Technology Basics

Topology

Media

Access Method

Product Categories

This report provides the technological details of the major types of LANs, including Ethernet, token-ring, and FDDI, and how the major components are deployed to implement them. Operational features of Local Area Network technology are discussed in depth, including topology, media, and access control. Characteristics of the components needed to assemble a Local Area Network: network interface cards, hubs, and network operating system software are then detailed.

What is a local area network? At the most basic level, it is a collection of hardware and software that connects individual PCs and workstations together for resource sharing, messaging services, and workgroup computing. Yet novices often see the world of local area network technology as an arcane realm populated by unfamiliar devices described only in confusing terminology. In truth, the technical details of local area networking are actually quite simple conceptually, and the components needed to implement a working LAN at the most basic level are few. A grasp of a few key ideas and a knowledge of the hardware and software that embodies those ideas are all that is required to understand networking.

The history of local area networking is relatively short - Ethernet was the first working LAN and it was developed at the Xerox Palo Alto Research Center (PARC), beginning in 1973, by a team headed by Dr. Robert Metcalfe. Ethernet was first widely employed commercially to network terminals to minicomputer systems - more specifically, to network Digital Equipment Corp. terminals to its VAX line of minis. UNIX-based engineering and scientific workstations were also connected by Ethernet early on.

Arcnet - which stands for Attached Resource Computer Network - first appeared in 1977, developed by Datapoint, a Texas-based manufacturer of minicomputer systems. Datapoint licensed Arcnet to several other vendors but required licensees to follow strict guidelines that ensured interoperability between various brands of Arcnet equipment. This Datapoint enforced standard is still in effect today, and Arcnet has never been approved by any independent standards-setting body, though Datapoint has recently sought ANSI standardization.

While these first LANs were being developed, the International Organization for Standardization (ISO) was busy working out a framework that all computer communications networks - local or wide area - could conform to in the interest of interoperability and eventual global networking. The seven-layer model proposed by the ISO in 1983, the Open Systems Interconnection (OSI) model, arranges functions common to all communications networks into a well-defined, hierarchical architecture. While a complete discussion of the OSI model is beyond the scope of this report, the model places networked applications and the user interface at the topmost levels, and the physical implementation of the network - the copper wire, adapter cards, and hubs - at the very bottom. Intermediate layers define how user interaction at the top level is translated into electrical signals on the network medium and then back into useful messages to a server, peripheral, or other user.

The major standards-setting body in the LAN industry is the Institute of Electrical and Electronic Engineers (IEEE) 802 committee. Formed in 1980 with the express task of defining industrywide standards for LANs, the 802 committee has been one of the most successful independent standards-setting bodies in the history of the computer industry, and its contributions to advancing network interoperability and the general state-of-the-LAN art have been immeasurable. IEEE 802 subcommittees are charged with defining standards for various segments of the industry. The 802.3 standard describes an Ethernet-like network; the 802.5 standard covers token-ring; 802.2 describes Logical Link Control - a method of interfacing network software to network hardware at the juncture between the physical and data link layers of the OSI model. Both 802.3 and 802.5 networks implement 802.2 Logical Link Control. Currently, 802 committees continue to proliferate - the 802.11 committee is close to finalizing standards for wireless LANs; the newest committees, 802.12 for Hewlett-Packard's 100VG ANYlan and a new 802.3 subcommittee for 100BASEX, are each dedicated to a type of so-called "Fast Ethernet,'' 100M bps transports on copper wiring. "802.13'' was passed over as a potential committee handle for unstated reasons - apparently no one wanted to standardize their LAN technology under that particular number.

With the introduction of the personal computer in the early 1980s, users recognized that PCs and networks were seemingly made for each other. Both Ethernet and Arcnet were quickly adapted to the PC. IBM's Token-Ring Network was submitted to the IEEE 802 committee in 1982. The first token-ring products were actually introduced in 1985.

In the mid-18th century, Dr. Samuel Johnson, writing his tongue-in-cheek dictionary, defined a network as "anything reticulated and decussated at equal distances with interstices at the intersections.'' While it is certain that Dr. Johnson was not referring to personal computer networks, to the uninitiated, many definitions of local area networks are not much clearer. When terms like "carrier detect multiple access,'' "fiber distributed data interface,'' and "early token release'' are bandied about, Johnson's joke seems succinct.

The problems associated with local area networking, and the confusing terminology as well, can be reduced to a few simple questions about what we are trying to accomplish when we design and build a computer network. Those questions define three basic technological issues associated with local area networking.

This report discusses each of these topics before turning to a discussion of four particular types of networks - Ethernet, Arcnet, token-ring, and FDDI - that are the most widely used.

Networking on a Bus

The earliest commercially viable network, Ethernet, used (and still uses) a bus - a single data path to which all workstations directly attach, and on which all transmissions are available to every workstation. Only the workstation to which the transmission is addressed can actually read it, however. A bus cable must be terminated at both ends to present a specified impedance to the network workstations. Therein lies one of the major disadvantages of a bus network, since any break in the bus cable takes the entire network down.

Physical Star/Logical Bus

A new version of Ethernet - 10BASE-T - appeared in the late 1980s. It uses unshielded twisted-pair wiring and is arranged in a hub-based star topology. Yet the network logically acts as a bus; that is to say, signals transmitted by any workstation are available on the network at all workstations. Only the station for which a transmission is meant can read it. The hub-based star has no single point of failure like the coaxial bus - a cable break on a star disables only the station that the broken cable connects to the hub. Another common Ethernet fault that is alleviated by the hub-based topology is "jabber.'' Sometimes a network adapter card will fail in such a way that causes it to begin flooding the network with meaningless data packets. While a jabbering node will disable an entire bus-based network, and can be extremely difficult to identify, on a hub-based star, an intelligent hub can simply turn off the jabbering node.

Networking on a Ring

The simplest form of a ring topology is one in which each workstation is connected to another adjacent workstation in a closed loop. Such a ring, however, shares with bus-based Ethernet the disadvantage of a single point of failure. For this reason, all ring networks in use today are actually implemented as physical stars.

Physical Star/Logical Ring

Both token-ring and FDDI implement a ring topology on a physical star. At the center of the star is a device similar to a hub called a multistation access unit, or a media access unit. Multiple-MAU networks actually form a ring, since the MAUs are connected via ports at either end of the devices. These ports are called "ring in'' at one end, and "ring out'' at the other. In this case, however, if a cable break occurs between MAUs, the internal circuitry of the MAU can loop the ring back on itself to bypass the break and keep the ring functioning.

Coaxial Cable

Coaxial cable is the original LAN media - both Arcnet and Ethernet use it. In fact, there are two variations of Ethernet on coaxial cable - "thick'' Ethernet, which runs on RG-8 coaxial cabling with a diameter of 0.4 inch, and "thin'' Ethernet, which uses RG-58 coaxial cable with a diameter of 0.25 inch. Both types present an impedance of 50 ohms to the network interface card. Arcnet runs on RG-62 93-ohm coax and uses BNC connectors.

Thick Ethernet workstations use a device called a transceiver to attach to the coaxial media. An installer first pierces the insulation of the coax to expose the central conductor. The transceiver is then clamped on to the cable so that a small spike is in contact with the exposed central conductor. The transceiver is connected to the workstation by a nine-conductor, noncoaxial cable that can be up to 50 meters long. At the workstation, the Ethernet network interface card is equipped with a 15-pin "D'' connector called an attachment unit interface (AUI) to which the transceiver cable is connected.

Thin Ethernet network interface cards are equipped with an onboard transceiver unit, and the coaxial bus cable attaches directly to the card by means of a BNC "T'' connector. This T-shaped connector attaches to a male BNC connector that juts out of the back of the interface card. Two coaxial cable ends are then attached to the arms of the "T.'' Workstations at either end of the bus require a 50-ohm terminating resistor "endcap'' on one arm of the "T'' connector.

Twisted-Pair Wire

Twisted-pair copper wire, also known as common telephone wire, while similar to the wire used in telephone systems, is not really the same. The twisted-pair wire required for reliable data transmission is of a heavier gauge than telephone wire - 24 to 26 American Wire Gauge (AWG). Nonetheless, the idea is the same - the two conductors are twisted around each other, a technique that minimizes the amount of interference the pair causes in adjacent pairs.

There are two varieties of twisted pair in wide use. Shielded twisted pair (STP), also known as IBM Type 1 cabling, is the most common media for token-ring networks. Both 4M bps and 16M bps token-ring networks can also run on unshielded twisted-pair wire (UTP). Shielded twisted pair allows longer cable runs between the MAU and the workstation and is less susceptible to interference of any kind. Shielded twisted pair is also more expensive and more difficult to pull into walls than unshielded twisted pair.

Fiber Optic

Fiber optic is the media of the future - cheap; relatively secure; completely immune to elect<m8>rical interference; and, perhaps most important, theoretically capable of supporting virtually unlimited transmission speed.

Until recently, optical fiber cable was expensive and difficult to install - attachment of a connector required polishing the fiber end and ensuring precise alignment of the media in the connector. For this reason, many considered fiber a viable media only for backbone LANs - central data paths connecting many subnetworks within a building or campus. Newly developed connector attachment methods cut the installation time significantly, however, and the phrase "fiber to the desktop'' has taken on a new plausibility, especially for LANs where high volumes of data must be transmitted on a regular basis. Mainstream LAN users are moving to implement document imaging and multimedia, and these data-intensive applications require the high transmission speeds that fiber make<m8>s possible.

Two types of fiber media are in common use in the communications industry - single-mode and multimode fiber. Multimode fiber allows the use of light-emitting diodes (LEDs) as the light source and is capable of transmission speeds in the hundreds-of-megabits-per-second range. Single-mode fiber requires a laser as the light source and is generally used for long-distance communication, such as in the public telephone network.

How can only one computer at a time be allowed to transmit on the network? Access to the network - the right to transmit - can be allocated in two ways: randomly or in a deterministic order. In a random access method, any station can initiate a transmission at any time - unless another station is already transmitting. In a deterministic access method, each station must wait its turn to transmit. Only one example of each type of access method is widely used. Carrier sense multiple access (CSMA) is the random access method used with Ethernet. Token-passing is the deterministic method used in token-ring, Arcnet, and FDDI networking.

Carrier Sense Multiple Access

Ethernet uses CSMA, a random access method that requires a workstation to "listen'' to the network media before attempting to initiate a transmission. If the workstation "senses'' activity on the line, it defers its transmission for a short period before trying to access the network again. When it senses a clear line, it begins sending. A major shortcoming of this access method is that two workstations can sense an unused line and begin transmitting at approximately the same time. The result is a collision. Ethernet implements carrier sense multiple access with collision detection (CSMA/CD). A collision is "detected'' when the voltage level on the network equals or exceeds the level generated by two stations sending at the same time. Any station can detect this condition, and when one does, it begins sending a jamming signal that forces each transmitting station to stop transmission. The transmitting stations then wait a random amount of time before attempting to transmit again.

CSMA/CD and random access methods in general work best when network traffic tends to be unpredictable and bursty, consisting of many short transmissions. Performance of a CSMA/CD network can degrade quickly under the kind of sustained heavy loads generated by many workstations sending large files. Paradoxically, Ethernet is still most widely used in just such an environment - connecting powerful UNIX-based engineering and graphics workstations.

Token-Passing

Arcnet, FDDI, and token-ring all use token-passing, a deterministic access method that allows only one station at a time the right to access the network. A special data structure called the token passes from station to station in sequence. A station that has data to transmit grabs the token and changes a bit, making the token into a packet header. When the data is received, the altered token is placed back on the ring as an acknowledgment from the intended recipient that the data was received without error. The transmitting station then generates a new token and passes it to the next station on the network.

The basic defining concepts of networking are topology, media type, and access control. There are four different network types commonly used today.

Network interface cards - boards that fit in the expansion slots of the popular types of microcomputers - represent, in practical terms, almost the only way to connect these machines to a network. While adapter cards can be discussed in terms of what network type they support, there are certain common characteristics that the various types share, and we will discuss these features first.

Both microcomputer bus supported and bus size are largely defined by the types of microcomputers in widespread use. Following are a few familiar types.

Media type and the type of connector supplied with the card vary, depending on the type of network the card is designed to support.

Hubs, MAUs, and concentrators are the central elements of LANs based on a star topology, including 10BASE-T Ethernet, token-ring, most Arcnet, and FDDI. In general, when employed in an Arcnet or 10BASE-T Ethernet network, these devices are referred to as hubs; while in token-ring and FDDI, they are called media access devices (MAUs), or multistation access devices, by some vendors, including IBM. Whatever they are called, they perform similar functions and, in many respects, have similar characteristics. They can be standalone devices that support, usually, eight to twelve connections, or, at the high end, they can be modular, multislot devices that support many more than eight connections, though the cards they house are most often sold in eight-port configurations. The low-end, standalone hubs are aimed at the workgroup, while the high-end, modular concentrators are facilitywide devices, capable of connecting many networks - even networks of different types - at a single location. Indeed, most vendors of multislot concentrators offer cards for all varieties of Ethernet, token-ring, and FDDI, as well as network management hardware, bridges, and routers. One recent advance in hub design, socalled "stackable" hubs, adds a backplane connection on to a low-end, eight to twelve port hub making it easy to daisy chain together two or more. Earlier low end hubs had to be connected over the network --- even if they were stacked one atop the other --- taking a up a valuable network port and adding to the traffic on the network too. Stackable hubs have enjoyed a wave of popularity among users over the past year.

Network operating systems are what finally make all that hardware function as a network. Originally, network operating systems functioned only to allow sharing of printers and disk files, and only one workstation at a time could access a particular disk volume. Today's network operating systems allow much more than that, providing the basis for client/server applications, integrating computers of various sizes and types, and allowing the formation of workgroups based on the electronic messaging capabilities of the LAN.

There are two network operating system paradigms - dedicated server-based systems and peer-to-peer systems. Dedicated server networks are hierarchical to a degree, centered around a powerful server machine that stores all files and applications. Specialized servers can be implemented as well - print servers can connect several printers and offer spooling of print jobs to network users without placing any additional load on the file server. Communication servers can connect many users to the outside world, or to the network from remote locations, by providing access to banks of modems. Fax servers can allow users to print files directly to the fax line, eliminating the necessity of getting up and walking to the fax machine to push papers into it. Dedicated applications servers for such processing-intensive software as database management systems typically run on high end .systems under the UNIX or Windows NT operating systems.

Peer-to-peer network operating systems allow all workstations to act as servers. Any user can elect to place any file on his or her hard disk in a "public'' partition that all users can access. While peer-to-peer networks are generally restricted to small network installations, they can provide simple-to-use installation and operation for the small business or workgroup at an economical price.

The local area network knows no application boundaries - nearly any application that can run on a standalone PC can run on a network. Although some applications, such as database management systems and mainframe gateways, can better take advantage of the benefits the LAN has to offer, than, for example, word processing, many of the benefits LANs offer are not so obvious. For example, they allow network managers to ensure that all users are working with the latest version of their application in a comparatively easy manner.

The major benefits of local area networks have been widely touted - increased economies through sharing of expensive resources and increased efficiencies through global access to corporate data and workgroup computing. The risks involved are not as well known, but mainly center around security issues. The argument of the traditional MIS sector has been that with corporate data dispersed around a network on microcomputer platforms, it is much more susceptible to corruption and loss due to both security violations and accidents. That LANs either do not yet implement the degree of security and fault-tolerance associated with the mainframe computer world, or implement them only at prohibitive cost levels, is certainly true. Yet it is only a matter of time before the demands of LAN implementation in Fortune 500-size corporations will make such features mandatory.

At the higher end of the computer business, it can be difficult to identify alternatives to local area networks, since most candidates for that title are widely regarded as obsolete. There is the centralized and hierarchical mainframe or minicomputer network of dumb terminals. There is the heterogeneous collection of isolated PCs, each relying solely on its own resources, where files are shared by hand-carrying diskettes from machine to machine - an arrangement that was dubbed "SneakerNet'' long before LANs had the wide corporate presence they enjoy today. And that is about all.

At the very low end, there is a growing interest in multiuser DOS systems - powerful PCs based on Intel Corp. 386 or 486 processors that support as many as 128 terminals, which can be simple dumb terminals, intelligent color graphics terminals, or other PCs. These multiuser systems can also be attached to local area networks. Activity in this niche market has been brisk enough to spark the formation of the MultiUser DOS Federation by the 20 or so vendors involved in producing and marketing operating systems, multiport serial cards, and graphics terminals.

There was a time when simple printer-sharing devices were touted as a low-cost alternative to a local area network. No more. Today, the price of a simple, peer-to-peer network can be lower than the price of a printer-sharing device, and the advantages enjoyed by network users are much greater.

Likewise, the telephone Private Branch Exchange (PBX) was once viewed as a likely candidate to become the central element in networks that delivered integrated voice and data communications to the desktop. Telephone hardware manufacturers that introduced their own - mostly proprietary - local area networking schemes in the mid-1980s, have seemingly abandoned such development in recent years. Today, most PBXes offer LAN connection to wide area links that can be shared by both voice and data communication traffic.

Rather than seeking an alternative to local area networks, it makes much more sense for potential users of all sizes to realize that there exist a broad array of alternatives within the realm of the LAN itself - from LocalTalk or Arcnet, up through 10BASE-T Ethernet and 16M bps token-ring, through 100M bps FDDI running on copper or on fiber.