NETWORK TOPOLOGIES

The physical layout of a network is referred to as its topology. Twenty or so years ago, there was little, if any, choice in designing a network's topology.

While there were (and are) are many ways to do it, let's look a typical "old-school" LAN. To build a LAN, you ran a fat coaxial cable called a Thicknet through your building and tapped hosts directly into it. The Thicknet cable was the network backbone. Connecting hosts directly to the network backbone resulted in a so-called bus topology. A bus is a cable (or a printed circuit board acting like a cable), and a bus topology is where most or all of a network's devices are connected to a single cable-which is like having everybody's driveway empty into one major thoroughfare instead of side streets. Two typical early network topologies are depicted in Figure 5-2.

Figure 5-2: Early network topologies employed thickand thin-wire connections

As network technology developed, topologies evolved a bit with the introduction of terminal servers, which made it possible to indirectly connect dumb terminals to the LAN. This was a good thing, because it gave individual users easy access to more than one minicomputer or mainframe. Another advance was the introduction of a thinner kind of coaxial cable called Thinnet, which was cheaper and far easier to work with than Thicknet cabling. But these improvements were only incremental; network layouts were still basically a bus topology.

The trouble with physical bus topologies was that if something failed along the trunk, the whole network went down (or at least a big part of it). Another drawback was that connecting hosts meant crawling into the ceiling plenum, finding the trunk cable, making the tap, dropping a second cable from the tap down to the device, and then testing the connection to see if it worked. Not only were early networks prone to failure and hard to install, but the equipment was also bulky and expensive.

Things have changed a lot since then. Nowadays, most hosts are connected to networks through switches. Switches give network administrators more choices in both the physical and logical layout of networks. They are modular in the sense that devices and hosts can be added without having to change anything on the network backbone. But above all, switches do away with physical bus topologies by allowing the easy installation and management of multiple LANs. Today, network designers use star topologies in place of one overtaxed LAN.

Note 

The term "LAN" can be confusing. In the old days, a LAN was a central trunk running through a building with everything on the network connected directly to it. Today, that same building might have dozens of switches serving as host connection points, with the switches, in turn, connected to a backbone. Any shared network medium is a LAN. A switch is a shared medium, and so is the backbone's central trunk cable. Even experts use the term "LAN," or "local area network," to loosely describe a local network comprised of multiple LANs. To avoid confusion, in this book, we use the term "LAN segment" to describe a shared network medium, which is the basic building block of network topologies. A LAN segment is defined by a switch or cable. Nowadays, the term "LAN" most often refers to a collection of LAN segments within a building or campus.

Breaking things up into smaller LANs makes it easier to meet current needs and still leave room for future change and growth. Network segmentation improves network performance by isolating traffic. Users within a workgroup or department are most likely to send messages to one another, so putting them on their own LAN segment means others won't get caught in their traffic. Reliability is better, because what happens on one LAN segment doesn't affect the overall network; the fault is isolated within the segment where the trouble started. Network administrators can better identify where the trouble is because of the transition points between LAN segments-an important feature in complicated networks. Also, the modularity of hierarchical networks naturally enhances security and manageability because devices can be grouped in ways that best fit management needs.

Note 

Logically, each switch port is a LAN segment. LAN segments are the individual parts of a LAN separated by a networking device, like a switch.

For all these reasons, networks today use switches to concentrate multiple hosts into a single network connection point-an approach called the star configuration. Star configurations are the building blocks with which hierarchical networks are constructed. Figure 5-3 shows common variations on the basic star topology.

Figure 5-3: The star topology is the network's basic building block

In stark contrast to the bad old days of trunk pulling and cable dropping, connecting a host to a network now is as simple as plugging in a phone-style jack. Each star-topology building block meets certain needs:

• A small business or department might use just one switch to form a LAN-in effect, putting the entire network inside a box-which is called a single-star topology. With switches, backbone cabling is no longer necessary to form small networks.

• A star-hierarchy topology is used to make more connection ports available within an office. Plugging outlying switches (access switches) into a master switch (Distribution or Core switch) gives more hosts a place to plug in without having to pull additional cable into the area.

Note 

The terms "backbone" and "central trunk" evoke the image of a single, unbroken span of cable. In reality, most backbones are made up of many cable spans spliced together. At the other extreme, some backbones aren't made up of cable at all, but instead are contained inside a box entirely on circuit boards, which is called a collapsed backbone. By definition, though, a backbone is the part of a network that acts as the primary path routing traffic between LAN segments. In general, only switches and routers are connected directly to backbones. In large networks, a backbone usually runs at a higher speed than the LAN segments it connects.

The topology of a network is, of course, most closely tied to the enterprise's geography-who's on what floor, which server sits where, and so on, but other considerations also come into play. Table 5-1 lists network design factors and how they affect decisions about what to do when designing a network.

Table 5-1: Topology Design Factors (Besides Geography)

Factor	Network Design Consideration
Pre-existing cable plant	To save time and money, network designers frequently try to run networks over wiring already installed in the walls and ceiling spaces of a building. Sometimes, they have no choice, and the type of network devices that can be used is dictated by pre-existing cabling.
Performance goals	Projected network traffic loads and end-user "need for speed" can influence the class of network devices and cabling plant used.
Platforms	The installed base of network operating systems and computer platforms frequently dictates network design decisions.
Security	Topology layout is often used as a way to help enforce security.

Network design decisions are most often constrained by the amount of money to be spent and by such logistical issues as how long the enterprise plans to stay in a building. Platforms, on the other hand, are always strategic and are usually a condition that network designers have little or no control over. If the enterprise is heavily invested in Novell NetWare or IBM SNA or Apple, the network equipment must adapt to the hardware and software platforms already installed.

Nowadays, however, no matter what your budget is or which platforms you're using, segmenting networks is not only an option, but also is the preferred design approach.

The Importance of Network Domains

The domain is one of the most fundamental concepts in internetworking. Although the term has many uses, for our purposes, what's important are the two most basic kinds of domains: the collision domain (or the token domain in Token Ring LAN segments) and the broadcast domain.

LAN segments run over shared media. In physical terms, member hosts in a LAN segment share a switch. To stave off the electronic chaos that would otherwise ensue from sharing a medium and "talking" all at once, some form of control must be enforced over access to it. This is called media access control (from whence the MAC address takes its name, as mentioned earlier).

Note 

For the literal-minded out there, thinking that the name should be "medium access control" because LANs, by definition, share only a single switched segment, you're right-to a point. Keep in mind, though, that MAC addresses are routinely exchanged between segments. Besides, somehow "medium access control" sounds half-hearted in the world of strict networking rules.

Ethernet and Token Ring are both shared media LAN technologies, but they use two sharply contrasting access control methods: Ethernet's CSMA/CD method (Carrier Sense Multiple Access/Collision Detection) and Token Ring's token-passing method.

Ethernet Collision Domains

Ethernet lets network hosts randomly contend for bandwidth. A host may send a message at will, but if it collides with a message sent by another host, both must back off and retry after a random wait period. An Ethernet collision domain is any segment or port in which collisions can take place. The more traffic there is on a collision domain, the more likely it is that collisions will occur. Increased collisions, in turn, result in hosts spending more and more time futilely attempting to retransmit.

Token Domains

A token domain is a Token Ring LAN's shared medium. Token Ring uses a deterministic method for controlling media access called token passing. In token passing, each host must wait for the token to be passed around the LAN's ring before it can grab it and transmit. Although they have no packet collisions, Token Ring LANs are not immune to traffic congestion. The more hosts connected to a ring, the longer each must wait for the token to come back around to be able to transmit. A good analogy to help understand Token Ring technology is a traffic light at the top of a highway on-ramp. You're forced to wait for a green light before entering the highway; thus, the heavier the traffic, the longer the wait. You're not going to be stuck in a jam down on the highway, but if traffic's heavy, you still must do your waiting on the ramp.

The majority of new LAN installations are Ethernet, so throughout the remainder of this chapter, we'll focus on Ethernet switches and refer to Token Ring only where appropriate.

Note 

There are three kinds of messages in IP-based networks, all ending in "cast": (1) a unicast message is a message sent to a single network address, (2) a multicast is a single message copied and sent to a specific group of network addresses, and (3) a broadcast message is sent to all nodes on a network. IPv6 networks, you might recall, introduce a fourth kind of message, called anycast. Anycast messages are transmitted by a host to the closest destination host.

Broadcast Domains

A broadcast domain is a set of all stations (network devices and hosts) that will receive any broadcast message originating from any device or host within the set. The key differentiation between broadcast and collision domains is that they are defined by the type of message they encompass. Collision domains encompass messages of any kind, while broadcast domains encompass only broadcast messages. As the lower-left part of Figure 5-4 illustrates, for two switches to join in the same broadcast domain, they must somehow be internetworked (routers usually block broadcasts).

Figure 5-4: Broadcast domains are implemented differently

The right side of Figure 5-4 shows how broadcast domains can be very different in switched networks. Using switch technology, a broadcast domain can be specifically configured through logical connections instead of physical ones. This is called a virtual LAN, or VLAN for short. The "virtual" in VLAN means that the LAN's domain is not defined by a physical connection. In fact, VLANs usually aren't even local at all (more on that in the section "VLANs" later in the chapter).

Collisions waste bandwidth because they abort transmissions. In contrast, broadcast messages indeed reach their destinations, but are still wasted bandwidth if the receiving hosts discard them as irrelevant. Obviously, then, broadcasts also play a central role in traffic congestion. Think of broadcast domains as internetworking's version of the ZIP code: the more addresses within a ZIP code, the longer it takes to deliver all the mail. Internetworking is no different. The larger the broadcast domain is, the slower the network tends to be.

The Need to Segment Networks

LAN segments should be kept small in order to help guarantee throughput speed by limiting the frequency of collisions. Small is also good when it comes to network flexibility, security, and maintainability. The trend is to divide networks into more and more LAN segments as network bandwidth comes under increasing strain. More users are becoming members of networks, and, on average, individual users are generating more network traffic. In addition, the mix is changing to more bandwidth-intensive applications, such as graphics, voice, and video. All this has combined to push network managers to deliver more bandwidth by both installing faster networking media and breaking up networks into ever smaller segments.

Network managers are doing both, but demand for bandwidth is outstripping the ability of network manufacturers to create faster technology, so network infrastructures are being reconfigured to incorporate more hierarchy and segmentation. This trend is reinforced as the cost of the hardware needed to segment networks plummets. The trend's greatest reinforcement is that the tools needed to integrate and manage heavily segmented networks have improved significantly. These tools are so good, in fact, that switches are being used to "microsegment" networks into tiny LAN segments.

Cabling Defines Network Speed and Distance

You can't appreciate traffic management without understanding the basics of road building. So, before we go into how Cisco switches address these problems, it's necessary to learn about the physical media over which networks operate: the cabling.

The most fundamental fact about wired networks is that they run over either two kinds of physical transport media: copper wire or fiber-optic cable. The vast majority of all LANs installed in the world today are on some form of copper wire. Fiber-optic cabling-often called fiber or glass for short-is mostly used for high-speed backbones.

Note 

Though copper and fiber are the most prevalent types of connecting media, wireless is picking up steam as a way to connect network hosts and devices without being tethered to a desk. We talk about wireless connectivity and networking in Chapter 8.

The proliferation of network users and bandwidth-hungry applications has driven the industry to introduce a steady stream of newer and faster transport technologies. A review of network cabling and terminology will help you keep things straight.

A Brief History of LAN Cabling

As mentioned, the earliest LANs ran over Thicknet coaxial cables. Thicknet was costly and hard to work with, so in the mid-1980s, Thinnet coaxial cable (also called Cheapernet ) became more popular. When used to run 10-Mbps Ethernet, Thinnet has a maximum length of 185 meters. Thinnet LANs can be extended beyond that distance using repeaters to link segments. ( Repeaters are devices placed along a LAN cable to amplify electrical signals and extend maximum operating length. Simple repeaters are rarely used now.) Also, coaxial cable requires that there be a certain minimum amount of spacing between connections, which cramps topology design choices.

Hubs were introduced in the late 1980s. Also called concentrators, hubs make hierarchical network topologies possible and simplify the installation and management of a cable plant. Hubs also hastened the introduction of a new type of cabling called twistedpair, which is inexpensive and easy to work with. One of the reasons using twisted-pair became possible is that its relatively short operating limit of 100 meters is extensible using hubs. For example, an office space 300 meters in length could be wired with twistedpair by placing two hubs into the topology.

LAN Cabling Today

Most larger networks today use a combination of fiber and twisted-pair. Twisted-pair is used to connect hosts to switches, while fiber is used for network backbones. Thanks to technological advances, even though twisted-pair uses less copper and shielding than Thinnet coaxial cable, it supports faster data rates. About the same time twisted-pair was taking over desktop connectivity, fiber-optic cabling established itself as the preferred medium for high-speed network backbones. Fiber is used to connect floors or major areas within an office building, and twisted-pair is used to connect LAN segments spanning from the backbone. As Figure 5-5 depicts, switches funnel the LAN segments into the backbone through various star-hierarchy configurations. "Backbone" is a relative term, however. For example, the fiber trunk interconnecting the buildings of a campus LAN is referred to as its backbone, while the cable connecting the floors of one of the buildings is referred to as a riser.

Figure 5-5: Enterprise networks today combine fiber and twisted-pair cable media

Twisted-pair cable comes in two basic types:

• STP (Shielded Twisted-Pair) A two-pair cabling medium encased in shielded insulation to limit electromagnetic interference of signals.

• UTP (Unshielded Twisted-Pair) A four-pair cabling medium not encased in shielding. UTP is used in most networks.

Generally, the more tightly twisted the copper wire strands are, the less likely it is there will be interference or signal loss. As its name suggests, STP has shielding to protect content on the wires. UTP, on the other hand, has no shielding. Because UTP is fast, reliable, and inexpensive, it has become the predominant type of cabling used in networking today. Use of the more expensive STP is limited to environments made hostile by high levels of electromagnetic interference.

Cabling Specifications Table 5-2 explains the categories of twisted-pair specified by an international standards organization called TIA/EIA (Telecommunications Industry Association/Electronics Industry Association). These cabling specifications are important in that the rate at which data can be reliably transmitted is determined by a combination of factors, such as:

• How tightly twisted the copper wire is

• The quality of the cable's copper

• The type of insulation used to encase the cable

• The design and quality of the cable connectors

Table 5-2: TIE/EIA Twisted-Pair Specifications

Category	Cable Description	Cable Application
Cat 1	Traditional telephone cable.	Not usable for networking; no longer installed for telephones.
Cat 2	Four twisted-pairs.	4 Mbps; not recommended for networking.
Cat 3	Four twisted-pairs with three twists per foot, rated up to 16 MHz.	10 Mbps Ethernet and 4 Mbps for Token Ring; also used for new telephone cabling.
Cat 4	Four twisted-pairs, rated up to 20 MHz.	16 Mbps; used for Token Ring.
Cat 5	Four twisted-pairs with eight twists per foot, rated up to 100 MHz.	100 Mbps; used for Fast Ethernet; fast becoming ubiquitous in networked buildings.
Enhanced Cat 5	Four twisted-pairs with eight twists per foot, but made of higher-quality materials and rated up to 200 MHz.	Rated to have up to twice the transmission capability of regular Cat 5.
Cat 6	Four twisted-pairs with eight twists per foot, but made of higher-quality materials.	Rated to have up to six times the transmission capability of regular Cat 5.
	Rated to support Gigabit Ethernet.
Cat 7	Four twisted-pairs, made of high-quality materials; rated up to 600 MHz.	Super-fast broadband applications, like Gigabit Ethernet; allows multiple applications operating at different frequencies.

In Table 5-2, Categories 3 and 5 represent the lion's share of twisted-pair networks today-especially Cat 5. But keep an eye on the relative newcomers, Cat 6 and 7. As bandwidth needs increase, so will their adoption rate.

Note that higher category numbers indicate higher speeds. Most new LAN installations use Cat 6 in order to accommodate Gigabit Ethernet, but many still run on older Cat 3 because it's so widely installed in network infrastructures.

The alternative to copper cabling is fiber-optic cabling. Although it's employed mostly as a backbone medium, it's sometimes used all the way out to the desktop for demanding applications, such as high-end graphics, although lower-cost Gigabit Ethernet over copper is now more commonly used in those scenarios. The advantage of fiber is that it can sustain very high speeds over long distances, but its use is constrained by relatively high costs.

Network Technologies Cabling specifications, such as Cat 6 or 7, describe the physical medium. Network specifications describe what is to happen over a medium, and are built around the capabilities and limitations of one or more cabling specifications.

There are several Ethernet specifications, each designed to guarantee efficacy on the physical medium over which it's designed to operate. Any networking technology's ability to function properly depends on how well matched it is to the physical medium. The faster a network must run-or the greater the distance over which it will operate-the better the underlying cable plant must be.

Network specification names seem mysterious until you've been introduced to the logic behind them. The following illustration breaks down the name of the Ethernet 10BaseT specification.

Sorting out the various network specifications shows that some media are used only for certain speeds, some are legacy specs no longer used in new installations, and others are specs that never quite caught on. Table 5-3 lists network specifications (mostly Ethernet) in their approximate order of importance, based on:

• Percentage of new LANs being installed

• Percentage of all installations

• Probable future importance as a technology

Table 5-3: LAN Specifications with Cable Types and Distance Limits

LAN Spec	Description
10BaseT	10-Mbps Ethernet using UTP Cats 3, 4, or 5 cabling; most new installations during the 1990s; in the process of being eclipsed by 100BaseT; 100-meter limit
100BaseTX	100-Mbps Fast Ethernet using UTP Cat 5 cabling; most new installations going in now are 100BaseT; 100-meter limit
100BaseFX	100-Mbps Fast Ethernet using two strands of multimode fiber-optic cable per link; most new high-speed backbones are 100BaseFX; 400-meter limit
FDDI	100-Mbps Fiber Distributed Data Interface token-passing LAN using either single-mode or multimode fiber-optic cabling (or sometimes either STP orUTP copper, called CDDI, for Copper Distributed Data Interface); 100-kilometer limit over fiber, 100-meter limit over copper
ATM	622-Mbps Asynchronous Transfer Mode over fiber-optic cabling; popular as a backbone for its sustained throughput and its proven ability to move multimedia applications at speed
1000BaseFX	1-Gbps Gigabit Ethernet over fiber-optic cabling; although 1000BaseX is now being reengineered to run over Cat 5 copper (to be called 1000BaseTX)
10GBASE-SR	Designed to support short distances for multimode fiber with a range of between 26 and 82 meters
10GBASE-LRM	Designed to support distances up to 220 meters on multimode cable
10GBASE-LR	Designed to support single-mode fiber for distances up to 25 kilometers
10GBASE-ER	Designed to support single-mode fiber with distances of up to 40 kilometers
10GBASE-LX4	Designed to support multimode fiber of up to 300 meters or 10 kilometers on single-mode fiber
10GBASE-CX4	Uses 4-laned InfiniBand connectors for distances of up to 15 meters
10GBASE-T	Provides 10-Gigabit connections over conventional UTP or STP cables
100VG-AnyLAN	100-Mbps Fast Ethernet and Token Ring using UTP Cats 3, 4, or 5 cabling; developed by HewlettPackard; can be run over any existing 10BaseT networks
10Base2	10-Mbps Ethernet using Thinnet coaxial cabling; widely installed in the 1980s; eclipsed by 10BaseT; 185-meter limit
10Base5	10-Mbps Ethernet using Thicknet coaxial cabling; widely installed in 1970s and 1980s; 500-meter limit
100BaseT4	100-Mbps Fast Ethernet using four pairs of UTP Cats 3, 4, or 5 cabling; 100-meter limit
10BaseFB	10-Mbps Ethernet using fiber-optic cabling; used as a LAN backbone (not to connect hosts directly); two-kilometer limit
10BaseFL	10-Mbps Ethernet using fiber-optic cabling; twokilometer limit, one-kilometer with FOIRL (FiberOptic Inter-Repeater Link, a precursor signaling methodology that FL replaces)
10BaseFP	10-Mbps Ethernet using fiber-optic cabling; used to link computers into a star topology without using repeaters; 500-meter limit
10Broad36	10-Mbps Ethernet using broadband coaxial cable cabling; 3.6-kilometer limit

The specifications reflect the worldwide trend toward Ethernet technologies. The two major exceptions are FDDI and ATM.

Table 5-3 shows that many network specifications are either old or are contending standards that meet with limited market acceptance. The trend in networking technology is, of course, toward ever faster speeds running over cheaper cabling plant. Not including high-speed backbones, most new LANs today use 1000BaseX Fast Ethernet running over Cat 6 or 7 cabling.

Things are less clear about which technology is winning out as the backbone medium of choice. ATM supplanted FDDI, probably due in part to the recent sharp increase in the demand for multimedia applications. 1000BaseX, commonly called Gigabit Ethernet, is the current popular medium. Planners not only like Gigabit Ethernet's rated speed of 1000 Mbps, but they also like its compatibility with most installed Ethernet networks. Chapter 2 discusses competing network technologies in detail.

Note 

Ever wonder how data travels over a cable? In simple terms, electrical pulses going over a wire are measured for plus or minus voltages to track signals. Special encoding schemes-for example, the Institute of Electrical and Electronics Engineers (IEEE) schemes for Fast Ethernet and Gigabit Ethernet-are used to translate data from identifiable bit patterns represented by the voltage fluctuations. Fast Ethernet uses a three-level encoding scheme to track data; Gigabit Ethernet uses a five-level encoding scheme. The two major problems facing network communications are return-loss and near/far-end crosstalk. Without getting bogged down in engineering details, return-loss is when a signal echoes back to the transmitter, confusing it; crosstalk, on the other hand, is when signals leak between wire pairs, creating electrical noise. Network engineers are always looking for improved encoding schemes to squeeze more bandwidth into smaller wires. It ain't like connecting two tin cans with baling wire, is it?

Straight-Through vs. Crossover Cables and Devices

Network hardware documentation frequently refers to straight-through cables and crossover cables. Network devices have transmitter (TX) pins and receiver (RX) pins. In a straight-through cable, the wire pair does not cross from TX to RX between interfaces. In a crossover cable, however, wire pairs are crossed over from TX to RX between connections. You must use crossover cables to connect hosts with identical interfaces. If a straight-through cable is used, one of the two devices must perform the crossover function. If neither device has a crossover connector, then a crossover cable must be used. In other words, signals must be crossed over either in one of the devices or in the cable. Figure 5-6 illustrates the two ways.

Figure 5-6: Signals can be crossed over either in the cable or in the device

Think of a signal traveling RX-to-RX or TX-to-TX as being like a conversation in which two persons mouth words, but no sound reaches their ears. Crossing over signals between devices makes networking possible by moving the signal from "mouth (TX) to ear (RX)."