Networks are classified as
Many network applications run by invoking functions of the network software, and with little or no concern about the physical characteristics of the specific network platform in use. Characteristics of the major types of network systems do intrude on application development in some cases. Also, concerns such as privacy, security, reliability, response time, and accessibility often depend on the type of network platform in use. Local Area Networks
Local Area Networks extend the usefulness of one computer by connecting it to others. Often a computer on a network is connected to other networks as well, and has essentially unlimited access to resources and services around the world. At the local level, the network may be any of several popular types. Local area networks have very low error rates and propagation delay that is negligible for most purposes. Transmission in most local area networks is by broadcast -- every station on the network receives every transmission. Thus, there are no routing decisions to make. Every packet (message unit) follows the same path, so reordering does not occur. These characteristics make LANs suitable for applications that depend on timely results, for example interactive processing involving files or other resources located on several different systems or time-critical response to a monitoring device. The broadcast mode of operation is suitable for applications requiring message exchange among a number of stations, like checking individual calendars to schedule a meeting or searching for a processor able to share in a demanding computational task.
A combination of technology and the performance expectations of LANs limits the distances they can cover. A small LAN might connect a few computers in an office, or in a home; a large LAN could extend over an office park or university campus, connecting computers and other devices in a number of buildings. Common speeds are 10Mbps (mega bits per second) or 16 Mbps.
Very simple network systems that allow limited sharing between two or more personal computers (for example, at home) are sold at most computer stores. The simplest of these use the computers' serial ports, standard telephone wire, and software that allows printer sharing, file transfer, and sending messages between computers.
This system has the three essential components of any network: a communication medium (telephone wire), an interface between the computer and the medium (the serial port), and software to send and receive on the interface. In the simple network shown in Figure daisychain , each station is connected to the next in a line. Each computer can send a message to, access a file, or use a printer on any of the other connected computers. If RICK uses the printer attached to LINDA, AMY and DARREN participate in the effort by passing on the messages. (This is not a broadcast type LAN, since a message does not go beyond the intended destination station.)
Most local area networks are more complex than this. Many local area networks have servers -- designated stations that provide access to disk storage or printers or other resources shared by users at other stations. Regular use of the servers increases traffic on the network and suggests that individual LANs be kept small, limited to a small cluster of computers that can efficiently share a common set of resources. Communication and sharing among clusters requires that the small LANs be connected. A popular way to interconnect local area networks is to attach each to a high speed backbone network such as shown in Figure backbone . Though the backbone could be the same type of LAN used in the clusters, more of them are using fiber-based network technology. Fiber Distributed Data Interface (FDDI) operates at 100 Mbps. It outperforms a slower LAN in providing high data rates required to connect other LANs as a backbone. The distributed queue dual bus (DQDB) is another high speed fiber-based network for times when high data rates are required. Because of their potential for city-wide installation, high speed fiber-based networks are sometimes called Metropolitan Area Networks ). (See Section MAN Halsall92
All local area networks have the same components: communications medium, an interface between the computer and the medium, and software to drive the interface to send and receive transmissions. The software coexists in the computer with other hardware drivers, such as those for printers and modems; and with other software such as compilers and databases, text editors and spreadsheets, as seen in Figure netparts . All the network services we will see in this book are simply extensions to the large collection of services already offered by modern operating systems.
Local area networks are sometimes described by the topology used to connect the machines. For the most part, the topology is transparent to the user and to the application developer. Sometimes, there are implications that affect the development effort. In Figure daisychain , the stations are connected in a daisy chain: each is tied to the next and if any station is lost, the chain is broken and communication cannot continue across the break. The most popular topologies for LANs are the bus , the ring , and the star .
In the bus topology, the stations are connected to the medium, which can be run in any convenient shape -- just like Figure daisychain . The difference is that each connection is independent of the others. Individual computers attached to the network do not participate in passing each others' messages along. The way the stations share the single communication medium is similar to the way people share multiple extensions on the same phone line (listen to see if the line is in use; if so, try again later; if not, talk). The rules for sharing the bus are codified in the standard IEEE 802.3, derived from and nearly identical to Ethernet
In the ring topology, the medium is distributed as a closed loop. The stations are attached to the ring and share the ring by passing a ``token'' that specifies whose turn it is. IEEE 802.5 defines the way the token ring operates.
IEEE 802.4 defines a special-purpose network called the token bus. It requires network stations to behave as a logical ring while connected to a physical bus. The bus is a convenient network type to install in factories and processing plants, while the token-passing method of gaining access to the transmission medium provides a guarantee for the maximum time a station must wait to receive a chance to transmit.
There are several approaches to the star topology. The idea is to have a central location where all trouble shooting can be done. To understand the desirability of this, suppose you are sitting at station RICK in Figure daisychain and attempting to print at station LINDA. If the print request does not succeed, there are several possible problems. Any of the stations between you and the printer may be turned off or malfunctioning. If the stations are all lined up in the same room, it is not difficult to find the problem. However, if each is locked in a separate room, and you don't have the keys, the solution is not so simple.
When the network topology is a star, every transmission goes through a central site. When trouble occurs, locating the problem requires disconnecting suspected links and trying again. When the faulty unit is located, it can be repaired and then reconnected, all from the central site. Meanwhile the rest of the network continues to function.
The bus, ring, and star topologies are the basic building blocks of network shapes. More complex figures, trees and snowflakes, are constructed from these. Figure tree shows several ways to join bus segments to form a tree shape. The boxes shown between the bus segments are one of several types of network connecting devices. The type of connection is determined by the protocol layers at which the box participates in the network operation. Though the word gateway is sometimes used as a generic term for any device that connects networks, three types of connecting devices have specific names:
If the boxes consist of only the physical layer, they are repeaters ; they just enhance the electrical signal that carries the communications on the networks. In that case, any transmission sent on any of the network segments arrives at all the others. Networks joined at by repeaters are really one logical network.
The use of repeaters allows a network to spread over a larger distance than it could otherwise span, and also allows more flexibility in getting the cable where it is needed. Multiple port repeaters also make it easy to expand a network. A new segment can be run where the additional stations are needed, then joined to the network by connecting to a port in the repeater. When the boxes shown in the figure are repeaters, the tree shape is a wiring decision, but the network is logically a single bus. Just like a bus with only one segment, all the stations receive every message; and all the stations must share the same communications medium.
If the boxes in Figure tree include network protocols implemented to the data link layer, they are called bridges . Bridges not only repeat the transmission, they also examine the packet to determine if it should be sent on to the networks on the other side of the box or confined to the network where it began. This filtering operation reduces the traffic on the individual segments of the network. Further, by keeping messages exchanged by two nodes on the same segment on that segment only, filtering provides a limited privacy function.
Routers are connecting boxes that participate in network protocols to the network layer. Routers can decide the path a packet should take to reach its destination, can participate in distributed congestion control schemes, and can fragment large packets and reassemble them when necessary to get through intermediate networks. Routers rely on information in the data packet to determine what action to take, so routers are considered protocol sensitive, i.e. a separate router is needed for each protocol used on the connection. Witkowicz88 Schnaidt88
Networks can be connected at higher layers as well. The higher the protocol layer implemented in the gateway, the deeper the gateway looks into the data that it transfers. An example of a Layer 7 gateway is a mail gateway, which connects two incompatible mail systems and allows them to exchange messages. Figure Gateways shows network connections through repeaters, bridges, routers, and a layer--7 gateway.
Figure snowflake shows several star-shaped networks making a more complex structure. This arrangement is convenient when a network includes stations on several floors of a building. A wiring closet on each floor provides a convenient place to look for problems, without requiring the extensive wiring that a single hub needs. In later chapters we will look more thoroughly into interconnecting networks, including implications for applications that span the connections. Wide Area Networks
Unlike local area networks, wide area networks consist of specific point-to-point links. Messages pass from one network station to another through a communications subnet. The communications subnet consists of nodes whose function is to pass the message from its origin to its intended destination quickly and accurately.
Wide area networks are characterized by significant propagation delay in message transmission and by high incidence of lost or damaged transmissions. Message units, called packets, are passed from one intermediate node to another until they arrive at the destination. Effective routing techniques are important to network performance. Packets from a single message may travel by different routes and arrive at the destination out of order. Typical transmission speeds range from 56 Kbps to 1.54 Mbps.
Since a wide area network is composed of point-to-point connections, connectivity is an important design consideration. Figure links shows some of the ways to join a set of five nodes in a communication subnet by point-to-point links. Figure links a) shows the minimum number of links to allow each node to communicate with each other. It is really the same topology as we saw in Figure daisychain , and consists of a simple daisychain. Routing is easy: each station only has to know which direction to go to reach the destination. The obvious weakness is vulnerability: in case a single node (other than the ends of the chain) fail to function correctly, communication is lost across the faulty node.
Figure links b) shows a fully connected configuration of five nodes. Again, routing is easy: every node connects directly to every other node. Vulnerability to failure is minimized, since no node is dependent on any other. The problem here is cost. From graph theory, we know that a complete graph with n vertices has edges, where is the combination of n things taken two at a time. Fully connecting five nodes ( ) might be extravagant; fully connecting a hundred nodes ( links) is probably unreasonable. Networks of many thousand nodes clearly require another approach.
Figure links c) shows a compromise configuration for the five node example. Every node has at least two connections. No single node loss would cut off any operating node from the others. On the other hand, loss of two nodes could interrupt communication among the remaining nodes, depending on which two nodes failed. Routing is no longer obvious in this network. Since there is a direct link from node A to node B, it is likely that is the best route for communications between those nodes. To get from A to E, however, a transmission could go by way of B or C. Many factors could enter the decision. One path might support higher transmission rates. Security concerns may dictate the better choice. Different costs may be involved. Even these issues would not be difficult if the best choice were always the same. Transient conditions such as heavy congestion complicate the decision process. Processes running in the nodes respond to instructions from the host computers regarding routing; but they also monitor the behavior of the network and learn what paths are best.
Often the station connected to the subnet is a gateway to a local area network. Thus, the user with access to the resources of a local network can reach out to a larger network for still more resources. Local area networks (or Metropolitan Area Networks) connected to other networks combine to form very complex patterns with characteristics similar to point-to-point wide area networks. See Figure localnets . Similar problems concerning routing decisions, dealing with failures, temporary conditions, etc. arise. A further complication concerns the differences among the types of local area networks: the required format for transmissions, maximum lengths, conventions for acknowledging successful delivery of a packet, etc. The stations that join two or more local area networks and address these issues are called routers .
Transparent access to very remote files and printers is not impossible, but is rare. Although the most common application of a wide area network is explicit communication (electronic mail and file transfer) another popular use is to login to a remote computer that offers a particular service. Thus, for instance, the NSFnet gives access to supercomputers to researchers working at sites far removed from such a resource. Travelers also use wide area networks to login to their own computers while away from the home or office. Metropolitan Area Networks
The term Metropolitan Area Networks is often applied to the new high-speed network technologies to distinguish them from wide area and local area networks. These technologies have pushed the carrying capacity of the communication links into the gigabits range. They can be used over larger areas than LANs, and are often used to connect LANs together to form a greatly extended LAN environment.
Another important potential for these networks is the ability to carry information in forms that require many more bits than conventional text or simple graphics. An ordinary computer monitor screen displaying 25 lines of 80 characters contains only 2000 characters. At 8 bits per character, only 16,000 bits are needed to fill the screen. At 10Mbps, the standard speed of LANs, only 1.6msec would be needed to fill the screen. However, when that same screen displays high resolution, bit-mapped images at 1024 by 768 bits, 786,432 bits are required to fill the screen. At 10Mbps the screen is refreshed every 78.6 seconds. If the display includes color (256 color images are common), the time required to deliver the information for a full-screen display (at 10Mbps) grows to 20.13 seconds -- hardly suitable for smooth motion video. Clearly, the standard LAN speeds are not adequate. High speed networks, approaching gigabits per second, are needed to provide the ability to deliver the full potential of multi-media displays to the user from a source on a different machine and to connect LANs together and deliver timely access to resources.
In the later chapters of this book, we will look at the high speed network protocols, to see how they overcome some of the limitations of the older LAN systems, and to see the potential for increased capability in the services networks can deliver.
The entire history of computing is a brief period in human development; the history of networking is even briefer. The first commercial installation of a general purpose electronic digital computer came in 1951. The connection of different types of computers through networks began in 1969 with ARPAnet.
Our historical treatment is deliberately incomplete. We highlight only a selection of the developments in networks that evolved into, or were instrumental in, current systems or architectures. References are included to enable you to pursue any development in more depth.
ARPAnet, ISO-OSI, and SNA have already been introduced. ARPAnet consisted of four operating nodes by the end of 1969 Denning89 and grew to a worldwide network of more than 60,000 nodes in 1989. At that time, it was succeeded by a combination of networks using its protocols and called the Internet. Part of the legacy of the ARPAnet is the pair of protocols IP and TCP. The TCP/IP suite, sometimes referred to as ``the ARPA protocols'' has greatly influenced protocol development approved for worldwide use by ISO.
In the late 1960s an effort was begun to determine whether radio communications were preferable to wire communications (telephone systems) for communicating by computer in the Hawaiian Islands. The study, and the resulting system design decisions led to the Alohanet, which began operating in June 1971. Abramson85 Early in the process, the fortuitous decision was made to use a direct form of transmission in single high-speed packet bursts. The ALOHA system became the direct ancestor of the bus protocols in very common use today. One and a half years after it began operating, the ALOHA system was connected to the ARPAnet.
Work on the ALOHA system was instrumental in the evolution of the Ethernet, developed by Metcalfe and Boggs in 1973 while working at Xerox Palo Alto Research Center (Xerox PARC). Metcalfe76 The Ethernet permitted more efficient use of the shared channel communication than the ALOHA approach. In 1980, Digital Equipment Corporation, Intel, and Xerox Corporation joined forces to promote and distribute the Ethernet specification. The resulting very broad early market penetration is largely responsible for the great commercial success of Ethernet and its dominance in the Local Area Network arena. Ethernet is the basis of the bus network system mentioned in section nettypes .
The Cambridge Ring, developed at Cambridge University in the mid 1970s, moves data in small, fixed length packets among stations linked in a ring. The Cambridge Ring has had significant use in England, but suffers from very high overhead compared to the amount of data that each slot can carry. We mention it here because recent high speed network protocols incorporate a similar slotted ring access method (for example, the distributed queue dual bus, DQDB).
The token ring network, released by IBM in 1985, is the standard mentioned in the context of the ring topology in section nettypes . Unlike the slotted ring, the token ring supports variable length packets with an arbitrary maximum length, determined by the length of the time the station may retain the token. Originally operating at 4 Mbps, the token ring's speed increased to 16Mbps late in 1988.
In most countries, the government maintains a monopoly on communications. As the need to communicate from one country to another grew, so did the need for standards. The Consultative Committee in International Telegraphy and Telephony (CCITT), housed under the International Telecommunications Union (ITU) of the United Nations, makes technical recommendations about interfaces. Because of the CCITT ties with telephone transmission, its recommendations can have more telecommunications than computer networking flavor. The ISO incorporates CCITT recommendations into standards for the networking community. One significant example of this cooperative process was the protocol X.25, developed by CCITT to mimic traditional voice telephone links on top of an underlying network that transports packets. X.25 is widely used internationally, and is available in the United States. In 1975, X.25 was incorporated into the ISO standards.
Other significant work of the ISO includes the process of defining world-wide standards for inter operability of networked systems begun in 1977 Becker91 and the publication of the ISO seven-layer OSI Reference Model in 1984. OSIModel
A major effect of the technologies that have advanced in the years of computing and of networks connecting computers is the growing reliance on networks and the changes in the way people work because of computer networks. For example, electronic mail was not a major goal in the development of the ARPAnet, but within two years was the principal source of network traffic. By 1988, loss of electronic mail contact was considered a major disaster. Denning89 As a retrospective in LAN Magazine said in 1991
The network has gone from the technology toy of the 1980s to the corporate utility of the 1990s. LANs have transcended their roots as an inexpensive way to share files and printers. Networks are a communications infrastructure. They are the utility that delivers a corporations's information assets. Schnaidt91 [p35--36]