Introduction to Data

Communications

In the minds of most people, data communication is a recent phenomenon

associated with computer communications, but data messaging predated the

telephone by several decades. Samuel F.B. Morse invented the first data communication

device, the telegraph, in 1844. The telegraph and its successor the

teletypewriter played an important role in communications for the first century of

their existence. These devices were the principal means of sending messages over

long distances until computers appeared.

Computers began as text-based systems that used punched cards for input

and magnetic tape for storage. At first, the mainframe computer was confined to

a self-contained processing center. Video display terminals (VDTs) gradually

replaced keypunch machines as the principal input device. VDTs are known as

“dumb” terminals because they have no processing power of their own. It all

resides in the host computer. Output devices of the time were punched cards,

tape, or high-speed line printers.

The methods of remote computer communications were primitive at first,

and the means of cramming information into voice channels was little understood.

One of the earliest applications involving computer communications was the U.S.

Air Force’s Semi-Automatic Ground Environment System (SAGES). SAGES integrated

data from radar stations into control centers that later formed the basis of

the air traffic control system. Development of SAGES began in 1957, a scant dozen

years after development of ENIAC, which was the ancestor of today’s computer

systems.

As computer applications evolved from centralized to distributed, enterprises

needed to move terminals to outlying offices to support applications such

as reservations and banking. Computer ports could easily be extended to distant

terminals with a modem running over a telephone circuit. Remote printers,

successors to the teletypewriter, had printing speeds in the order of 100 words

per minute, more than the speed of most typists. One hundred wpm is only about

80 bps, and a telephone circuit supports 2400 bps with the simplest of modems. It

was not cost-effective to use a fraction of the capacity of a voice circuit for a single

VDT. Moreover, 1 bit in error could render a message stream unusable. The industry

needed a reliable means of sharing circuits and detecting and correcting errors.

Two methods of solving this problem evolved. In one solution, IBM’s Binary

Synchronous Communications (BSC), a remote controller communicated with the

host computer. The host sent short polling messages to determine if any of the

VDTs attached to the controller had traffic to send. If so, the controller sent a block

of information with an error-detection block attached. If the host received the

block error-free, it returned an acknowledgment. If not, the controller resent the

block until the host acknowledged receipt. This method provided reliable communications,

but since the transmitting end could not send a block until the previous

one was acknowledged, it was slow. IBM upgraded BSC to SNA in later

years. SNA permits the sender to launch multiple packets without waiting for

acknowledgment. The service that supports BSC and SNA is called multidrop, a

name derived from the fact that multiple controllers share a single circuit back to

the host.

The second method, which most non-IBM computers used, was a data

multiplexer. A multiplexer connected to terminal ports on the host computer and

communicated with a matching unit that supported terminals at the distant end.

The multiplexer pair had two functions: one, they subdivided a voice channel into

multiple data streams, and two, they carried on an error detection and correction

dialogue to ensure data integrity.

Both of these methods are still used with variations, but only to a limited

extent because servers have replaced most mainframes and minicomputers. As

long as the keyboard remained the primary input device, a voice circuit was more

than adequate. Applications, however, evolved into file transfers and local area

network (LAN) interconnection, both of which involve computer-to-computer

communications. The mismatch in speed with which a hard disk can deliver information

and a modem can transmit it puts a severe damper on efficiency. LANs

move data within the office at much higher speeds. The earliest LANs were in the

order of 1 Mbps, but these soon increased to 10 and 100 Mbps. As organizations

interconnected their LANs and distributed information over wide areas, the bandwidth

of an analog circuit became a severe choke point. Users, having no idea

where information was stored, noted the vast difference in response time between

a local file delivered over a high-speed LAN and one delivered over a voice-grade

wide area network (WAN). Distributed applications needed a better match

between LAN and WAN speeds.

The common carriers’ first solution was to provide digital circuits. The basic

digital circuit in North America supports 56 Kbps of bandwidth. This is more

than twice as fast as an analog circuit, but a LAN is many times faster. The digital

56 Kbps circuit was barely a waypoint on the trek to higher bandwidths.

36 PART 1 Introduction

The carriers’ difficulty lay in the availability of transmission resources. Fiber

optics was in its infancy in 1980 and long-haul circuits were transported over analog

microwave radio. Digital microwave was available, but its error rate limited it

to a span of about 500 miles (800 km). The restrictions of the analog voice network

remained until fiber optics relieved the bottleneck.

Another major change in computer applications was the transition from textbased

information to graphics. Computers had been migrating toward graphics

for several years, but the Web accelerated the change. Desktop computers

replaced dumb terminals as the I/O device, and graphics-based operating systems

replaced command-driven systems. File sizes grew by orders of magnitude

and with the growth of desktop computing and the need for Internet access, network

managers had little choice but to expand bandwidths to meet the demand.

This brings us to the current state of data communications, which is mixed,

but moving toward a unified network. Desktop computers have replaced most of

the older VDTs. Multiplexers have largely been supplanted by devices that interconnect

LANs, and the older data networks have evolved into networks of interconnected

LANs. To begin the discussion of data networks, an understanding of

some key terminology is in order. We start the chapter with a discussion of

protocols, which data devices use to communicate with one another. Following

that, we discuss networks in overview, focusing on LANs, which are private data

networks that operate at high speed over a narrow range and WANs, which have

a global reach. Then we review the common carrier services that subscribers use

to interconnect their LANs so users can access resources from the desktop without

regard to where they are located.

PROTOCOL TERMINOLOGY AND FUNCTIONS

Protocols are specialized computer programs that either exist in software or in

firmware chips. Most desktop computers contain both varieties. The Ethernet

LAN protocol, which comes with all new computers, is encoded into firmware.

For users to access the Internet they must load the software protocol, Transport

Control Protocol/Internet Protocol (TCP/IP), which is provided with the computer’s

operating system.

Most protocols are written in modules or layers, each of which has defined

functions and interfaces. A modular structure allows developers to write application

software without concern about how the program operates over the wide

area. If a function or specification changes, it is not necessary to change the entire

protocol stack; only the affected layer and its interfaces with other layers are

changed. Many international standards are based on the OSI model, which is an

ISO structure that defines the functions of each layer.

LAN standards are a good example of the efficacy of a layered protocol.

LANs operate at the first two layers of the OSI model: physical and datalink.

Network interface cards (NICs) connect computers to the physical layer, which is

CHAPTER 3 Introduction to Data Communications 37

twisted pair wire, fiber optics, coaxial cable, or wireless. Network operating systems

use higher layer protocols that talk to any NIC. The NIC manufacturers provide

software drivers to enable the functions in their cards to communicate with

the network operating system across a protocol known as the logical link layer

(LLC), which is a subset of the datalink layer.

Protocol Functions

Diplomatic protocols prescribe a complex set of rules that dictate the way people

interact in international relationships. Diplomatic protocols prescribe such

customs as who is seated next to whom, how officials of different ranks are

addressed, and what kind of response is appropriate to another’s statement. The

protocol metaphor is apt for data because devices must behave according to

defined procedures. In a layered protocol these functions are usually assigned to

one layer, but the rules regarding this are not rigid. For example, every Ethernet

NIC has a unique and exclusive protocol and address known as media access control

(MAC) embedded in firmware. The MAC address identifies the station within

its own LAN segment, but to communicate across a WAN a higher level address

is required, the most prevalent of which is IP.

A fundamental purpose of protocols is to establish a session across a network.

A session begins when a user logs on to a distant computer, and ends with

log-off. The protocol authenticates the parties before permitting communication to

begin. Data networks handle sessions in two distinct ways: connectionless and

connection-oriented. In a connection-oriented protocol, the devices establish a physical

or logical connection across the network. The connection is set up at the start

of the session and remains for its duration. The connection can be circuit switched

as it is in the telephone network, or it can be a virtual connection that is defined in

a software path and shares bandwidth with other sessions. In a connection-oriented

session the packets contain a path identifier, but they do not need to carry

the address of either the sender or the receiver after the path is set up.

In a connectionless session, a single packet known as a datagram is launched

into the network and delivered to the distant end based on its address. The postal

service is an example of a connectionless operation. The user does not know or

care how a letter gets to its destination provided the post office delivers it to

the addressee intact and in a timely manner. LANs are connectionless. A station

launches a stream of information onto the network. All stations copy the message

but retain it only if they are the addressee. Connectionless operation requires that

each packet or frame contains the address of the sending and receiving stations.

The following points discuss the generic functions of protocols. The application

and the communications medium dictate whether a specific function is

needed. Within a LAN, which has a short and relatively secure transmission

medium, stations can communicate without many of the protocols that a session

needs across the more hostile environment of a WAN.

38 PART 1 Introduction

Communications control: Protocols can be classified as peer-to-peer

or master–slave. In the latter case, the master controls the functioning

of the datalink and controls data transfer between the host and its

endpoints. All communication between slave stations goes through

the master. Peer-to-peer protocols do not use a controller. Devices can

communicate directly with one another.

Error detection and correction: Protocols check for errors, acknowledge

correctly received data blocks, and send repeat requests when blocks

contain an error. Sophisticated protocols can acknowledge multiple

packets using one of two procedures. A selective repeat acknowledgment

enables the receiving device to request the sending end to resend

specific packets. In the go-back-n method the receiver instructs the

sender to resend an errored packet and all subsequent packets.

Link management: After the session is set up, the protocol controls the

traffic flow and data integrity across the datalink.

Setting session variables: The protocol determines such variables as

network login and authentication and whether the session will be

half-duplex, meaning information flows in one direction at a time or full

duplex, meaning information can flow in both directions simultaneously.

Synchronizing: At the start of a session data devices exchange signals to

determine such factors as the data transfer rate and whether they will

use compression or encryption. Modems exchange signals to determine

the highest speed at which they can exchange data and fall back to a

lower speed if the circuit will not support the maximum.

Addressing: Every session requires an address to set up a connection if the

protocol is connection oriented or to route packets if it is connectionless.

Not all protocols contain addresses. Many of them rely on higher or

lower layers for addressing.

Routing: In data networks that have multiple routes to the destination,

the protocol determines the appropriate route based on conditions such

as cost, congestion, distance, and type of facility.

Data segmenting and reassembly: The protocol segments a continuous data

stream from the source into frames, cells, or packets as appropriate.

The term protocol data unit (PDU) refers to the unit into which the data

stream is divided. The PDU is equipped with header and trailer records

for transmission over the network. Headers and trailers contain extra or

overhead bits that contain information the protocol uses for routing and

error checking. At the distant end, the protocol strips the overhead and

reassembles the data stream for delivery to the receiver.

Data formatting: The bit stream may require conditioning before

transmission and restoration after reception. For example, conditioning

could include encryption or compression.

CHAPTER 3 Introduction to Data Communications 39

Supervision: The protocol establishes a connection, determines how the

session will be started and ended, which end will control termination of

the session, and how billing will be handled.

Flow control: Protocols protect networks from congestion by sending

signals to the source to halt or limit traffic flow.

Failure recovery: If the session terminates unexpectedly, the protocol sets

markers and enables recovery without starting over.

Sequencing: If data blocks arrive out of their original sequence, the

protocol delivers them to the receiving device in the correct order.

DATA NETWORKS

Unlike voice sessions, which have predictable and uniform bandwidth requirements,

data sessions have unique requirements that depend on the characteristics

of the application. Users have many network alternatives to match the needs of

data sessions. Data sessions have much different characteristics than voice. For

one thing, they are lengthy. Contrasted to voice sessions, which typically average

3 or 4 min, data sessions extend from log-on to log-off, which may be days or

months apart. During the session the user may download a file, which puts a

momentary high demand on the network, but then suspend any further activity

while working on the file or attending to other business. Meanwhile, the network

makes its bandwidth available to other sessions that have noncoincident peaks.

Here are examples of typical data sessions:

Inquiry–response: Client-server applications such as Web browsing are

characterized by a small transmission upstream to a server, which may

result in a large downstream transmission.

Remote data entry: Access to a bank ATM machine is an example.

Transmissions are short, and may be approximately symmetrical, but

data flows in only one direction at a time.

Database backup: A computer may back up its hard disk to a storage area

network. Data flow is heavily weighted in the upstream direction.

LAN interconnection: This application has symmetrical data flow on the

average, but it is characterized by bursts of heavy flow.

The bursty nature of many data applications means that fixed-bandwidth

circuits impose limitations because the larger the file, the longer it takes to transfer

the data. Most data applications work best in an environment that provides a

variable amount of bandwidth, expanding to meet short-term needs, and during

low-usage periods, making it available to other sessions. Also, circuit setup time,

which is relatively insignificant to a voice session, is lengthy in data terms. Many

data applications need a fast response from the other end. If the devices had to

40 PART 1 Introduction

carry out a lengthy handshaking routine each time they needed access, the application

would be handicapped.

Network Access Methods

If multiple devices share the transmission medium, some means of allocating

access is needed. Circuit switching can be used, and for some applications such as

dial-up Internet access and occasional connections to remote access servers and

computer ports it is effective. For many data applications, however, the setup time

is too long, the bandwidth is too narrow, and dial-up access costs too much. Most

data applications need a full-time connection to the network.

One common method of allocating access is polling in which a device at the

head end of the network sends short messages to each node in turn, asking if it

has traffic to send. IBM’s BSC and SNA use this method.

A third method is contention access, which has the characteristics of a party

line. Devices listen to the network to see if it is busy and defer their transmissions

until it is idle. Ethernet uses this method. This method works well if all devices

can hear all of the transmissions on the network. If they cannot, however, transmissions

may collide, necessitating the devices to recognize the collision and

retransmit. Collisions may be reduced by permitting devices to transmit only during

their allocated time slots.

Another method of regulating access is by using an intelligent device such

as a router or switch to accept packets from the sources, buffer them briefly, and

release them in a disciplined fashion that does not exceed the capacity of the

transmission medium. The TCP/IP protocol operates in this way. IP is a connectionless

protocol that allows packets to traverse the packet as individual PDUs

known as datagrams. A datagram is an independent, unacknowledged packet,

which can arrive out of sequence, with an error, or not at all. TCP, as we will discuss

in Chapter 6, provides the stability that teams with the unreliable IP to form

the protocol that most of the world uses.

Data Network Addressing

A major difference between the PSTN and data networks lies in the addressing

method. A voice station address is the familiar E.164 telephone number, which is

unique for each station throughout the world. Data addresses are also unique, but

they are designed for machines to read. Humans would find it difficult to work

with an address such as 00-05-E9-3F-88-4A, which is expressed in a form known

as hexadecimal. All computers use the binary form of addresses—hexadecimal,

which consists of a series of 4 bits, is just a convenient way to make it less daunting

to humans. Hexadecimal has a number base of 16. It uses the digits 0 to 9 plus

letters A to F as symbols. For example, binary 1111 equals hex F.

CHAPTER 3 Introduction to Data Communications 41

Data protocols operate with a variety of addressing methods. For example,

the address mentioned above is typical of an Ethernet address, a unique address

that is permanently encoded in the card. The computer that contains the card can

be named to make the address easy for humans to use, but this raises a complication.

Any computer can be connected to an Ethernet port anywhere in the

world, and its owner could request access to the Internet or a private network. The

computer name and Ethernet address must be correlated somehow. The Address

Resolution Protocol (ARP), which runs on every LAN, translates the computer

name to its MAC address.

The Internet has a completely different addressing scheme. IP addresses are

expressed in decimal form such as 23.128.55.3. Network users would find this

address complex, but network technicians use it every day, and often must know

its binary equivalent. For users, IP addresses are expressed as a uniform resource

listing (URL) such as username@company.com. Domain name service (DNS)

translates between the URL and the IP address. Every computer that connects to

the Internet must inform the network through its setup routine of the URL of its

DNS. We will discuss this in more detail in Chapter 6. For now it is sufficient to

understand that many different protocols have unique addressing methods and

where they must interoperate, some method of address resolution is provided.

Local Area Networks

A LAN is a high-speed privately owned data network that links computers and

peripherals over a short range. The primary motivation for LANs arose from a

need to share peripherals such as printers, and to access files that are shared in

servers. The industry proposed several proprietary LAN solutions, but none

caught on until the IEEE 802 committee standardized two LAN protocols. One,

802.3, is an offshoot of Ethernet, a proprietary product that Xerox developed and

offered to the committee. The 802.3 standard that IEEE adopted is similar to

Ethernet, but not compatible. Nevertheless, 802.3 is universally known by the

Ethernet name. IBM promoted a different standard, which became 802.5 token

ring. An 802.5 network uses a token-passing protocol that is a variation on polling.

Before a station can transmit, it must capture a software token that circulates

around the ring. The vast majority of LANs use Ethernet.

LANs form the foundation of nearly every network today. Computers come

equipped with Ethernet ports, which enable them to connect directly into an office

network. The Ethernet port connects via high-quality twisted-pair wire to a

shared hub or, more frequently, to a switch that connects ports long enough to

pass a data frame. Within the range of an Ethernet segment, which is nominally

100 m from station to hub, data travels at 100 Mbps or more. A variety of highspeed

methods, the most common of which is a fiber-optic link where it is practical,

or a WAN can expand segment diameters where common carrier facilities

are used.

42 PART 1 Introduction

Wide Area Networks

When LANs must extend outside the boundaries of privately owned facilities,

they normally use common carrier facilities. Facility is a generic term that

describes the combination of local loops and long-haul circuits that make up a

connection between subscribers’ endpoints. A variety of services and associated

protocols are available for connecting LANs. The industry typically shows the

WAN as a cloud as in Figure 3-1. The LAN owner is concerned with how to feed

the service provider’s devices at the network edge, but has little knowledge of

how the data flows inside the WAN. In this figure, routers at each customer

location connect LANs to the WAN. The industry has a sub-category of the

WAN called a metropolitan area network (MAN). The major difference between

the two is in the span of the network. A MAN typically does not extends outside

a metropolitan area. We can confine the discussion to WANs because the services

and protocols are essentially identical. The distinctions between the MAN and

WAN are blurred and the differences are unimportant for now. We briefly discuss

the types of facilities available for WANs here, and will elaborate on them in later

chapters.

Common carrier WANs facilities can be classed as one of seven types: pointto-

point, multidrop, circuit switched, message switched, packet switched, frame

switched, or cell switched. Any one or a combination of these can serve as the

WAN cloud.

Point-to-Point Circuits

A point-to-point circuit is a dedicated private line that is directly wired between

two endpoints. Analog voice-grade circuits are available, but rarely used

today. Much higher bandwidths are used for data transport. Digital circuits of

56 Kbps in North America or 64 Kbps in Europe are available, with bandwidths

extending as high as 40 Gbps. Some carriers provide “dark” fiber, which subscribers

can illuminate with their own equipment and use up to its capacity.

Point-to-point circuits can be terminated on a variety of different devices; data circuits

usually terminate on routers. Carriers typically price dedicated or private

line facilities in three segments: originating loop, interoffice channel (IOC), and

terminating loop.

Multidrop Circuits

The simplest point-to-point circuits connect directly between two endpoints

that have exclusive access to the circuit. If the endpoints do not have enough

traffic to use the capacity of a circuit, multiple circuits can be bridged in the central

office in a configuration similar to Figure 3-2. Legacy data networks such as

IBM’s SNA have such a topology. A front-end processor polls controllers in

round-robin fashion. If a terminal attached to the controller has traffic to send, the

controller forwards it. Otherwise, the controller responds that it has no traffic.

The central unit also sends output messages to the controllers when it has downstream

traffic. Circuit time is consumed with polling messages and negative

responses, adding to the overhead. A multidrop circuit is an effective way of

sharing capacity in a distributed network where no single station needs more than

a fraction of the circuit time. Automatic teller machines often use this type of

network.Circuit Switching

In a circuit-switched network, stations connect to a central switch in a star configuration.

Fax machines use the PSTN almost exclusively. Other data devices may

employ it as a matter of convenience; otherwise, it is rarely used for data except

for access to a WAN or LAN. Circuit switching has a major advantage compared

to other sharing methods in that the endpoints have exclusive use of the link to

the central switch, which reduces security problems. The usage-sensitive nature of

circuit switching and its bandwidth restrictions limit its use in data communications.

Also, circuit switching wastes bandwidth. When a station has no data to

send, the circuit’s bandwidth cannot be allocated to another station.

Message Switching

Message switching networks are sometimes called store-and-forward. Stations connect

to a computer that accepts messages, stores them, and delivers them to their

destination. The storage turnaround time can be either immediate for interactive

applications or the message may be delayed for forwarding when circuits are idle,

rates are lower, or a busy device becomes available. Data networks used message

switching in the past, but it is not common today except for electronic mail, the

essence of which is store-and-forward.

Packet Switching

A packet network consists of a collection of nodes connected with links as shown

in Figure 3-3. Subscriber stations connect at the network edge with either dedicated

or dial-up access circuits. The interface between the network and the subscriber

station is the X.25 packet switching protocol. Data travels across the network in

packets, which consist of address and control headers, a data payload block, and

an error-checking trailer. Each node checks incoming packets for errors, and forwards

them to the next node only after they are error-free. Inside the network, the

service provider determines how to interconnect the nodes and how much bandwidth

to provide. Stations communicate across a packet network with virtual

circuits. A virtual circuit is a connection between endpoints defined in software over

a path that is shared with other stations. Packet networks offer both permanent

virtual circuits (PVCs) and switched virtual circuits (SVCs). The carrier provisions a

PVC. An SVC is a temporary circuit set up for the duration of a session.

Common carrier packet networks are known generically as public data network

(PDNs). Users access the network over either dedicated or dial-up connections.

Before the Web became popular, many public databases allowed their users

to dial local connections into a PDN node to set up a connection to the database

provider.

Frame Switching

Packet networks were popular in the 1970s when physical circuits were analog,

error rates were high, and most applications were text-based. When fiber opticsbecame available, errors dropped to a fraction of their previous level. The error

rate was so low, in the order of one error or less per billion bits, that link-by-link

error checking was a waste of time and processor resources. The frame relay protocol

was the next phase of packet network development. The architecture is similar

to that of a PDN, except that dial-up access is rare. If a frame relay node

discovers an error, it discards the packet and leaves it to the endpoints, which are

usually routers, to arrange retransmission.

The method of charging on frame relay networks is also different. Packet

networks typically base their charges on kilopackets transmitted. By contrast,

frame relay costs are based on bandwidth. The subscriber pays the cost of the

access circuit, the bandwidth of the access port, and the cost of the PVC. For each

PVC, the customer chooses a committed information rate (CIR). The CIR is the

minimum rate the network guarantees to carry. The frame relay protocol allows

the data rate to burst up to the speed of the access port if capacity is available.

A user could, for example, purchase a 64 Kbps access circuit and port, but choose

a lower CIR, such as 16 Kbps. The network would guarantee to carry at least

16 Kbps. If capacity was available, the network would permit the subscriber to

send bursts up to 64 Kbps. If capacity is not available, the network can mark

frames as discard eligible, carry them if capacity permits, but discard them to

protect itself from overload.

46 PART 1 Introduction

Node

Link

SD

Professional Workst ation 6000

PRO

SD

Professional Workstation 6000

PRO

Access

Circuit

Access

Circuit

F I G U R E 3-3

A Packet Network

Another major departure from the packet network is in the bandwidth

offered. Packet networks typically support dedicated or dial-up analog access.

Most PDNs had a maximum access bandwidth of 56 Kbps. Frame relay always

uses digital access. The minimum access circuit speed is 56 Kbps, and bandwidths

of up to 45 Mbps are available.

A major advantage of frame relay is that it offers the security of a dedicated

network. A customer’s packets are never accessible on another customer’s premises.

Circuits are shared only in the backbone, and all backbone nodes are confined

to common carrier premises. Frame relay is discussed in more detail in Chapter 34.

Cell Switching

Cell relay, the predominate example of which is asynchronous transfer mode

(ATM), is a combination multiplexing and switching protocol. The architecture of

the network is similar to a packet network except the protocol slices data streams

into short cells, 48 octets in ATM, and forwards them across the network with a

short header, which is five octets in ATM.

ATM is used in both private and public networks. Unlike other data protocols,

cell relay is designed to be used in both voice and data networks. ATM

circuits form much of the basis for the Internet and for transporting voice across

common carrier networks. It is also commonly used as the access protocol for

digital subscriber line (DSL).

IP Networks

A growing tendency in the industry today is to use IP networks. Architecturally,

an IP network is similar to the packet-switched network in Figure 3-3. In fact, it is

a packet network, but with some important differences compared to a PDN or

frame relay. The nodes in a packet network are usually routers, which are specialpurpose

computers that are optimized for data communications. Routers are

available with a variety of input ports, including frame relay and ATM as well

as dedicated lines. Since IP is a connectionless protocol, packets can be sent to

any station that has an IP address. This is both an advantage and a disadvantage.

The main advantage is that it is not necessary for the service provider to define a

PVC. The major disadvantage, however, is that an IP network is exposed to all the

security threats that plague the Internet.

The Internet service provider (ISP) is the Internet’s counterpart of the LEC in

the PSTN. ISPs serve the customers and provide access into networks of regional

or backbone carriers. Customers may connect to the ISP with either dedicated or

dial-up service. Dedicated service may be a point-to-point digital line, but smaller

users use either cable access from a CATV provider or DSL from a LEC. The ISPs

may choose to connect themselves through private peering arrangements, or they

may connect through a network access point (NAP) or a metropolitan area

exchange (MAE). NAPs are currently located in New York, Miami, Chicago, San

Francisco, and Washington, DC, with large carriers providing the hosting facilities.

CHAPTER 3 Introduction to Data Communications 47

MAEs are located in most major cities worldwide. The main requirement for an

MAE is collocation space in a building that is secured against all of the usual hazards

such as fire, sabotage, and power failure, plus the availability of broadband

fiber facilities with protected bandwidth. Figure 3-4 illustrates how the Internet is

configured.

Routers attempt to find the most effective path for packets to reach the

destination. They do their work in two stages. The first is to consult their routing

tables, which may be extensive databases of the ways to reach a destination

address. If the destination cannot be reached over a route contained in the table,

the router forwards the packet to another router that is closer to the NAP or a

peering point. Packet forwarding is the second stage of this two-step process.

IP addresses can be either public or private. If public, the address is directly

reachable over the public Internet, opening the network to security concerns.

The first step in insulating the LAN from the Internet is network address translation

(NAT). As its name implies, NAT translates the public address space to

private addresses behind a firewall. The private address space cannot be reached

directly from the public Internet. A firewall is either a hardware device or a

software applet that enables the network administrator to restrict who can

access the internal network. One reality of security is that convenience and

protection are usually diametrically opposed. The best firewalls do not necessarily

block viruses and worms because holes in the firewall must be opened to allow

e-mail and others must be opened for voice and other applications. To derive a

private network over IP, users can form virtual private networks (VPNs) by

“tunneling” through the Internet. A VPN tunnel uses encryption and authentication

to ensure that even if communications are intercepted, the intruder will not

be able to read them. IP networks and security will be addressed further in

Chapters 10 and 36.

Network Hierarchies

Large data networks are usually hierarchical in form. ISO defines terminology

for three levels of network entities: end system (ES), intermediate system (IS) or

area, and autonomous system (AS). An ES is a device such as a terminal, personal

computer, or printer that does not perform traffic forwarding or routing functions.

ESs are typically connected to an IS, which is a device such as a router,

switch, or bridge that routes and forwards data packets. There are two types of

IS: intradomain and interdomain. An AS, also known as a domain, is a collection of

networks under common administrative control. An intradomain IS communicates

within a single AS, while an interdomain IS communicates between ASs.

An AS may be subdivided into areas, which is a logical grouping of network

segments. Figure 3-5 illustrates the concept. The devices connecting the areas

are generically known as gateways; the gateway function is usually contained in

a router.Data Network Performance Measures

Acrucial question for every network manager is how much bandwidth is enough.

Bandwidth is cheap in a LAN, but in the MAN and WAN, where it is leased from

a common carrier, it is costly. Insufficient bandwidth results in lost productivity as

users wait for the network to respond. Response time is defined as the interval

between the time the user presses a key or clicks a button to launch a demand and

the screen begins to fill. Response time can usually be improved by increasing

bandwidth, but at a higher service cost.

The most effective data networks make bandwidth available to meet the

demand. Since traffic demands come in peaks that do not coincide, the bandwidth

is allocated to multiple sessions. Throughput, which is defined as the number of

data bits correctly transferred per unit of time, and reliability are increased by

using mesh networks. The nodes in data networks are either computers or routers

that have the intelligence to move traffic to another route if the primary route is

congested. The links in a full-mesh network connect every node to every other

node. Full-mesh connectivity between nodes is usually unnecessary. Providing

several paths to and from each node offers high reliability as well as the ability to

avoid congestion, but at a higher cost.

Every network has a bandwidth ceiling that is set by some choke point,

which is usually the bandwidth of the access circuit. If the access circuit is uncongested,

it can devote its entire bandwidth to the application that requests service.

In a lightly loaded circuit, the response time is easy to predict—it is a function of

the throughput and the size of the file. If several applications are using the access

circuit simultaneously, each receives some portion of the bandwidth. How much

they get depends on the router programming. Some packets are more important

than others. Perhaps the importance depends on who is using the network, or

maybe it depends on the nature of the application. This issue becomes more

important as data and voice attempt to share the access circuit.

Differences in Quality Requirements

The PSTN has evolved to the point that quality is rarely a problem for voice.

Quality is subjective, and carriers measure it by taking controlled opinion surveys.

The standard of reference is the so-called “toll-quality” circuit, which earns a

mean opinion score of 4.4 on a scale of 1 to 5. For the Internet to approach toll

quality, several measures known as quality of service or QoS must be applied. Users

do not always demand toll quality as evidenced by the fact that they accept lower

quality on cell phones in favor of the convenience.

Data, voice, and video can and do share the same circuits, but they assess

quality in completely different ways. Reliable data communications must be free of

errors; that is, every bit must arrive at the destination exactly as it left the source.

Any deviation constitutes a bit error. Microwave radio has a much higher bit-error

rate than fiber optics, which usually is nearly error free, but no transmission

medium is completely devoid of errors. Data systems detect errors and correct

them by retransmitting the errored information. Voice sessions, by contrast, can

50 PART 1 Introduction

tolerate a bit-error rate that would render a circuit unusable for data. Error correction

is essential for data, but it is a waste of time and resources for voice. Delay-sensitive

applications such as voice and video make no attempt to detect and correct

errors and dropped packets.

TCP/IP protects itself from congestion in a variety of ways, but the input

buffers of routers may fill to capacity before the router can forward packets to the

next hop. When this happens, the protocol permits the router to discard packets.

Voice can tolerate a certain level of packet loss, but interactive video is highly sensitive

to lost packets. The percentage of lost packets is an important measure of

network quality.

Delay or latency, which is the time required for the signal to travel from source

to destination, is also a critical quality measurement for voice and interactive video.

If the latency in a voice session is excessive, the participants tend to talk over each

other and natural conversation becomes difficult. Voice connections ideally have a

maximum of 150 to 200 milliseconds (ms) of delay. Satellite circuits, most of which

have been replaced by fiber optics, have a delay of about 250 ms. Although people

can talk over a satellite circuit, it can be disconcerting and carriers avoid it if terrestrial

circuits are available. To the data user, delay is relatively unimportant. If the

response time is half a second longer than normal, it may hardly be noticeable.

Noise and loss are two phenomena that have an adverse affect on both data and

voice. They are characteristic of analog circuits, but affect digital circuits to a minor

degree. Loss refers to the drop in volume from one end of a circuit to the other. When

loss is excessive, a voice user has difficulty hearing and the difficulty increases with

a higher noise level. Data are more tolerant of loss than voice. Since data signals terminate

in electronic apparatus, they can be amplified to the level the equipment

requires, but noise is amplified with it. Noise is important to data only if it causes

errors. To both voice and data, the signal-to-noise ratio is the important factor.

Echo is another quality impairment that is important to both voice and data.

Echo occurs when a signal hits an irregularity that reflects a portion of the signal

back to the source. Excessive echo can render a circuit unusable for both voice and

data. It is generally resolved by using echo cancellers, which are devices that look

at the forward signal, compare it to the reverse signal, and cancel the reverse path

if the signals match.

Differences in Session Requirements

The nature of a voice session is much different than a data session. Voice sessions

begin with setting up a connection, which is exclusive for the duration of the call,

and then tearing the connection down. When a data user dials up a connection, such

as remote access to an ISP or a corporate network, the session is, for all practical

purposes, a voice session, with the modem emulating the human voice. Dial-up sessions

are acceptable for some data applications, including Internet access, but dialup

is usually an expedient that is used only when dedicated access is unavailable.

Voice sessions are usually user to user, extending to multiple users in the

case of a conference call, but easily confined to fixed bandwidth. The nature of a

CHAPTER 3 Introduction to Data Communications 51

voice session leaves half the bandwidth of the circuit unused. In addition, during

pauses in conversation, no information flows in either direction. Data sessions can

make use of this idle bandwidth, but it is important that long data packets not

block short voice packets.

DATA APPLICATION ISSUES

This chapter has touched on most of the concepts needed for a broad understanding

of how data networks function and how their quality is measured. Data and voice

both ride over the same physical fiber-optic backbone, but at the circuit level they are

mostly separate. That, however, is changing. The trend is toward convergence of all

media, including voice, data, video, graphics, and multimedia applications over a

combined network. Before that transition can occur, several issues remain to be

resolved, so the networks will remain separate for the next several years.

Today, frame relay is the most popular data service. Frame relay also carries

a limited amount of voice and video. The main reason for its popularity lies in its

simplicity and security. Aframe relay network is provisioned within common carrier

facilities, and is fully as secure as a network composed of dedicated point-topoint

lines. It is economical, and offers bandwidth on demand up to the capacity

of the input port.

The major frame relay carriers also offer IP networks, and are recommending

that their customers move their service from frame relay to IP VPNs. For the customer,

an IP network has a major attraction. The typical enterprise is widely distributed

and international frame relay is expensive. Most sites already have

dedicated Internet access. If the same circuit can be used for access to both the

Internet and internal enterprise network, the cost may be significantly less. A transition

to an IP network would be easy to justify except for security and privacy

concerns. To implement an IP VPN, the customer either must set up and manage

appliances that provide the tunnels or turn the task over to the carrier. Either

approach raises the cost to the point that much of the expected saving evaporates.

Like the telephone network, the IP network is a complex and intricate

collection of circuits, routers, and servers. The Internet is most effective for those

that have full-time connections to it, but dial-up access through the PSTN is an

alternative for those that do not.

For carriers the choices are point-to-point circuits, MPLS, or ATM. For enterprise

networks, the choices are point-to-point circuits, frame relay, or IP, with the

latter gaining in popularity. The three services have a narrow range of overlap in

their applications and differences that must be clearly understood.

Access Channel

The service providers usually obtain the access channel from the customer premises

to the carrier’s point-of-presence (POP) from the LEC or a CAP. The access

channel cost within an LEC’s wire center is based on bandwidth. If the POP is in

52 PART 1 Introduction

a different wire center than the point of service delivery, the access channel may

also be distance sensitive, meaning the longer it is the more it costs. Therefore, the

location of the service provider’s POP is important. To be price competitive, the

carrier may select its POP for rating purposes at a different location than the point

of connection to its backbone network. Nevertheless, the cost of the access channel

is apt to vary widely among carriers and types of service. IP networks can use

DSL or cable for the access channel, which offers inexpensive access that carriers

can provision in short intervals.

Pricing Structure

Unlike frame relay and IP, which are not distance sensitive, the cost of point-to-point

circuits varies with the length of the IOC. The cost of all three alternatives varies

with the amount of bandwidth. IP is connectionless, so packets can be routed to any

endpoint. Frame relay has a PVC charge, which defines the endpoints that can

communicate across the network, although stations that lack a PVC can still communicate

through the central site. In addition, a CIR charge applies to each PVC.

Multiple PVCs to the same location may share access channels.

Bandwidth

Point-to-point bandwidth can be obtained up to the capacity of a fiber-optic

wavelength or lambda. Frame relay providers usually offer a maximum bandwidth

of DS-3.

Security

From a security standpoint, a dedicated channel offers the maximum security.

Frame relay is a close second. With both dedicated circuits and frame relay, the circuit

is accessible only on the carrier’s premises. Security is the weakest point in an

IP network and is the primary factor weighing against its use in sensitive applications.

Although IP can theoretically be made as secure as frame relay or dedicated

circuits, IP is vulnerable to such hazards as denial-of-service attacks that

cannot affect frame relay or dedicated circuits. Some applications require firewall

holes that destructive viruses and worms can penetrate. IP networks can be

secured, but to preserve security levels they require more administrative attention

than frame relay.

Connectivity

A major advantage of IP is its connectivity. Any IP address that is part of the public

addressing structure can be reached from any other location. This is a major

advantage compared to frame relay, which offers the equivalent of a dedicated connection.

Frame relay PVCs are provisioned using an identifier known as a datalink

CHAPTER 3 Introduction to Data Communications 53

connection identifier (DLCI), which defines the circuit. Anode on a frame relay network

can communicate with another node only by defining a PVC to that node or

by communicating through another node.

Service Level Agreements (SLAs)

The three services are much different in the ability of the carriers to provide tight

SLAs. SLAs are generally quoted on the basis of factors listed below.

Availability

Availability is defined as a percentage of time the network is available for service.

For example, a network with 1 h of the outage per year has an availability of

99.989%. This is calculated with the formula:

Percent availability = 1 – (outage hours/8760 h/year)

Availability is normally quite high for all three types of facility. Point-topoint

circuits are not subject to degradation during overload periods as are IP and

frame relay. IP reliability is somewhat less than frame relay, which is not subject

to hacking attacks.

Order Intervals

The user is concerned only with obtaining the access channel in an IP network.

Once the channel is up, the endpoint can connect anywhere in the world without

carrier intervention. Access channels can be provisioned quickly by using DSL or

cable. If a dedicated access channel is required, the order interval is the same as

for dedicated or frame relay circuits, but a major advantage of IP is the speed with

which it can be provided.

The longest intervals can be expected with point-to-point circuits, which must

have two access circuits and an IOC provisioned, often by multiple carriers. To some

degree the same is true of frame relay, except that the service provider can provision

the DLCI quickly, and can make changes to the port speed in a day or two.

Technical Variables

The tightest limits on such variables as noise, bit-error rate, latency, and packet

loss can be obtained with point-to-point circuits, and to a marginally lower

degree, with frame relay. Some frame relay providers offer classes of service with

latency and packet loss limits that are tight enough to support VoIP.

The ability of ISPs to offer SLAs on technical variables depends on the

degree of control they have over the network. Large ISPs may have their own

backbone, and therefore be able to control the network, but smaller ISPs may have

little or no control after they hand off packets to a backbone provider. In any case,

latency and packet loss will be higher on an IP network than frame relay.