Inside TCP/IP, Third Edition Network Support for TCP/IP

Transcription

1 Inside TCP/IP, Third Edition -- Ch 3 -- Network Support for TCP/IP Page 1 of 47 Inside TCP/IP, Third Edition Network Support for TCP/IP One of the reasons for the popuarity of TCP/IP networks is that TCP/IP can run on a variety of networks and computing devices. The TCP/IP protoco suite can run on computers such as main frames, mini computers, workstations, persona computers, hand-hed computers and organizers, and even ceuar phones! The actua network hardware and software is usuay transparent to the user of the network. In fact, most users are not even aware that they are using the TCP/IP protoco et aone the type of network hardware. However, this book is written for you, the network professiona, who wants to earn about TCP/IP, the "gue" that ties together the different network components. The networking hardware and how TCP/IP runs on top of it is therefore of concern to you and is discussed in this chapter. Overview of Networking Hardware that Run TCP/IP By networking hardware, what is meant is ayers two and one of the OSI mode. Reca from the discussion in Chapter 2, "TCP/IP Protoco Layering Concepts," that ayer two of the OSI mode is the Data Link ayer and ayer one of the OSI mode is the Physica ayer. The Data Link ayer transmits and receives data units caed frames and interacts with the Physica ayer that is responsibe for sending and receiving the frame over a physica media. Network interface cards are the pieces of network hardware that go inside a computer and connect the computer to the physica network. Network interface cards typicay impement the functionaity of the Data Link and Physica ayers of the OSI mode. Before describing the variety of networking hardware avaiabe for running TCP/IP appications and protocos, you might ask a very fundamenta question: Why do I need so many different types of network hardware to run TCP/IP? The motivation for having so many different types of hardware is that it is difficut for a singe type of hardware to satisfy a the requirements of a rea-ife network. Most organizations use a Loca Area Network (LAN) at a given ocation, because LANs provide high speeds at reativey ow impementation costs. However, LANs are restricted to imited distances and cannot span the distances needed by widey separated sites that need to be connected by a network. For spanning ong distances, Wide Area Networks (WANs) that connect remote sites are used. Most WANs use a fundamentay different technoogy fromlans. As their name impies, they are designed for connecting networks over ong distances. Whie WANs can span onger distances than LANs, they typicay have imited speeds and are sower than LANs. So we see that a singe LAN or WAN may not meet the needs of an organization. What is needed in this case is a LAN/WAN combination (see fig. 3.1). fie://i:\chapters\z\zc648.htm 3/21/01

2 Inside TCP/IP, Third Edition -- Ch 3 -- Network Support for TCP/IP Page 2 of 47 Such a network consists of one or more LAN and WAN networks. The LANs and WANs differ from each other in the distances they can span, the speeds they provide, and the fundamentay different network hardware technoogies they use. Networks that use such different network hardware technoogies need to be tied together by specia "gue" devices. These devices are caed bridges, routers, or gateways. (This chapter wi discuss the roes of bridges and routers to connect different network hardware technoogies. You wi earn more about routers, gateways, and Internet routing protocos in ater sections of this book.) One of the goas of connecting different types of networks is to provide users access to resources on the network regardess of the ocation of the user. This access shoud be provided in a manner that is transparent to the user. In other words, the compexity of the network and its underying technoogies, protocos, hardware devices, and cabing shoud be hidden from the user. Whie the network detais are hidden from the user, they cannot be hidden from the network designer/impementer. From a network designer/impementer perspective, the network components and resources must interoperate and be easy to manage and maintain. The network designer or impementer needs to be knowedgeabe about the network hardware and protoco components to determine which kinds of devices can be connected to each other and how they interoperate. FIGURE 3.1 LAN/WAN combination network. Figure 3.1 shows a network that consists of LANs and WANs. In each of these categories (LANs, WANs), there are severa different types of hardware that are avaiabe. The network designer or impementer must make this choice based on the business needs of the organization. The foowing are some of the more common types of LAN technoogies: IEEE LANs Ethernet Token Ring Switched networks FDDI The foowing are some of the more common types of technoogies: X.25 Frame Reay SMDS MAN SLIP/PPP/CSLIP fie://i:\chapters\z\zc648.htm 3/21/01

3 Inside TCP/IP, Third Edition -- Ch 3 -- Network Support for TCP/IP Page 3 of 47 ATM TCP/IP on IEEE Loca Area Networks As you work with TCP/IP network hardware, you wi encounter references to the Institute of Eectrica and Eectronic Engineers (IEEE) 802 standards. These are ubiquitous standards for LANs and are cosey connected with the first two OSI ayers: the Data Link ayer and the Physica ayer. The IEEE standard defines Ethernet LANs by the name of IEEE standard, and Token Ring LANs by the name of IEEE standard. These standards constitute a arge number of LAN networks that run TCP/IP. A brief background on the evoution of IEEE LAN standards wi be hepfu here, and this section offers just that. The IEEE undertook Project 802 in February of 1980 to identify and formaize LAN standards for data rates not exceeding 20 megabits per second (Mbps). Standardization efforts resuted in the IEEE 802 LAN standards. The number 802 was chosen to mark the caendar date when IEEE undertook the LAN standardization efforts (80 for 1980, 2 for February). Figure 3.2 shows the IEEE LAN standards in reationship to the OSI mode discussed in Chapter 2. You can see that the primary emphasis of the IEEE committee was to standardize the hardware technoogies used at the Physica and Data Link ayers. This is not surprising considering that networking hardware such as network interface cards and LAN wiring can be modeed competey by the two ower OSI ayers. The IEEE standards divide the OSI data ink ayer into two subayers: Media Access Contro (MAC) Logica Link Contro (LLC) FIGURE 3.2 Reationship of the IEEE 802 Standard to the OSI mode. (Courtesy IEEE Standard ) Media Access Contro (MAC) The MAC ayer deas with media access techniques utiized to contro access to a shared physica medium. Token Ring and Ethernet have different impementations of the MAC ayer because they use different methods to share the physica media. This aso means that if your TCP/IP network needs to ink Ethernet and Token Ring networks, you must use one of the "gue" devices such as a bridge or router between the Ethernet and Token Ring networks. Logica Link Contro (LLC) A IEEE LANs have the same LLC ayer as defined by standard The advantage of a common subayer such as the LLC is that upper-ayer mechanisms can be the same regardess of what kind of networking hardware you use. Figure 3.2 shows the interface between Upper Layer Protocos and the LLC ayer defined by Link fie://i:\chapters\z\zc648.htm 3/21/01

4 Inside TCP/IP, Third Edition -- Ch 3 -- Network Support for TCP/IP Page 4 of 47 Service Access Points (LSAPs). LSAPs are ogica data ink addresses. A singe MAC address, such as an Ethernet address, can have mutipe LSAP addresses. These mutipe addresses aow mutipe end-point connections between two nodes on a LAN. The LLC ayer is the upper subayer within the Data Link ayer. Reca from the previous chapter that the Data Link ayer is responsibe for transmission of data between two adjacent nodes on a network. These adjacent nodes have LSAP addresses that are caed the Destination Service Access Point (DSAP) and Source Service Access Point (SSAP) within the LLC ayer. The LLC ayer aso provides the options of virtua circuit (connections-oriented) or datagram (connectioness) services or a combination of these two. Datagram Services Datagram services are modeed after posta services. In the datagram approach, every packet contains compete addressing information such as destination and source addresses. No specia effort is made to ensure that packets arrive intact or in the correct order. Datagram services are connectioness; there is no attempt to create a connection before transmitting the data. Datagram services may or may not use acknowedgments. Acknowedgments are specia frames sent by the receiver that announce to the sender the frames that have been received correcty. They are used to retransmit frames that were not received correcty. This coud happen if there are errors in transmission because of physica hardware or interference on the ine. Chapter 2 contains a discussion on how the CRC mechanism is used to detect errors in frames. Unacknowedged datagram services means that datagrams are sent but no attempt is made at the Data Link ayer to retransmit frames that were incorrecty received. For appications that require a guarantee that packets were received correcty, this guarantee is deivered by the Transport ayer of the OSI mode (refer to Chapter 2). In TCP/IP networks, the Transport ayer is impemented by TCP. Unacknowedged datagram service for IEEE LANs is caed Type 1 service. The Internet Protoco, which impements ayer 3 of the OSI mode, aso does not use acknowedgments; that is, it is a connectioness service. This makes it easier to map the IP datagram into the frames used by LANs. Typicay, a singe IP datagram fits inside a LAN frame (see fig. 3.3). Where the size of the IP datagram is too arge to fit inside a singe LAN frame, the IP datagram is fragmented into sizes suitabe to fit inside a LAN frame, and each IP datagram fragment is sent inside a LAN frame (see fig. 3.4). You wi earn in ater chapters that there are mechanisms in the header of the IP datagram to keep track of the individua IP data fragments. FIGURE 3.3 An IP datagram is transported inside a LAN frame. FIGURE 3.4 IP datagram fragments are transported inside LAN frames. Virtua Circuit In a virtua-circuit approach a specia effort is made to ensure that packets arrive error-free in the order they were sent. Virtua circuits are modeed after the teephone system and require that a connection be estabished between two nodes before data can be exchanged between them. When data transfer is compete, this virtua circuit needs to be cosed or terminated. Virtua-circuit services are caed Type 2 services. In TCP/IP networks, virtua circuits are sedom used at the eve of a MAC ayer in LANs. This is because the TCP protoco is used to buid an end-to-end virtua circuit (see fig. fie://i:\chapters\z\zc648.htm 3/21/01

5 Inside TCP/IP, Third Edition -- Ch 3 -- Network Support for TCP/IP Page 5 of ). The difference between an end-to-end virtua circuit provided by a TCP and the Type 2 service of LANs is that the LAN virtua circuits are between two nodes on the same LAN or two compatibe LANs connected by a bridge. LAN virtua circuits cannot provide end-to-end virtua circuits across WAN inks of the type shown in figure 3.5. FIGURE 3.5 TCP virtua circuits versus LAN virtua circuits. Acknowedged Datagram Services Acknowedged datagram services, a combination of datagram and virtua circuits, are caed Type 3 services, in which an effort is made to correct data errors by retransmitting packets that have data errors. Type 3 services are used in IBM networks for transmitting SNA data over LANs. In summary, the types of services provided by LLC are as foows: Type 1. Unacknowedged, datagram service. Supports point-to-point, mutipoint, and broadcast transmission. Type 2. Virtua-circuit service. Provides sequenced, fow-controed, error-free services between LSAPs. Type 3. Acknowedged, datagram service. Provides datagram point-to-point service with acknowedgments. Figure 3.6 shows how the IEEE committee has identified the choices at the different ayers. FIGURE 3.6 Services defined by various IEEE 802 standards. Each of the choices represents a standard protoco or specification. Their IEEE numbers and meaning are described in tabe 3.1. TABLE 3.1 IEEE Standards IEEE Standard Meaning IEEE LAN bridging IEEE Logica Link Contro (LLC) IEEE Standardization of Ethernet Technoogy IEEE Token Bus standard IEEE Token Ring standard IEEE Metropoitan Area Network (MAN) IEEE Broadband technica advisory IEEE Fiber optic technica advisory IEEE Integrated Voice/Data (IVD) IEEE LAN security IEEE Wireess LANs IEEE VG-AnyLAN fie://i:\chapters\z\zc648.htm 3/21/01

6 Inside TCP/IP, Third Edition -- Ch 3 -- Network Support for TCP/IP Page 6 of 47 LAN Wiring Rues in TCP/IP Networks Each LAN standard has its own rues for LAN wiring. These rues define the connecting media, the hardware requirements, and the way the various components are arranged. Two primary concerns exist with regard to media: Arrangement of cabes Media type (usuay some type of cabe) Arrangement The geometrica arrangement of the wiring scheme is caed the topoogy. The topoogies that are common in LANs are the foowing: Star Bus Ring These are shown in figure 3.7. FIGURE 3.7 LAN topoogies. The Star Topoogy In the star topoogy, communication between any two nodes must go through a centra device or switching eement. The devices that connect to the centra switch tend to be simpe, with a of the compexity residing in the centra switch. The centra switching eement shoud be reiabe and provide signa isoation between ports so that faiures at any one port are not propagated to other ports. Cassic exampes of star topoogy are mainframe and minicomputer architectures in which the host is the centra switch. If the host breaks down, communication between the nodes in the network is broken aso. This points out the vunerabiity of the star wiring topoogy: it is vunerabe to a singe point of faiure. If, on the other hand, the centra switching eement both is reiabe and provides signa isoation between the ports, the star topoogy is one of the best topoogies. This expains why it is used in Token Ring, FDDI, and 10BASE-T LANs. Another advantage is that it is easy to connect or remove stations from a centra ocation. In many LANs, these centra eements (hubs) come with advanced network management features ike Simpe Network Management Protoco (SNMP). The Bus Topoogy The bus topoogy consists of a inear cabe to which stations are attached. Signas sent by a station on fie://i:\chapters\z\zc648.htm 3/21/01

7 Inside TCP/IP, Third Edition -- Ch 3 -- Network Support for TCP/IP Page 7 of 47 the bus propagate in both directions. Every transmission is avaiabe to every station on the network more or ess simutaneousy. A cassic exampe of a bus topoogy is Ethernet. The Ring Topoogy The ring topoogy consists of a cabe in the form of a oop with stations attached to it. Signas are sent in ony one direction, and the ring can be impemented by point-to-point simpex (one direction fow) inks. Stations see ony the transmissions that happen to pass by them in the ring. An exampe of a network that uses the ring topoogy is a Token Ring LAN. An important distinction shoud be noted between physica and ogica network topoogies. The physica topoogy of a network describes the way in which the actua cabes are aid out. The ogica topoogy describes the way that the network behaves. You can see and touch the physica topoogy of the network. The person who instas the cabe and hardware sees the physica topoogy. You cannot see the ogica topoogy; it is the network from the perspective of how data is sent through the network. Token Ring is described as a ring topoogy because data is passed from station to station unti it returns to the starting point. Data behaves as though it traves around a ring. Token Ring networks aways are wired with an individua cabe that extends from a centra wiring hub to each workstation. Because the wiring system ooks ike a star, Token Ring has a star ogica topoogy. Simiary, Ethernet aways has a ogica bus topoogy even when it is wired in a star using the new and popuar 10BASE-T system. Media Choices This section briefy reviews common LAN media choices in TCP/IP networks, such as coaxia, twisted pair, and fiber optic, as seen in figure 3.8. Coaxia Cabe Coaxia (coax) cabe consists of an inner conductor (usuay made of a copper aoy) that is used for sending a signa. The return signa fows through the shied that is separated from the centra conductor by a dieectric (eectricay insuating materia). The shied provides good bandwidth capabiities and eectrica noise immunity. This cabe type is the "granddaddy" of LAN media because some of the eariest LANs were buit using it. Coax cabes are typicay found in bus LAN topoogies. FIGURE 3.8 Twisted pair, coaxia, and fiber optic cabes. Twisted Pair Wiring Twisted pair (TP) wiring consists of a pair of wires wrapped around each other. These wires are twisted to minimize radiation and eectrica interference. Twisted pair wiring can have a shied around it to improve its data transmission quaity. Both shieded twisted pair (STP) and unshieded twisted pair (UTP) wiring are avaiabe for LANs. One wire is used for sending the signa, and the other wire acts as a signa return. Twisted pair wiring is cheap and easy to insta. Many buidings are aready pre-wired with data-grade twisted pair wiring. fie://i:\chapters\z\zc648.htm 3/21/01

8 Inside TCP/IP, Third Edition -- Ch 3 -- Network Support for TCP/IP Page 8 of 47 Shieded cabes surround the center conductors with a jacket of fine, braided wires. The shied heps prevent outside eectrica interference from affecting the conductors and aso reduces the risk of broadcasting signas interfering with nearby eectronic devices. Athough shieded cabes once were required for neary a oca area network instaations, today, use of unshieded twisted pair wire is more popuar. Unshieded twisted pair wire for LANs is simiar to that used for teephone communications. Networks can often use teephone wire that is aready instaed. Fiber Optic Cabe Fiber optic cabes consist of a strand of fiber materia, usuay gass, inside a protective jacket. Signas are transmitted in the form of refected ight puses. Signas can propagate over ong distances before they need ampification (provided by repeaters). Fiber optic has the best noise immunity characteristics compared to other wiring, is secure because it cannot be tapped easiy, and has the best bandwidth characteristics. What is noise immunity? If the signas representing data being transmitted in the media are unaffected by eectrica interference, caed noise, then the media has good noise immunity. The sources of eectrica interference or noise coud be wireess devices, fuorescent ights, eectrica motors/generators, etc. Fiber optic cabes use ight signas, which are unaffected by eectrica noise; therefore fiber optic cabes have good noise immunity. Today s high-speed LANs use fiber optic media. The end-component costs for fiber optic cabes and the required connecting equipment, however, are higher than twisted pair and coaxia cabes. Fiber optic cabes, therefore, are most commony used for high-speed connections or in situations requiring ong cabes or better immunity to eectrica interference. Ethernet LAN Operation in TCP/IP Networks Robert Metcafe, aong with David Boggs and others who worked for Xerox Corporation, deveoped a LAN based on carrier sensing mechanisms. This LAN spanned a distance of one kiometer, supported 100 persona stations, and achieved data rates of 2.94 Mbps. This system was caed Ethernet in honor of that eusive substance caed ether through which eectromagnetic radiation was once thought to propagate. Ethernet was proposed as a standard by Digita Equipment Corporation, Inte, and Xerox. The first Ethernet standard was pubished in September 1981 and was caed the DIX 1.0. DIX stands for Digita (DEC), Inte, and Xerox. DIX 1.0 was foowed by DIX 2.0 pubished in November The DIX 2.0 standard is aso caed Ethernet II. Meanwhie, Project 802 from the IEEE had undertaken LAN standardization efforts. Not surprisingy, Digita, Inte, and Xerox proposed the adoption of Ethernet as a standard. IBM proposed the Token Ring as a standard, based on prototypes buit at IBM s Zurich ab. The Ethernet proposa became known as the IEEE and the Token Ring proposa became the IEEE True to the nature of committee design, the IEEE standard is not quite the same as the Ethernet standard; there are important differences. Athough and Ethernet are incompatibe standards, the term Ethernet is used in TCP/IP LANs to designate compiant networks. This book bows to common usage and uses the term Ethernet for both standards, making distinctions as required when a fie://i:\chapters\z\zc648.htm 3/21/01

9 Inside TCP/IP, Third Edition -- Ch 3 -- Network Support for TCP/IP Page 9 of 47 specific standard is discussed. Ethernet is aso known by other names. In 1982, the European Computer Manufacturers Association (ECMA) adopted it as the ECMA 80/82/82 standard. The U.S. Federa Government adopted "Ethernet" in the FIPS PUBS 107 pubication, in In 1989, the Internationa Organization of Standards and Internationa Eectrotechnica Commission adopted it as the ISO/IEC standard. The foowing sections discuss Ethernet as it reates to LAN operation in TCP/IP networks in greater detai. Ethernet Operation Before an Ethernet station transmits, it istens for activity on the transmission channe (see fig. 3.9). Ethernet frequenty is described as a "isten before taking" protoco. Activity is any transmission caused by other Ethernet stations. The presence of a transmission is caed a carrier. The station eectronics can sense the presence of a carrier. FIGURE 3.9 Carrier-sense mechanism in Ethernet. If a station detects a busy channe, the station refrains from transmission. After the ast bit of the passing frame, the Ethernet data ink ayer continues to wait for a minimum of 9.6 microseconds to provide proper interframe spacing. At the end of this time, if a data frame is waiting for transmission, and the channe is free, transmission is initiated. If the station has no data to transmit, it resumes the carrier sense (istening for a carrier) operation. The interframe gap provides recovery time for other Ethernet stations. If a station tried to transmit when the channe is busy, a garbed transmission woud resut. Garbed transmissions are caed coisions. If the channe is free (no carrier detected), the station is free to transmit. Because mutipe stations attached to the Ethernet channe use the carrier-sense mechanism, it is caed a Carrier Sense with Mutipe Access (CSMA). What if two stations decide to transmit at the same time and there was no activity on the channe? A coision woud occur. Coisions occur during the norma operation of Ethernet LANs because stations transmit based ony on one fact: the presence of a carrier on the channe. They do not know if packets are queued for transmission on other stations. Furthermore, the CSMA operation is compicated by the fact of propagation deay in LANs. In Ethernet, for exampe, signas propagate at 0.77 times the speed of ight for standard (thick) cabes and 0.65 times the speed of ight on thin Ethernet cabes. A deay occurs before a transmission is heard by a stations, and a station may transmit because it has yet to hear another station's transmission. Coisions are a fact of ife in Ethernet LANs. Ethernet stations minimize the effects of coision by detecting the coisions as they occur. Hence the name CSMA/CD to describe the Ethernet media access mechanism (CD stands for Coision Detect). The stations invoved in the coision abort their transmissions. The first station to detect the coision sends out a specia jamming puse to aert a stations that a coision has taken pace. After a coision occurs, a stations set up a random interva timer. Transmission takes pace ony after this interva timer expires. Introducing a deay before transmission can reduce the probabiity of coisions. fie://i:\chapters\z\zc648.htm 3/21/01

10 Inside TCP/IP, Third Edition -- Ch 3 -- Network Support for TCP/IP Page 10 of 47 What happens when successive coisions occur? The average random time out vaue is doubed. This doubing takes pace up to 10 consecutive coisions. Beyond that, doubing the average random time out vaue does not improve the performance of the network significanty. This mechanism is caed the truncated binary exponentia back-off agorithm. How ong does a station have to wait under heavy oad conditions to transmit a frame? A station may experience a string of "bad uck" during which every time it transmits some other station has the bus. When coisions occur, stations introduce a deay using the random timer. But what if a station has the misfortune of timing out after the other stations have aready timed out? Under the worst-case scenario, a station may have to wait indefinitey. This is not acceptabe for rea time appications. Hence Ethernet is not suited for rea-time appications. Because Ethernet technoogy is simpe, cheap, and easy to impement, many rea-time networks used in factory automation use Ethernet with a modified CSMA/CD mechanism to guarantee network avaiabiity. This modified Ethernet is of course no onger the standard Ethernet discussed in this book. The next section examines different Ethernet options. Ethernet Cabe Options Coaxia cabe serves as the medium for two variations of Ethernet: the Standard Ethernet and the Thin Ethernet. Ethernet can aso can run on UTP wiring. These options are shown in figure FIGURE 3.10 IEEE options for Ethernet. Standard Ethernet Wiring Rues Another name for Standard Ethernet is Thick Wire Ethernet because the coaxia cabe it uses is much thicker than that used for Thin Wire Ethernet. The cabe type used for Thick Wire Ethernet is RG-8. The IEEE version of standard Ethernet is caed 10BASE5. The 10 stands for 10 Mbps operation; the BASE stands for baseband operation; and the 5 stands for 500 meters per segment. Figure 3.11 shows some standard Ethernet components. The network board shown in this figure has an AUI connector socket and a coaxia connection. The coaxia connection is used to connect to Thin Wire Ethernet. This particuar card can be used with both Thick/Thin Wire Ethernet. Stations on Thick Ethernet communicate to the externa network through externa transceivers attached to the shared media. The shared media is caed the trunk segment cabe or just the segment. Because of signa attenuation, a segment cannot be onger than 500 meters. The externa transceiver and the Network board are connected by a transceiver cabe. The DIX connector pug mates with the DIX connector socket on the Network board. FIGURE 3.11 Thick Ethernet network cabe and hardware. A side ock is used to secure this connection. The other end of the transceiver fits into a connector on the externa transceiver. Figure 3.12 shows the Thick Ethernet cabe used to make up the trunk segments. Thick Ethernet cabe fie://i:\chapters\z\zc648.htm 3/21/01

11 Inside TCP/IP, Third Edition -- Ch 3 -- Network Support for TCP/IP Page 11 of 47 is a 0.4-inch diameter, 50-ohm cabe and is avaiabe in various precut engths with an N-series connector pug attached to each end. You aso can purchase Thick Ethernet cabe in spoos or buk quantities. These come without the N-series connectors attached at the ends. Figure 3.12 aso shows the N-series barre connector that can be used to join two engths of Ethernet cabe. A trunk segment must be terminated with an N-series terminator. The N-series terminator is a 50-ohm resistor that bocks eectrica interference on the segment. Additionay, it cances out any signa refections caused by signas reaching the end of the cabe. The N-series terminator is attached to the mae N-series terminator on the end of the segment. N-series terminators come with a grounding wire. One end of the cabe must be grounded; the other end must remain ungrounded to avoid ground-oop currents. Figure 3.13 shows an exampe of a Thick Ethernet network. Two trunk segments are joined together by a device caed a repeater in Thick Ethernet networks. A repeater is an active device that aows an Ethernet LAN to expand beyond a singe segment by inking two segments together. The repeater ampifies and regenerates the signa so the signa can be transmitted over onger distances. A mutiport repeater such as a DEMPR (Digita Equipment s mutiport repeater) can ink a number of Ethernet segments together. FIGURE 3.12 Thick Ethernet coaxia cabe connectors. FIGURE 3.13 Exampe of a Thick Ethernet network. Tabe 3.2 describes the rues you must foow with Thick Ethernet wiring. TABLE 3.2 Thick Ethernet Parameters and Wiring Rues Thick Ethernet Parameters Vaue Maximum data rate 10 Mbps Maximum repeaters without IRLs 2 Maximum repeaters with IRLs 4 Maximum coaxia segment ength 500 meters Maximum transceiver cabe ength 50 meters Maximum number of ink segments 2 Maximum combined ink segment ength 1000 meters Maximum stations per segment 100 Maximum number of stations 1024 Distance between stations Mutipes of 2.5 meters To trave from one station to another station on an Ethernet LAN that consists of coaxia trunk segments ony (see fig. 3.14), a signa cannot trave through more than two fu repeaters. A fu repeater joins two coaxia segments together directy. A coaxia segment is distinct from a ink segment. A ink segment made of fiber optic or twisted pair cabe can be used to join two coaxia segments over a onger distance. The purpose of a ink segment is to extend the range of an Ethernet LAN. You can have a maximum of two ink segments on an Ethernet LAN. Link segments do not have stations attached to them and are connected to coaxia segments by repeaters. Another name for fie://i:\chapters\z\zc648.htm 3/21/01

12 Inside TCP/IP, Third Edition -- Ch 3 -- Network Support for TCP/IP Page 12 of 47 them is Inter-Repeater Link-segments (IRLs). A haf-repeater joins a coaxia segment to a ink segment. Another name for a haf-repeater is a remote repeater. The trunk coaxia segment ength cannot exceed 500 meters. The combined engths of the two ink segments cannot exceed 1000 meters. Using these wiring parameters, you can deduce the maximum ength of an Ethernet LAN. FIGURE 3.14 Longest Thick Ethernet network formed by using fu repeaters ony. Figure 3.15 iustrates the ongest possibe Ethernet. T1 through T6 represent transceivers. Using this diagram, you can cacuate the ength of this network: Coax Segment 1 ength = 500 meters Coax Segment 2 ength = 500 meters Coax Segment 3 ength = 500 meters Combined Link Segment 1 and 2 ength = 1000 meters Tota Ethernet Length = 2500 meters Some peope take advantage of the considerabe ength of transceiver cabes to extend the range of the LAN even farther by adding the transceiver cabe engths to transceivers T1, T2, T3, T4, T5, and T6 in figure Because the maximum transceiver cabe ength is 50 meters, this gives a combined transceiver ength of 300 meters. The foowing cacuations show how to arrive at this number: FIGURE 3.15 Longest Ethernet possibe. Transceiver cabe ength of transceiver T1 50 meters Transceiver cabe ength of transceiver T2 50 meters Transceiver cabe ength of transceiver T3 50 meters Transceiver cabe ength of transceiver T4 50 meters Transceiver cabe ength of transceiver T5 50 meters Transceiver cabe ength of transceiver T6 50 meters Combined transceiver cabe ength 300 meters By using maximum transceiver engths, the maximum Ethernet ength is 2500 meters pus 300 meters, which is 2800 meters. The maximum number of stations that you can attach to a Thick Ethernet segment is 100, and the fie://i:\chapters\z\zc648.htm 3/21/01

13 Inside TCP/IP, Third Edition -- Ch 3 -- Network Support for TCP/IP Page 13 of 47 tota number of stations cannot exceed The repeater attachment to a segment counts as one station. The minimum distance between any two stations is 2.5 meters. It is recommended that you separate stations at distances of mutipes of 2.5 meters to minimize interference caused by standing waves on an Ethernet segment. Standing waves are formed by the presence of eectrica signas on the segment. Thin Wire Ethernet Wiring Design Rues Other names for Thin Wire Ethernet are Thinnet and aso Cheapernet (because it is cheaper than Standard Ethernet). The coaxia cabe it uses is much thinner than that used for Thick Wire Ethernet. The IEEE version of Thin Wire Ethernet is caed 10BASE2. The 10 stands for 10 Mbps operation; the BASE stands for baseband operation; and the 2 stands for approximatey 200 meters (actuay, 185 meters) per segment. Figure 3.16 shows some of the Thin Wire Ethernet components. The network board or Network board, in this figure, has a coaxia connection. FIGURE 3.16 Thin Ethernet components. The transceiver functions for a Thin Wire Ethernet are performed by the on-board network eectronics. No externa transceiver connections are made to the network. BNC T-connectors are used to connect the Network board with the cabe. The two opposing jacks of the T-connector are used to join two engths of Thin Wire Ethernet cabe. The remaining pug is attached to the BNC connector jack on the Network board. Due to signa attenuation, a thin wire segment cannot be onger than 185 meters. Thin Ethernet cabe has a 0.2-inch diameter and RG-58 A/U 50-ohm cabe, and is avaiabe in various precut engths with a standard BNC pug attached to each end. Thin Ethernet cabe aso can be purchased in spoos or buk quantities that come without the BNC connectors attached at the ends. Using specia crimping toos you can construct cabes of arbitrary engths. You aso can use the BNC barre connector to join two engths of Ethernet cabe. A trunk segment must be terminated with a BNC terminator. The BNC terminator is a 50-ohm resistor that bocks eectrica interference on the segment. Additionay, it cances out any signa refections caused by signas bouncing off the end of the cabe. The BNC terminator is attached to one of the two jacks on a T-connector to which no cabe is attached. There is a grounded BNC terminator that has a grounding wire. One end of the cabe must be grounded; the other end must remain ungrounded to avoid ground oop current. Figure 3.17 shows an exampe of a Thin Ethernet network. In this network, there are two trunk segments that are joined together by a repeater. The repeater in figure 3.17 has two ports to attach a maximum of two segments. FIGURE 3.17 Exampe of a Thin Ethernet network. There are a number of rues reated to Thin Ethernet wiring. These are summarized in tabe 3.3. TABLE 3.3 Thin Ethernet Parameters and Wiring Rues fie://i:\chapters\z\zc648.htm 3/21/01

14 Inside TCP/IP, Third Edition -- Ch 3 -- Network Support for TCP/IP Page 14 of 47 Thin Ethernet Parameters Vaue Maximum data rate 10 Mbps Maximum repeaters without IRLs 2 Maximum repeaters with IRLs 4 Maximum coaxia segment ength 185 meters Maximum number of ink segments 2 Maximum stations per segment 30 Maximum number of stations 1024 Minimum distance between stations 0.5 meters The repeater rues for Thin Ethernet are the same as for Thick Ethernet. The trunk coaxia segment ength for Thin Ethernet cannot exceed 185 meters. The maximum number of stations that can be attached to a Thin Ethernet segment is 30, and the tota number of stations cannot exceed The repeater attachment to a segment counts as one station. The minimum distance between any two stations is 0.5 meters. 10BASE-T Wiring Design Rues An increase in interest for 10BASE-T began in 1990 due to the ower cost components and ease of configuring networks based in UTP wiring. 10BASE-T is a technoogy that is used to impement Ethernet. 10BASE-T is an impementation option for the IEEE standard. The 10 stands for 10 Mbps operation; the BASE stands for baseband operation; and the T stands for twisted pair wiring. In figure 3.18, the Network board has a teephone-type RJ-45 port, which is officiay caed a Media Dependent Interface (MDI) port. The Network board shown in the figure aso has a DIX connector. The DIX connector is used to connect by means of a transceiver to Thick Wire Ethernet. This particuar card can be used with both 10BASE-T and Thick Ethernet. Many network boards require a switch setting to enabe either the 10BASE-T or DIX port, whereas others have an auto-sense mechanism. The transceiver functions for a 10BASE-T are performed by the onboard network board eectronics. The 10BASE-T uses a physica star topoogy with the 10BASE-T concentrator (aso caed hub) serving as the centra switching eement. The 10BASE-T pug and connector are shown in figure Each concentrator accepts cabes to severa workstations. UTP wiring is used to connect a 10BASE-T concentrator to the workstation. This wiring normay consists of 0.4 to 0.6 mm diameter (26 to 22 AWG) unshieded wire in a mutipair cabe. The performance specifications are generay met by 100 meters of 0.5 mm teephone twisted pair. FIGURE 3.18 Connectors on a 10BASE-T network adapter card. FIGURE BASE-T pug and connector. (Source: IEEE Standard 802.3i-1990). There are two twisted pairs (four wires) between each network board and the concentrator, as shown in figure Each two-wire path forms a simpex ink segment. One simpex segment is used for transmitting and the other for receiving. fie://i:\chapters\z\zc648.htm 3/21/01

15 Inside TCP/IP, Third Edition -- Ch 3 -- Network Support for TCP/IP Page 15 of 47 FIGURE 3.20 Simpex segments used in 10BASE-T. (Source: IEEE Standard 802.3I-1990). Tabe 3.4 shows the pin assignments for a 4-pair twisted pair wiring. Ony two pairs, one for transmission (TD) and another for receiving (RD), are used. TABLE 3.4 Pin Assignments for the MDI Connector Contact MDI signa 1 TD+ 2 TD_ 3 RD+ 4 Not used by 10BASE-T 5 Not used by 10BASE-T 6 RD_ 7 Not used by 10BASE-T 8 Not used by 10BASE-T A crossover function is impemented in every twisted pair ink so that the transmitter at one end wi be connected to the receiver at the other. Figure 3.21 shows the two ways of impementing crossover functions. One way to do this is to use an externa crossover UTP cabe that reverses the transmit and receive pairs at the RJ-45 connector at one end of the UTP cabe. A second way is an interna crossover function, in which the crossover is designed as part of the interna circuitry in the 10BASE- T device. An MDI port with this function is marked with the symbo "X." FIGURE BASE-T crossover wiring. Figure 3.22 shows a singe concentrator 10BASE-T network. FIGURE 3.22 A singe concentrator 10BASE-T network. The concentrator has 12 RJ-45 ports. If the station's Network board has a 10BASE-T (RJ-45) connector, the connections can be made directy through UTP cabe. For stations with 10BASE5 network boards, a 10BASE-T MAU (10BASE-T transceiver) is needed to connect the AUI cabe to the station. The 10BASE-T concentrator serves the roe of a repeater. It performs the foowing functions: Data packet retiming (IEEE standard) Per-port Link Integrity Test ("Good Link Test") Per-port autopartitioning, which disconnects the port in the event of 30 consecutive coisions, an excessivey ong singe coision, or jabber input The proper operation of the CSMA/CD 10BASE-T network requires network size to be imited to fie://i:\chapters\z\zc648.htm 3/21/01

16 Inside TCP/IP, Third Edition -- Ch 3 -- Network Support for TCP/IP Page 16 of 47 contro round-trip propagation deays (the time it takes a signa to reach extremity of network and come back). The configuration rues for more than one concentrator are as foows: Maximum of four concentrators in the data path between any two stations UTP segments shoud be no onger than 100 meters UTP Wiring Considerations Use of UTP (unshieded twisted pair) wiring for data communications has come a ong way from its initia use of transmitting anaog voice signas to its use in 10BASE-T and CDDI (Copper Distributed Data Interface), which is a variation of the FDDI LAN that runs at 100 Mbps. UTP can aso be used for a 100 Mbps version of Ethernet caed 100BASE-T. Athough using UTP for LAN wiring needs can simpify instaation and reduce wiring costs, it can, if not done propery, do just the opposite: compicate instaations and increase maintenance costs. The factors to consider for an effective UTP instaation are discussed in the upcoming paragraphs. The ack of a shied in UTP makes it cheaper than other types of wiring and aso easier to insta because it is more fexibe than the shieded twisted pair wiring. However, because it is unshieded, UTP can become a good antenna and susceptibe to Eectro-Magnetic Interference (EMI) and Radio- Frequency Interference (RFI). At such frequencies as 10 to 100 Mbps, UTP wiring resuts in oss of signa due to attenuation. Inductance and capacitance effects aso become dominant at these high frequencies. The inductance is caused by the eectromagnetic fied that surrounds the UTP wire when the highfrequency signas pass through it. It can be ikened to the transformer effect that induces a votage on the secondary of the transformer due to eectromagnetic couping. The capacitance effect is caused because the conductors that make up the twisted pair wire are separated by an insuating materia. These effects reduce the quaity of the signa and imit the distance that can be used between devices connected by UTP. The twists that are used in twisted pair wiring hep reduce inductance by creating a magnetic fied that essentiay cances out inductance. For this reason, an important parameter in measuring the quaity of a cabe is the twists per inch of the wire. This can reduce the amount of cross-tak, which is the inductive couping to other pairs of wires or noise sources. Cross-tak can ead to signa distortion (often caed jitter) and, in the case of Ethernet networks, can be mistaken for coisions, which coud degrade the network performance. In Token Ring networks, cross-tak can generate hard errors that can cause the Token Ring networks to go through reconfigurations. Reconfigurations are time-consuming and resut in sow networks. Fat siver satin wire, which works just fine in ow-speed data networks such as 19.2 Kbps, has zero twists per inch. This type of cabing is common in teephone networks; however, if voice teephone cabe is used in data networks that operate in the Mbps range, it can resut in disaster. Besides causing the network to fai, it can create a great dea of EMI noise that can cause other devices to fai aso. The signa that is used in both Ethernet and Token Ring networks is a baseband signa. Baseband signas are digita signas that have sharp edges. The capacitance effect in a wire causes the signa to fie://i:\chapters\z\zc648.htm 3/21/01

17 Inside TCP/IP, Third Edition -- Ch 3 -- Network Support for TCP/IP Page 17 of 47 ose some of its sharpness so that it becomes rounded. The resistance effect causes the signa to ose its strength (attenuation). The inductance and capacitance effect can make the signa vunerabe to externa noise sources to the point that the signa can be competey distorted. Figure 3.23 iustrates these effects. FIGURE 3.23 Signa distortion and noise in cabes. Another factor to consider is that signas with sharp edges or rapidy changing signas resut in highfrequency harmonics. Mathematicay speaking, the ds/dt--the rate at which the signa changes--is high for the edges of the baseband signa. If the signa is periodic, it can be expressed as the sum of sine wave harmonics of the fundamenta frequency of the signa, in which the sine wave may have different phase (starting point) differences. This means that a 20 Mbps signa is reay not just a 20 Mbps signa, but a sine wave with 20 MHz fundamenta frequency and harmonic components of 40 MHz, 60 MHz, 80 MHz, 100 MHz, 120 MHz, and so on. The higher harmonics are smaer in magnitude. What this means in practica terms is that the cabe must be abe to carry the higher harmonic components of the data signa. If it does not do this we, the signa can become distorted. Some of these factors can be mitigated by using high-quaity twisted pair wiring. The characteristics of cabes can be defined in terms of attenuation, which is measured in decibes per 100 feet. Decibes is a ogarithmic scae (to the base 10) for comparing power eves. It is defined as og (P2/P1) where P2 = Power at output P1 = Power appied at input LAN designers specify maximum distance engths for cabe segments based on attenuation characteristics of the cabe medium for the frequency of data transmission. For this reason, 10BASE- T networks have a imit of 100 meters between station and wiring concentrator. The type of sheath used to encose the twisted pair wire affects its penum rating. Penum rating determines whether or not the cabe must be encased in a conduit for fire resistance as required by some buiding codes. PVC (poyviny choride) is the most common coating used and is not fire resistant. Another type of cabe coating caed TFEP (tefon fuorinated ethyene propyene) is rated as fire resistant. It aso has a ower dieectric constant. The ower the dieectric constant, the ower the capacitance, and, therefore, the ower the signa distortion. Because of these characteristics, TFEPcoated wire can transmit the signa over onger distances with ess signa distortion compared to PVC-coated wires. In teephone networks, it is common to use a punch-down bock caed the 66 Bock. Whereas this type of punch-down bock works fine for teephone networks, it is not designed to carry data. For data networks that need to carry data in the Mbps range, you must use punch-down bocks specificay designed for data. These incude punch-down bocks known as the 110s, 3m 7000D, and Krone. Data grade punch-down bocks incude god-pated or siver-pated contact points, abeing, and so forth. Data-grade patch panes aso are avaiabe. Some of their features are cross-connect circuits etched on the wafer board itsef. These patch panes can carry high-frequency signas in a manner simiar to fie://i:\chapters\z\zc648.htm 3/21/01

18 Inside TCP/IP, Third Edition -- Ch 3 -- Network Support for TCP/IP Page 18 of 47 printed circuit boards. Mixed Media Ethernet Networks You can combine the different media (coaxia, twisted pair, and fiber) into one Ethernet LAN. If you combine mixed media networks, use a fiber optic, twisted pair, or coaxia cabe to impement the ink segment. Figures 3.24 and 3.25 show exampes of mixed media networks. FIGURE 3.24 An network using fiber optic cabe. FIGURE 3.25 An network with a coaxia backbone. Tabe 3.5 summarizes the maximum deays of the various media segments. This tabe is important for the LAN manager because Ethernet segments can be buit by combining cabes from different vendors, each of which may differ from the specifications by sma amounts. Test equipment ike Time Domain Refectometers (TDRs) can be used to see that the deays are within the specifications. TABLE 3.5 Maximum Deays for Ethernet Media Media Segment Type Maximum MAUs per Segment Maximum Segment Length (meters) Minimum Propagation Veocity (fraction of speed of ight) Maximum Deay per Segment (nanoseconds) Coaxia Segment 10BASE c BASE c 950 Link Segment FOIRL c BASE-T c 1000 AUI (Transceiver Cabe) AUI DTE/1 MAU c 257 c = 300,000,000 meters/second (veocity of ight in vacuum) FOIRL = Fiber Optic Inter Repeater Link The foowing network topoogy rues appy for mixed media networks: 1. Repeater sets are required for a segment interconnections. 2. The maximum transmission path between any two stations may consist of up to five segments, four repeater sets (incuding optiona AUIs), and two MAUs. 3. If a network path consists of five segments and four repeaters sets, up to three segments may fie://i:\chapters\z\zc648.htm 3/21/01

19 Inside TCP/IP, Third Edition -- Ch 3 -- Network Support for TCP/IP Page 19 of 47 be coaxia trunks and the remainder must be ink segments. If five segments are present and Fiber Optic Inter-Repeater Link (FOIRL) is used as the ink segment, the ink segment shoud not exceed 500 meters. 4. If a network path consists of four segments and three repeater sets, the maximum aowabe ength of the FOIRL segments is 1000 meters each. Rue 2 is iustrated in figure Notice that this rue does not te us how many segments are coaxia trunks with mutipe station attachments and how many are ink segments with no station attachments. Rue 3 carifies this probem. FIGURE 3.26 Maximum Ethernet transmission path with three coaxia segments. (Source: IEEE Std 802.3I-1990) Rue 3 is iustrated in figure This figure shows a mutimedia Ethernet network. The media used in this network is a combination of coaxia, fiber, and twisted pair. Notice in figure 3.27 that there are five repeater sets. This may at first gance seem to contradict the rue of a maximum of four repeater sets, but between any two stations, there are no more than four repeaters in the transmission path. There is a tota of ten segments: seven twisted pair, two fiber optic, and one coaxia. However, there are no more than five segments between any two stations. Aso, there is a maximum of one coaxia segment, which is within the maximum three coaxia segment rue. When the coaxia segment is incuded in the transmission path, the remaining four segments are ink segments: three twisted pair and one fiber optic ink segment. Because there is a maximum of five segments and four repeaters in the transmission path, the maximum FOIRL ength is 500 meters. This foows from Rue 3. The maximum span of this network is 1300 meters, not incuding AUI drops. Rue 4 is iustrated in figure There are three repeater sets and six segments: four twisted pair and two fiber optic. There are no coaxia segments in this figure. Between any two stations, there is a maximum of four segments and three repeaters. The four segments consist of two fiber optic and two twisted pair cabes. Each of the FOIRL inks has a maximum ength of 1000 meters. The maximum span of this network is 2200 meters, not incuding AUI drops. FIGURE 3.27 Maximum transmission path using coaxia segments, 10BASE-T ink segments, and fiber optic ink segments. (Source: IEEE Standard 802.3i-1990) If you are extending an existing network and pan to use a different cabe type, you can do so by using the methods discussed in the foowing sections. Combining Thin/Thick Cabe in a Segment You can combine thin and thick Ethernet cabe in a singe segment by using as much thin cabe as possibe. Thin cabe is cheaper and easier to insta than thick cabe. Figure 3.29 iustrates a network ayout using segments made up of a combination of thin and thick cabe. Combined thin/thick cabes are between 185 meters and 500 meters ong. The minimum ength is 185 meters because coaxia segments shorter than 185 meters can be buit with thin cabe excusivey. The maximum of 500 meters is the imit for a segment made out of thick coaxia excusivey. FIGURE 3.28 Maximum transmission path with three repeater sets and four ink segments. (Source: fie://i:\chapters\z\zc648.htm 3/21/01