM)
ATM was originally designed as an asynchronous transport mechanism for broadband ISDN. ATM's low latency and high bit rate, it was speculated, would make it equally ideal for use in local area networks. The subsequent market hype has almost completely cemented its reputation as a LAN technology, to the exclusion of its capabilities as a WAN technology.

As a cell-switched WAN technology, ATM is commercially available at 1.544Mbps (DS1) or 44.736Mbps (DS-3), although this availability varies geographically. Initially, wide area ATM was available using only permanent virtual circuits, much like the DLCIs of Frame Relay. Ultimately, however, wide area ATM will be a switched technology that's capable of forwarding individual cells without requiring the overhead of establishing a permanent virtual circuit or reserving bandwidth.

Choosing Communications Hardware
The communications hardware needed to build a WAN includes three basic categories: In this context, DCE refers to the telecommunications carriers' gear. As such, there's very little you can do to actually select DCE; therefore, it's not discussed in this section.

CPE refers to the physical telephony mechanisms that are used to tie premises communications equipment, including routers, LANs, switches, and so forth, to the commercial telephony network of the communications carrier.

Premises edge vehicles are those mechanisms that connect the LAN to the CPE. They generally operate at Layers 2 and 3 of the OSI Reference Model and are responsible for forwarding and receiving packets, based on internetwork addresses. Edge vehicles are the mechanisms that separate LAN from WAN, in the context of telecommunications. Both CPE and edge vehicles are customer provided.

Telecommunications carriers, of course, deploy a considerable amount of hardware to support the transmission facilities they provide to customers. Such hardware typically remains invisible to the users and LAN administrators so we don't need to discuss here.

Customer-Provided Equipment (CPE)
CPE is the Physical Layer telephony hardware that encodes signals and places them on the transmission facility. This hardware is almost always provided by the customers and is installed in their physical premises, on their side of the demarcation point. The demarcation point, referred to as the demarc, is the official boundary between the physical plant of the telecommunications carrier and the customer's physical plant that's connected to the carrier's telephony infrastructure.

The demarc is usually just a modular jack box that's labeled with the circuit identification numbers. The telecommunications carrier owns this box, as well as everything that's hard-wired to it. The customer is responsible for all equipment that connects to the modular jack receptacle. This customer-provided equipment is the CPE.

The types of CPE varies by transmission technology. The two most common forms of CPE are the CSU/DSU and the PAD. Both of these are further examined in the following sections.

Channel Service Unit/Digital Service Unit (CSU/DSU)

The typical WAN is constructed with leased line, circuit-switched transmission facilities. Therefore, the typical CPE is known as a CSU/DSU (Channel Service Unit/Digital Service Unit). The CSU/DSU assumes that the transmission facility is a leased line and that no dial-up connections are possible.

CSU/DSUs are data communications equipment that terminate channelized and digital transmission facilities. This termination typically takes the form of a modular jack. The CSU/DSU also features a serial connection to the router at the customer's premises edge. The CSU/DSU provides more functionality than simply transmitting and receiving the physical signals. Depending on the brand and model, CSU/DSUs can also perform line conditioning and respond to diagnostic queries from the central office. These units are essential in any leased circuit that supports transmission rates of 56Kbps or greater.

Packet Assembler/Disassembler (PAD)
Transmission facilities that use packet switching may require a different device to create and dismantle the packets. This device is known as a PAD. PAD is actually an acronym for Packet Assembler/Disassembler. A good example of a network technology that uses PADs is the X.25 network. X.25 typically used a 9.6Kbps transmission facility to interconnect a user's premises to the telecommunications carrier's switched network infrastructure. The terminating device on these relatively low-speed facilities was the PAD.

Today's packet-switching technologies tend to use circuit-switched transmission facilities. For example, Frame Relay has clearly evolved from X.25, yet it doesn't use a PAD. Instead, LANs can be interconnected via Frame Relay using logical subchannels carved out of a T-1 facility. Given that the T-1 provides 1.544Mbps of bandwidth, it must terminate in a CSU/DSU at the customer's premises, regardless of which transmission technology it supports. Therefore, a WAN built using Frame Relay features routers and CSU/DSUs at each site. The CSU/DSU at each site connects to a T-1 transmission facility that interconnects them via the Frame Relay network.

Premises Edge Vehicles
A premises edge vehicle is the equipment that interconnects a customer's local area network with the CPE. In the typical LAN environment, this is a router. Routers function as the boundary between the LAN and the WAN. As such, their primary responsibility is keeping track of the routes to known internetwork addresses. These addresses are stored in routing tables that correlate the address with the physical interface on the router that must be used to get to that address.

Understanding Internetwork Addressing
Wide area networking invariably creates the need for addressing devices that reside beyond one's local LAN. Internetworking addresses are constructs of Layer 3, the Network Layer of the OSI Reference Model. These addresses are used to access and exchange data with hosts on other subnetworks within the WAN.

The address architecture is determined by the routable protocol that's used within the WAN. Some of the possibilities include IPv4, IPv6, IPX, and AppleTalk. Each has its own unique addressing scheme. Therefore, the choice of protocol determines the possible address hierarchies that can be implemented.

Ensuring Unique Addressing

The single most important aspect of internetwork addressing is uniqueness! With the solitary exception of IPv6, any network protocol you select requires that at any given point in time, there's only one endpoint with any given address. Redundant internetwork addresses create routing errors and compromise the consistency of your user's networkbased operations.

Theoretically, if your WAN is not going to be directly interconnected with the Internet, or to any other network, internetwork addresses can be arbitrarily selected. Generally speaking, arbitrarily selecting intemetwork addresses is short-sighted and a dangerous decision. That being said, Request for Comment (FRC) #1597 was released in May, 1993 and posited a plan to the contrary. Three ranges of addresses that could be used for internal networking purposes only were identified and reserved. These ranges include one each of IPv4's Class A, B, and C addresses. They are as follows:

These ranges are reserved by the Internet Assigned Numbers Authority (IANA) for use in private networks. One stipulation of RFC #1597 is that these addresses can't be used when directly accessing the Internet. Companies that use these addresses, and subsequently find the need to access the Internet, can use a proxy server (a server which forwards requests on behalf of another) with a unique and registered IP address as an intermediary. Alternatively, Network Address Translation (NAT) can be used.

Internetworking with Different Protocols
Not every WAN has the luxury of using a single routed protocol. Multiprotocol networks present some basic challenges that must be overcome. The problem is providing connectivity across dissimilar protocols. Two approaches can be used: tunnels and gateways.

Tunnels
Tunnels are a relatively simple construct that can be used to pass data through an otherwise incompatible network region. Data packets are encapsulated with framing that's recognized by the network that transports it. The original framing and formatting is retained but treated as "data"

Upon reaching its destination, the recipient host unwraps the packet and discards the "wrapper". This results in the packet being restored to its original format, complete with its original internetwork addressing.
Example: the tunneling of IPv4 packets through an IPv6 network region. Because of the inherent difference in the length of these two protocols' addresses, they're not directly compatible. To overcome this incompatibility, IPv4 packets are wrapped in IPv6 by Router A for transmission through an IPv6 WAN. Router B removes the IPv6 wrapper and presents the restored IPv4 packet to the destination host in a form it can recognize.



Gateways
If your WAN requires the interconnection of subnetworks with dissimilar routed protocols, you need a gateway at the border of the dissimilar regions. A gateway is any device that can translate between the address architectures of the two protocols. Gateways can be routers or hosts. The only criterion is that the device must be capable of translating between the two protocols' address architectures.

Routers have two ways of performing such an address translation. First, they can use two different routing protocols. This requires the router to calculate routes, forward route information, and forward packets in both protocols. Routers were designed to operate in multiprotocol environments, so this should not represent any operational difficulties.

Alternatively, a router may have an integrated protocol that is simultaneously capable of routing two different protocols and addresses. Examples of this form of routing protocol are the emerging series of "routing" protocols that are designed to facilitate the migration between IPv4 and IPv6. Specific examples are OSPF(Cisco) and RIP(MicroSoft).

Using Routing Protocols
Dynamic routing protocols are used by routers to perform three basic functions: Dynamic routing protocols fall into three broad categories: distance- vector, link-state, and hybrids. Each are discussed in the sections that follow. Their primary differences are in the way they perform the first two of the three aforementioned functions. The only alternative to dynamic routing is static routing, which is described in the section titled "Static Routing."

Distance-Vector Routing
Routing can be based on distance-vector algorithms (also sometimes called Bellman-Ford algorithms) , which require that routers periodically pass copies of the routing tables to their immediate network neighbors. Each recipient adds a distance vector (its own distance "value") to the table and forwards it to its immediate neighbors. This process occurs onmidirectionally between immediately neighboring routers.

This step-by-step process results in each router's learning about other routers and developing a cumulative perspective of network "distances." For example, an early distance-vector routing protocol is Routing Information Protocol, or RIP. RIP uses two distance metrics for determining the best next path to take for any given packet. These distance metrics are time sensitive, as measured by "ticks" and hop count. The cumulative table is then used to update each router's routing tables. When completed, each router has learned vague information about the distances to networked resources. It does not learn anything specific about other routers or learn the network's actual topology.

This approach can, under certain circumstances, actually create routing problems for distance-vector protocols. For example, a failure in the network requires some time for the routers to converge on a new understanding of the network's topology. During the convergence process, the network may be vulnerable to inconsistent routing and even infinite loops.

Certain safeguards may mitigate many of these risks, but the fact remains that the network's performance is at risk during the convergence process. Therefore, older protocols that are slow to converge may not be appropriate for large, complex WANs.

Link-State Routing
Link-state routing algorithms, known cumulatively as shortest path first (SPF) protocols, maintain a complex database of the network's topology. Unlike distance-vector protocols, link-state protocols develop and maintain a full understanding of the network's routers, as well as of how they interconnect.

This understanding is achieved via the exchange of link-state packets (LSPs) with other directly connected routers. Each router that has exchanged LSPs then constructs a topological database using all received LSPs. A "shortest path first" algorithm is then used to compute reachability to networked destinations. This information is used to update the routing table. This process is capable of discovering changes in the network topology that may have been caused by component failure or network growth. In fact, the LSP exchange is triggered by an event in the network rather than running periodically.

Link-state routing has two potential areas for concern. First, during the initial discovery process, link-state routing can flood the network's transmission facilities, thereby significantly decreasing the network's capability to transport data. This performance degradation is temporary but very noticeable.

The second area for concern is that link-state routing is memory and processor intensive. Routers configured for link-state routing tend to be more expensive because of this problem.

Hybridized Routing
The last form of the dynamic routing discipline is hybridization. Although "open" balanced hybrid protocols exist, this form is almost exclusively associated with the proprietary creation of a single company, Cisco Systems, Inc. This protocol, Enhanced Interior Gateway Routing Protocol (EIGRP), was designed to combine the best aspects of distance-vector and link-state routing protocols, without incurring any of their performance limitations or penalties.

The balanced hybrid routing protocols use distance-vector metrics but emphasize more accurate metrics than conventional distance-vector protocols. They also converge more rapidly than distance-vector protocols but avoid the overheads of link-state updates. Balanced hybrids are event driven rather than periodic, thereby conserving bandwidth for real applications.

Static Routing
A router that's programmed for static routing forwards packets out of predetermined ports. After static routes are configured, routers no longer have any need to attempt route discovery or even communicate information about routes. Their role is reduced to simply forwarding packets.

Static routing is good for only very small networks that have only a single path to any given destination. In such cases, static routing can be the most efficient routing mechanism because it doesn't consume bandwidth trying to discover routes or communicate with other routers.

As networks grow larger and add redundant paths to destinations, static routing becomes a labor-intensive liability. Any changes in the availability of routers or transmission facilities in the WAN must be manually discovered and programmed. WANs that feature more complex topologies that offer multiple potential paths absolutely require dynamic routing. Attempts to use static routing in complex, multipath WANs defeat the purpose of having that route redundancy.

Protocol Selection
Selection of a routing protocol should be done carefully and with an appreciation for the long-terrn implications of your selection. Your selection of any given protocol directly affects the selection of a router vendor as well as the operational efficiency of the WAN. The preceding sections on the different classes of routing protocols, as well as the section on static routing, should have amply demonstrated the operational implications of each category of routing protocol. These implications should help you narrow down your options to a single category or class of protocols.

The next step is to determine whether you're going to use one or more router vendors in your WAN. If at all possible, try to select a single manufacturer's products. The reason for this is simple: Open routing protocols enable each manufacturer some latitude for variation. Therefore, one manufacturer's version of an open routing protocol is likely to be less than 100 percent interoperable with another manufacturer's version. Perhaps the best example of this is the well-documented differences between Bay Networks' and Cisco System's versions of the Open Shortest Path First (OSPF) protocol.

If you select a router manufacturer before you select a routing protocol, understand how doing so can limit your selection of protocols. Some routing protocols are proprietary and, consequently, available from only a single vendor.

Understanding WAN Topologies
The topology of a WAN describes the way the transmission facilities are arranged relative to the locations that they interconnect. Numerous topologies are possible, each one offering a different mix of cost, performance, and scalability. More subtly, some functional specialization may be introduced by a topology that has a direct bearing on the transmission facilities. The more common WAN topologies include the following: Although some of these may sound more like LAN topologies than WAN topologies, they are quite applicable in both arenas. Each of these is described and illustrated throughout the rest of this section. Their relative cost, performance, scalability, and technology implications are also examined.

Peer-to-Peer Topology

A peer-to-peer WAN can be developed using leased private lines or any other transmission facility. This WAN topology is a relatively simple way of interconnecting a small number of sites. WANs that consist of just two locations can be interconnected in this manner only. A small peer-to-peer WAN is depicted in Figure 10.4.

This topology represents the least-cost solution for WANs that contain a small number of internetworked locations. Because each location contains, at most, one or two links to the rest of the network, static routing can be used. Static routing can be time intensive to establish but avoids the network overheads of dynamic routing protocols. Given that there are no redundant routes to be had in this simple topology, the benefits of dynamic routing are limited.

Unfortunately, peer-to-peer WANs suffer from two basic limitations. First, they do not scale very well. As additional locations are introduced to the WAN, the number of hops between any given pair of locations remains highly inconsistent and has an upward trend. This results in varied levels of performance in communications between any given pair of locations. The actual degree to which performance varies depends greatly on many factors, including the following:

The second limitation of this approach is its inherent vulnerability to component failure. Only a single path exists between any given pair of locations. Consequently, an equipment or facility failure anywhere in a peer-to-peer WAN can split the WAN. Depending on the actual traffic flows and the type of routing implemented, this can severely disrupt communications in the entire WAN.

Another significant implication of the peer-to-peer topology's lack of route redundancy is that using a dynamic routing protocol to calculate routes and forward packets is a waste of time and CPU cycles. The route calculated between any two points can never change! Therefore, statically defining the routes may result in a better performing network.

Ring Topology

A ring topology can be developed fairly easily from a peer-to-peer network by adding one transmission facility and an extra port on two routers. This minor increment in cost provides route redundancy that can afford small networks the opportunity to implement dynamic routing protocols. Given that the cost of most transmission facilities is mileage sensitive, it would be wise to design the ring so as to minimize the overall distances of those facilities.

A ring-shaped WAN constructed with point-to-point transmission facilities can be used to interconnect a small number of sites and provide route redundancy at a potentially minimal incremental cost. The existence of redundant routes through the network means that the use of a dynamic routing protocol affords flexibility not available with static routing. Dynamic routing protocols can automatically detect and recover from adverse changes in the WAN's operating condition by routing around the impacted links.

Rings, too, have some basic limitations. Depending on the geographic dispersion of the locations, adding an extra transmission facility to complete the ring may be cost prohibitive. In such cases, Frame Relay may be a viable alternative to dedicated leased lines, provided that its performance limitations are acceptable relative to the projected traffic loads.

A second limitation of rings is that they're not very scalable. Adding new locations to the WAN directly increases the number of hops required to access other locations in the ring. This additive process may also result in having to order new circuits. For example, as shown previously in Figure 10.5, adding a new location, X, that's in geographic proximity to Sites C and D, requires that the circuit from location C to D be terminated. Two new circuits have to be ordered to preserve the integrity of the ring--one running from C to X and the other running from D to X.

The ring topology, given its limitations, is likely to be of value in interconnecting only very small numbers of locations. It's preferable to the peer-to-peer interconnection of locations only because of its capability to provide a redundant path to the locations within the ring.

Star Network Topology

A variant of the peer-to-peer topology is the star topology, so named for its shape. A star is constructed by homing all locations into a common location. One could argue that this, in essence, creates a two-tiered topology. The distinction between a star and a two-tiered topology is that the center router in a star topology may also be used to interconnect the LANs installed at that location with each other as well as the WAN.

In a two-tiered topology, as discussed later in this chapter, the second-tier router should be dedicated exclusively to interconnecting the transmission facilities of the other locations. More importantly, a two-tiered topology provides route redundancy by supporting the development of networks with multiple concentration points.

The star topology can be constructed using almost any dedicated transmission facility, including Frame Relay and point-to-point private lines. A star-shaped WAN is shown in Figure 10.6.

A star topology WAN with point-to-point transmission facilities is much more scalable than a peer-to-peer or ring network. Adding locations to the star does not require the reengincering of existing transmission facilities. All that's required is to provision a new facility between the concentration router and the router at the new location.

The star topology rectifies the scalability problems of peer-to-peer networks by using a router to interconnect, or concentrate, all the other networked routers. This scalability is afforded at a modest increase in the number of routers, router ports, and transmission facilities compared to a comparably sized peer-to-peer topology. Star topologies may actually be developed with fewer facilities than ring topologies, as Figures 10.7 and 10.8, later in the chapter, demonstrate.

The scalability of the star topology is limited by the number of ports that the router at the center of the star can support. Expansion beyond its capacity requires either a reengineering of the topology into a two-tiered topology or the replacement of the existing router with a much larger unit.

Another benefit of a star topology is improved network performance. Overall network performance in a star topology is, in theory, always better than in either a ring or peer-topeer network. This is because all network-connected devices are just three hops away from each other. These three hops include the router at the user's location, the concentrator router, and the router at the destination. This degree of consistency is unique to the star topology. However, there are two drawbacks to this approach:

It creates a single point of failure. The existence of a single point of failure means that all WAN communications can be disrupted if the concentrator router experiences a failure.

There is no route redundancy. The lack of route redundancy means that if the concentrator router fails, you're out of service until the failure is rectified. Dynamic routing protocols are not able to calculate new paths through the network because there are none!

Full-Mesh Topology

At the extreme high end of the reliability spectrum is the full-mesh topology. This topology features the ultimate reliability and fault tolerance. Every networked node is directly connected to every other networked node. Therefore, redundant routes to each location are plentiful. Implicit in this statement is that static routing is utterly impractical. You're virtually forced into selecting one of the dynamic routing protocols to calculate routes and forward packets in this type of network. A fully meshed WAN is illustrated in Figure 10.7.

This approach minimizes the number of hops between any two network-connected machines. Another benefit is that it can be built with virtually any transmission technology.

Some practical limitations are inherent, however, in a fully meshed topology. For example, these WANs can be fairly expensive to build. Each router has to be large enough to have a port and transmission facility for every other router in the WAN. This tends to make both startup and monthly recurring operational costs expensive. It also places a finite (although substantial) limit on the scalability of the network. Routers do have a limit on the number of ports they can support. Therefore, full-mesh topologies are more of a Utopian ideal with limited practical application.

One application would be to provide interconnectivity for a limited number of routers that require high network availability. Another potential application is to fully mesh just parts of the WAN, such as the backbone of a multitiered WAN or tightly coupled work centers. This option is described in more detail in the section titled "Hybrid Topologies."

Partial-Mesh Topology

A WAN can also be developed with a partial-mesh topology. Partial meshes are highly flexible topologies that can take a variety of very different configurations. The best way to describe a partial-mesh topology is that the routers are much more tightly coupled than any of the basic topologies but are not fully interconnected, as would be the case in a fully meshed network. This topology is illustrated in Figure 10.8.

A partially meshed WAN topology is readily identified by the almost complete interconnection of every node with every other node in the network. Partial meshes offer the capability to minimize hops for the bulk of the WAN's users. Unlike fully meshed networks, a partial mesh can reduce the startup and operational expenses by not interconnecting low-traffic segments of the WAN. This enables the partial mesh network to be somewhat more scalable-and therefore affordable-than a full-mesh topology.

Two-Tiered Topology

A two-tiered topology is a modified version of the basic star topology. Rather than single concentrator routers, two or more routers are used. This rectifies the basic vulnerability of the star topology without compromising its efficiency or scalability.

Figure above presents a WAN with a typical two-tiered topology. The worst-case hop count does increase by one, as a result of the extra concentrator (the backbone) router. However, unlike the peer-to-peer network presented earlier in Figure 10.4, the hop count is not adversely affected every time a new location is added to the WAN.

A two-tiered WAN constructed with dedicated facilities offers improved fault tolerance over the simple star topology without compromising scalability. This topology can be implemented in a number of minor variations, primarily by manipulating the number of concentrator routers and the manner with which they're interconnected. Having three or more concentrator routers requires the network designer to select a subtopology for the concentrator tier. These routers can be either fully or partially meshed, or they can be strung together from peer to peer.

Regardless of the subtopology selected, hierarchical, multitiered topologies function best when some basic implementation principles ate adhered to:

The concentration layer of routers should be dedicated to their tasks. That is, they're not used to directly connect user communities.

The user premises routers should intemetwork with only concentrator nodes and not with each other in a peer-to-peer fashion.

The interconnection of user premises routers to concentrator routers should not be done randomly. Some logic should be applied in determining their placement. Depending on the geographic distribution of the users and the transmission facilities used, it may be prudent to place the concentrator nodes so as to minimize the distances from the user premises.

Given that one or more routers are dedicated to route aggregation, this topology can be an expensive undertaking. This tends to limit the use of these topologies to larger companies.

Three-Tiered Topology

WANs that need to interconnect a very large number of sites, or are built using smaller routers that can support only a few serial connections, may find the two-tiered architecture insufficiently scalable. Therefore, adding a third tier may well provide the additional scalability they require. This topology is illustrated in Figure 10. 10.

A three-tiered WAN constructed with dedicated facilities offers even greater fault tolerance and scalability than the two-tiered topology. Three-tiered networks are expensive to build, operate, and maintain. They should be used for interconnecting only very large numbers of locations. Given this, it seems foolish to develop a WAN of this magnitude and not fully mesh the uppermost (or backbone) tier of routers.

Hybrid Topologies
Hybridization of multiple topologies is useful in larger, more complex networks. It enables administrators to tailor the WAN to actual traffic patterns, rather than try to force-fit those patterns into a rigid topological model. In other words, the basic topologies presented in this section are little more than academic constructs intended to stimulate creative thought. There are no limits on the topological variety that can be introduced to a WAN. The effectiveness of each topology, and the subsequent combination of WAN technologies, depends directly on your particular situation and performance requirements.

Multitiered networks, in particular, lend themselves to hybridization. As previously discussed, a multitiered WAN can be hybridized by fully or partially meshing the backbone tier of routers. Although there's no right or wrong way to build a hybrid topology, one example of this WAN is illustrated in Figure 10. 11. Due to space considerations, the building icons have been omitted from the backbone tier in this illustration.

An effective hybrid topology may be developed in a multitiered WAN by using a fully meshed topology for the backbone nodes only. This affords a fault tolerance to the network's backbone and can provide some of the hop minimization of a full mesh network without experiencing all its costs or incurring its limitations on scalability. Fully meshing the backbone of a multitiered WAN is just one form of hybridized topology. Other hybrids can also be highly effective. The key is to look for topologies and subtopologies that can be used in combination to satisfy your particular networking requirements.

Designing Your WAN
Designing a WAN requires the successful integration of all the technical components described in this chapter. Successful integration means that the performance of the finished network meets, or exceeds, performance requirements and user expectations. Therefore, it's imperative that you identify and quantify (to the extent that users cooperate) these performance criteria before you begin the design.

WAN Performance Criteria
Many different criteria, or metrics, can be applied to measure the success of a WAN. Many of these are fairly objective and can be automatically extracted from the networkmonitoring protocols native to virtually every network device. Others are subjective and can be next to impossible to determine in advance. Some of the more common metrics include the following: Component Uptime
Each physical component of the WAN can be monitored and measured for its availability using uptime. Uptime is the opposite of downtime: It's the amount of time that the device is functional and in service, relative to the users' requirements for its availability. It's quite common for uptime to be statistically overstated by measuring it on a 7 x 24 basis, even though the users' requirements may be for only 5 x 12. Remember to tailor this, and every other metric, as closely as possible to your users' stated requirements for network performance. All electronic devices even the most highly reliable-eventually fail. Most manufacturers provide a Mean Time Between Failure (MTBF) rating for their equipment as a reassurance of how reliable their products are. Typically, MTBF ratings are in the tens of thousands of hours. Conceivably, this could translate into years of trouble-free service. Unfortunately, these ratings are statistically derived. The actual time between failures of any given device depends greatly on a number of factors. These factors include the followings: In other words, your actual mileage will vary! Monitoring and tracking uptime of individual components enable you to demonstrate to your user community how well you are satisfying their requirements for the network's availability.

Trends in component uptime data can also be followed over time to identify potentially problematic components in your network infrastructure. Such trends can provide information about the general reliability of a given type or brand of hardware, which then can be used to identify individual components that may be at risk of failure.

Traffic Volumes
One of the more important metrics for any WAN is the volume of traffic it is expected to support. Volume is almost always volatile; it varies with time, business cycles, seasons, and so on. In other words, you can count on traffic volumes' being anything but constant. Given this volatility, it's important to measure volumes using maximum volumes and average volumes:

The maximum volume you expect the network to support is known as the peak volume. As its name implies, this is the greatest amount of traffic you expect the network to have to support.

Average volumes are the normalized traffic loads that you can reasonably expect during the course of a business day from any given work location. That is, the base traffic on an average day, not a numerical average of all your traffic!

Establishing these two traffic volumes is critical to the sizing of the WAN's transmission facilities as well as its routers. For example, if you expect any given location to generate a traffic load of 10OKbps during the course of a business day, it's clear that a 56Kbps transmission facility is inadequate.

Delay
Delay is one of the more common metrics that can be used to measure network performance. Delay is the time that elapses between two events. In data communications, these two events are typically the transmission and reception of data. Therefore, delay is the total amount of time required by the network to transport a packet from its point of origin to its destination. Given this definition, delay is an aggregate phenomenon, with many potential causes. Three of the more common causes include the following:

Propagation delays
This term refers to the cumulative amount of time that's required to transmit, or propagate, the data across each transmission facility in the network path that it must take. The size and quantity of each transmission facility in the network path directly contribute to the aggregate forwarding delay of any given transmission. An additional contributor to propagation delay is traffic volume. The more traffic that's flowing across a given facility, the less bandwidth that's available for new transmissions. Propagation delays are indigenous to terrestrial circuits, regardless of whether they traverse glass or copper media. Satellite uplink/downlink delays. Some transmission facilities are satellite based. They require the signal to be transmitted up to the satellite and back down from the satellite. Due to the potentially great distances between the terrestrial transmission facilities and the satellite, these delays can be quite noticeable.

Forwarding delays
The forwarding delay in a network is the cumulative amount of time that each physical device needs to receive, buffer, process, and forward data. The actual forwarding delay of any given device may vary over time. Individual devices that are operating at or near capacity ordinarily experience a greater forwarding delay than comparable devices that are lightly utilized. Additionally, forwarding delays can be exacerbated by heavy traffic or error conditions in the network. Forwarding delays are frequently identified as latency in individual components.

Resource Utilization
The degree to which the various physical resources of the WAN are being used is a good indicator of how well or how poorly the WAN is performing relative to the performance requirements. Two main categories of resource utilization rates should be monitored carefully:

Router Resources
Routers are among the most vital components of any WAN. Unlike the transmission facilities, they are outside the purview of the telecommunications carrier. Therefore, they're distinctly the responsibility of the customer. Fortunately, a router is an intelligent device that contains its own CPU and memory. These physical resources are indispensable in the calculation of WAN routes and the forwarding of packets. They can also be used to monitor the performance of the router.

If either CPU or memory utilization rates approach 100 percent, performance suffers. Numerous conditions can result in either utilization rate temporarily spiking upward, with consequential performance degradation. One example might be a sudden increase in transmissions from the LAN to the WAN. LANs can operate at speeds up to 1Gbps but usually only at 10, 16, or 10OMbps. Any of these speeds is a gross mismatch with the typical WAN transmission facility, which offers a paltry 1.544Mbps of bandwidth. This mismatch in bandwidth must be buffered by the router's memory. It doesn't take long for a router to become resource constricted, given a sustained period of heavy LAN transmissions.

If such situations are rarely experienced, they should be considered aberrations. Alberr ations should be monitored, but they shouldn't drive physical upgrades. However, if these resource constrictions recur or constitute a trend, something needs to be done.

Usually, this requires an upgrade to the next larger router or an expansion of memory. If a router is chronically at or near 100 percent of capacity with its memory, it's time to purchase additional memory.

Responding to chronically high CPU utilization rates might not be as simple as a memory upgrade. There are really only two options for improving high CPU utilization rates:

Manipulating traffic patterns is really a viable option only in larger WANs with complex topologies that afford route redundancy. Even so, if the router in question is a premises edge vehicle (as opposed to a backbone router), your only option is likely to be the forklift upgrade.

Transmission Facility Rates
Transmission facilities, too, can be monitored for utilization. Typically, this utilization rate is expressed in terms of the percentage of consumed bandwidth. For example, if you're using a T-1, a given sample might indicate that 30 percent of its 1.544Mbps of available bandwidth is currently being utilized.

These rates can be tricky to analyze and may be misleading. For example, it's not uncommon for network-management software packages to capture utilization data in time intervals. These can be one hour, five minutes, or just about any other interval. The sampling frequency, if set too coarsely, can miss short-duration fluctuations in bandwidth consumption. If the sampling is too frequent, you could find yourself mired in a meaningless morass of data points. The trick is finding the correct frequency that provides meaningful data about how the network is performing relative to the users' expectations.

Beyond merely selecting the sampling rate lies the issue of sampling windows. A sampling window is the timeframe within which samples are to be taken. Establishing a sampling window consists of establishing the frequency and duration of the sampling. The sampling window should be determined by the users' requirements for WAN availability. If the utilization samples are spread over a 24-hour day and a 7-day week, but the users work only 10 hours per day, 5 days per week, the statistical data will not be indicative of how well the users' requirements are being met.

Utilization rates are a wonderful statistical tool for monitoring and measuring the status of transmission facilities. They are not, however, the only metric for assessing a network's performance. The network is successful only if it satisfies the users' requirements. Therefore, a combination of performance metrics that provides a multifaceted, composite perspective is likely to provide a better assessment of the network's successfulness.

Costs of the WAN
Tempering any evaluation of these performance criteria is cost. The costs of owning and operating a WAN include the initial startup costs as well as the monthly, recurring expenses. Not surprisingly, the larger and more powerful network components are much more expensive than smaller, less robust components. Therefore, designing a WAN becomes an economic exercise in which a careful balance of performance and cost is achieved.

Achieving this balance can be painful. No one wants to design a WAN that will disappoint the users with its performance, but no one wants to design a WAN that blows the budget, either! Fortunately, a few truisms can help guide administrators as they choose the design of a WAN that satisfies existing requirements, provides flexibility for future growth, and doesn't exceed the budget:

The capital investments in routers and other network hardware become a fixed part of the network. After they're placed into operation, the logistics of replacing them become quite complicated. Plus, depending upon your depreciation schedule for capital equipment, you might find yourself obligated to use them for five or more years! It might behoove you to purchase a larger but relatively unpopulated router. You can add hardware (memory, CPUs, and interfaces) in the future, as the need for them arises. This makes future expansion possible at modest incremental costs and little (if any) operational downtime.

The transmission facilities are relatively easy to replace with other transmission facilities. They're expense items, not a capital investment, so there's no depreciation expense to retire. They can be replaced with other facilities as often as your lease agreement with the carrier permits. Therefore, you might want to explore your options for meeting performance requirements with the various available transmission facilities and technologies.

Applying the wisdom behind these truisms can help you meet your users' present and future expected requirements-all within the constraints of your budget.