Upgrading and Repairing Networks
Chapter 7. Major Network Types
Part 2

Fast Ethernet

As applications and data types become larger and more demanding, a need for greater network throughput to the desktop has become apparent, and several new networking standards have risen out of that need. Although network speeds of up to 100 Mbps have been achievable for some time through the use of FDDI fiber-optic links, this technology remains too complex and expensive for use in connections to individual workstations, in most cases. However, FDDI continues to be a popular choice for network backbone links, particularly when large distances must be spanned or remote buildings on a campus network linked together.

Another technology that promises to deliver increased throughput to the desktop is asynchronous transfer mode (ATM). This new communications standard is commonly recognized as being a dominant player in the future of networking, but the absence of ratified standards has made it far too tenuous a technology to draw major commitments from network administrators at this time. The first ATM products have hit the market, but chances of interoperability between manufacturers are slim, at this time.

The other great drawback of these alternative high speed networking solutions is that they both require a complete replacement of virtually network-related component in the system, from host adapter to hub and everything in between. What was needed for the average corporate network was a system that could allow for the use of cabling already installed at the site, when necessary, and provide a gradual upgrade path for the other networking hardware so that the entire network would not have to be rebuilt in order to perform the upgrade.

To satisfy this need, several competing standards have arisen that utilize existing 10BaseT cable networks to provide 100 Mbps to the desktop. Of these competing standards, only the IEEE 802.3u document, also known as the 100BaseT standard, can truly be called an Ethernet network. Other standards, such as the 100VG (voice grade) AnyLAN network being promulgated by Hewlett Packard and other vendors may run over the same cable types, but they use different methods of media access control, making them incompatible with existing Ethernet networks. 100BaseT utilizes the exact same frame format and CSMA/CD media access and collision detection techniques as existing 802.3 networks. This means that all existing protocol analysis and network management tools, as well as the investments made in staff training on Ethernet networks, can still be used. In addition, backwards compatibility with standard 10BaseT traffic and a feature providing auto-negotiation of transmission speed allow for the gradual introduction of 100BaseT hardware onto an existing 10BaseT network.

Ethernet host adapters that support both 10 and 100 Mbps speeds are becoming commonplace on the market. This market is a highly competitive one, and the additional circuitry required for the adapters is minimal, causing prices to drop quickly as this technology rapidly gains in popularity. Computers with such adapters installed can be operated at standard 10 Mbps speed until a hub supporting the new standard is installed. The auto-negotiation feature will then cause the workstation to shift to the higher transmission rate. In this way, workstations can be shifted to 100 Mbps as the users' needs dictate. This is a rare instance when the network can be made to conform to the user, instead of the other way around.

Indeed, virtually every part of the 100BaseT standard is designed around compatibility issues with existing hardware. Three different cabling standards are provided to accommodate existing networks, and even new installations can benefit from the fact that the fiber-optic wiring standard, for example, is adopted wholesale from the document specifying the wiring guidelines for fiber-optic cable used in FDDI networks. This prevents cable installers from having to learn new guidelines to perform an installation and also keeps prices down. This can be an important factor when you are contracting to have wire pulled for a new site.

Remember that people may have been pulling twisted-pair cable for business office phone systems for decades but still know little or nothing about the requirements for a data-grade installation. Be sure that your contractors are familiar with the standards for the type of cabling that you choose to install and that specific details about the way in which the installation is to be performed are included in the contract. This may include specifications regarding proximity to other service connections, signal crossover, and use of other wire pairs within the same cable, as well as the type and grade of materials employed.

Another factor of a cabling installation that may be of prime importance is when the work is actually done. Unless you are installing a network into brand new space that is not yet occupied, you will be faced with the dilemma of whether or not you should attempt to have the installation performed while business is being conducted. Standards such as 100BaseT have made it possible for network upgrades to be performed without interruption of business. The question of whether to pay overtime rates in order to have cabling installed at night or on weekends or whether to have the contractors attempt to work around your employees is one that must be individually made for every type of business. A good cabling contractor, though, should be able to work nights and still leave an unfinished job site in a state that is suitable for corporate business each day. This is often the mark of a true professional.

100BaseT Cabling Standards

The three cabling standards provided in the 100BaseT specification are designed to accomodate virtually all of the extant cabling installed for use in 10BaseT networks. Obviously, the primary goal is to allow 100BaseT speeds to be introduced onto an existing network without the need for pulling all new cable. Table 7.2 lists the three standards and type of cabling that is called for in each.

All of the standards listed above utilize a similar interface between the actual network medium and the Ethernet port provided by the NIC in the DTE. A medium dependent interface, or MDI, the same as one that can be used for a 10BaseT network, connects to the network cable and is linked to the Ethernet adapter with a physical layer device (PHY) and a media independent interface (MII). These two components may take several forms.

Many 100BaseT host adapter cards are now available that integrate all of these components as circuitry on the expansion card, allowing a standard RJ-45 connection from the network medium to the adapter itself. Other realizations of the technology may take the form of a daughter card that provides switchable 10/100 Mbps capability to an existing 10BaseT adapter. This connects to the network medium utilizing an RJ-45 jack and plug directly into the Ethernet adapter in the host machine. The third possible configuration is through the use of an external physical layer device, much like the seperate MAUs or transceivers used by thicknet systems. The MII of this device then connects to the Ethernet adapter using a short (no more than 0.5 meter) 40-pin cable.

In this way, a number of options are provided to accommodate the networking equipment already installed. Any one of these arrangements can be attached to any one of the designated cable types, providing enough flexibility to allow 100BaseT to be used as a high-speed networking solution. As we examine the three cabling standards in the following sections, notice the way in which they encompass virtually every twisted-pair cabling installation in place today, providing almost universal upgradability.

100BaseTX

Generally speaking, UTP cabling that conforms to the EIA/TIA Category 5 specification is recommended for use by data transmission systems running at high speeds. The 100BaseTX standard is provided for use by installations that have already had the foresight to install Category 5 cable. Using two pairs of wires, the pinouts for a 100BaseTX connection are identical to those of a standard 10BaseT network.

Although the cabling standard for 100BaseTX is based almost entirely on the ANSI TP-PMD wiring standard, the pinouts from the ANSI standard have been changed to allow 100BaseTX segments to be connected directly to existing Category 5 networks without modification. This ANSI standard also allows for the use of 150-ohm shielded twisted-pair (STP) cable, such as that used for token-ring networks. Thus, network types other than Ethernet can also be adapted to the 100BaseT standard, although without the interoperability and auto-negotiation provided to existing 10BaseT Ethernets. Cable of this type, using 9-pin D connectors is wired according to the ANSI TP-PMD specifications.

As with 10BaseT, the maximum segment length called for by the 100BaseTX standard is 100 meters but for a different reason. Segment length on a 10BaseT network is determined by loss of signal strength as a pulse travels over the network medium. Although 100 meters is used as a rule of thumb, a Category 5 10BaseT installation can often include segments of up to 150 meters, as long as the signal strength is maintained. Cable testers of various types can be used to determine whether the installed network maintains the signal strength necessary to extend the segment beyond 100 meters.

For a 100BaseTX segment, however, the 100 meter limitation is imposed to make sure that the round trip timing specifications of the standard are followed. Thus, it is not the strength of the signal, but the amount of time that it takes for the signal to be propagated over the segment that determines the maximum segment length. In other words, 100 meters is a strict guideline that should not be exceeded, even to the point at which the maximum 0.5 meter length of an MII cable (at each end) must be subtracted from the overall segment length.

As with 10BaseT networks, 100BaseTX segments must provide signal crossover at some location on the network. The connections for the transmit pair of wires at one end of the segment, must be attached to the receive connections at the other end, so that proper bi-directional communications can be provided. This crossover can be provided within the hub (in which case a port must be marked with an "X") or within the cable itself.

100BaseT4

In order not to alienate the administrators of the large number of installed 10BaseT networks that utilize voice grade Category 3 cabling, a cabling standard was provided to accommodate 100BaseT on networks of this type. To compensate for the decreased signal strength provided by the lesser quality cable, however, the standard requires the use of four wire pairs, instead of the two used by both 10BaseT and 100BaseTX.

Of the four pairs, the transmit (TX) and receive (RX) wires utilize the same pinouts as 100BaseTX (and 10BaseT). The two additional pairs are configured for use as bi-directional connections, labeled BI_D3 and BI_D4, using the remaining four connectors in the standard RJ-45 jack. Signal crossover for the transmit and receive pairs is identical to that of a 100BaseTX segment, but the two bi-directional pairs must be crossed over as well, with the D3 pair connected to the D4 pair and vice versa. Again, this crossover can be provided by the cable itself or within the hub.

In every other way, a 100BaseT4 segment is configured identically to a 100BaseTX segment. For installations that are limited by the quality of the cable that they are utilizing but which have the extra wire pairs available, creating 100BaseT4 segments can be more economically feasible than pulling new cable for an entire network. Transitional technology such as this allows a network to be gradually upgraded as time and finances permit. As we shall see, different 100BaseT segment types, along with 10BaseT segments, can be easily combined in a single network, allowing additional throughput to be allocated to users as needed.

100BaseFX

The 100BaseFX specification provides for the establishment of fiber-optic link segments that can take advantage of the greater distances and electrical isolation provided by fiber-optic cabling. The medium used is two separate strands of multimode fiber-optic (MMF) cable with an inner core diameter of 62.5 micrometers and an outer cladding diameter of 125 micrometers. Since the crosstalk and signal attenuation problems common to copper cabling are much less of an issue with fiber optic, separate strands of cable are used with no need for twisting, and the crossover connection can be provided by the link connections themselves, rather than inside the hub. A maximum signal loss of 11 dB over the length of the segment is specified by the standard, but the 400 meter maximum segment length is specified, again, by the need for a highly specific maximum round trip signal propagation delay, rather than concern for signal loss.

Several different connector types may be used for the 100BaseFX MDI, again to accommodate the different legacy networks that may be adapted to this technology. The connector type most highly recommended by the specification is the duplex SC connector, although a standard M type FDDI media interface connector (MIC), or a spring loaded bayonet (ST) connector may be used as well. Since they utilize the exact same signaling scheme, the 100BaseFX and the 100BaseTX specifications are collectively known as the 100BaseX specifications.

100BaseT Network Configuration Guidelines

The 100BaseT specification defines two classes of multiport repeaters or hubs for use with all of the various media types. As with the 10BaseT standard, these devices are defined as concentrators that connect disparate network segments to form a single collision domain, or network. A Class I hub can be used to connect segments of different media types while a Class II hub can only connect segments of the same media type. The standard dictates that the different hub types must be labeled with the appropriate Roman numeral within a circle, for easy identification.

The fundamental 100BaseT rules for connecting segments within a single collision domain are

• Two segments of the maximum length called for by the individual media types can be connected via a Class I or Class II repeater (subject to some further limitations for certain media types)
• No more than one Class I or two Class II repeaters may lie in the path between any two networked DTEs

Table 7.3 lists the maximum segment lengths allowed according to media type and repeater type:

As you can see, the copper media types provide for fairly consistent limits throughout the various configurations, but the introduction of fiber-optic cable extends the length limitations. The one exception to this is the 205 meter limit when connecting copper segments with two Class II repeaters. This figure is valid for Category 5 cabling only. Voice grade Category 3 cable is limited to an overall length of 200 meters.

Class I repeaters generally provide greater amounts of delay overhead when translating signals for use with the various media types, so they impose greater segment length limitations than repeaters of the Class II variety. When mixed networks using both copper and fiber segments are defined, the figures provided in the table assume a 100 meter copper segment as contributing towards the total listed. It should also be noted that, for all of the network types, the maximum total one meter length of any MII cables used (0.5 meters at each end) must be counted towards the total length of the segment.

These quibbles over what seem to be inconsequential variances in segment length should indicate how tightly these estimates are integrated into the 100BaseT specifications. 10Mbps networks do not tax the medium to the degree at which 100 Mbps ones do, and so a certain unofficial "fudge factor" can be assumed to exist on the slower systems. Long experience has determined that many standard Ethernet networks continue to function acceptably, despite physical layer installations that exceed the recommended specifications. 100BaseT is far newer technology, however, and a far narrower margin for variation is provided. It is recommended that these limitations be adhered to quite stringently, at least until experience has determined where variations can safely be made.

As with traditional Ethernet, the 100BaseT standard provides the means by which individual network segment limitations may be calculated mathematically. The primary limiting factor for 100BaseT, however, is the Path Delay Value, which is a measurement of the round trip signal propagation delay of the worst case path—that is, the two stations on the network that are the greatest distance apart, with the greatest number of repeaters between them. Cable delay values for the specific types of media used to form the network, along with the distances spanned, and the number of repeaters, are plugged into a formula, an extra margin is added for additional safety, and a specific value is derived that indicates whether or not the network meets the requirements of the 100BaseT standard. As with 10BaseT, exceeding the recommended values can result in late collisions and packet CRC errors that severely affect the efficiency and reliability of the network.

Obviously, the limitation in the number of repeaters allowed in a 100BaseT collision domain over that of 10BaseT will require a certain amount of redesign in some networks being retrofitted to the faster system. An existing network that is stretched to the limit of the 5-4-3 10BaseT guideline may have to have its repeaters relocated to confirm to the new restrictions, but the increased segment lengths allowed for most of the cabling types should make the task a possible one for most existing installations. In any case, it should be clear that migrating to 100BaseT is more than just a matter of replacing NICs and hubs.

Ethernet Switching

Note also that the limitations detailed earlier apply only to segments within a single collision domain. Hubs that provide packet-switching services between the segments are becoming increasingly popular, and have come to make up a large portion of the market. A packet-switching hub essentially provides the same services as a repeater, but at a higher level. Packets received on one segment are regenerated for transmission via another. All of the OSI model from the network layer up is shared by the two segments, but the data link and the physical layer is isolated, establishing separate collision domains for the two and providing what is essentially a dedicated network for each port on the device.

In this way, a centrally located switch can be used to provide links to multiport repeaters at remote locations throughout the enterprise. These repeaters are then linked to individual workstations in the immediate area. The network, in its most strictly defined sense, extends from the switch port to the DTE, with only one intervening repeater. More demanding installations may even go so far as to use switched ports for the individual desktop connections themselves, thus providing the greatest possible amount of throughput to each workstation. As you may expect, a packet switching hub will be more expensive than a simpler repeating device, but they may be the most economical means of adapting an existing network to today's requirements. Just the time and expense saved by not having to replace dozens of LAN adapters in workstations all over the network may be enough to lure administrators towards this technology.

Bear in mind that this switching technique is by no means limited to networks using 100BaseT. Switches are becoming a popular solution for 10BaseT and even token-ring networks. In fact, adding switches to a 10BaseT network may provide enough additional performance to obviate the immediate need for a large scale network upgrade program.

Full Duplex Ethernet

Another technique that is being used to increase the efficiency of both 10BaseT and 100BaseT links is the establishment of full duplex Ethernet connections. Ethernet networks normally communicate using a half duplex protocol. This means that only one station at a time can be transmitting over the network link. Like a two-way radio, a single station on the network may transmit and then must switch into a listen mode to receive a response. Managing this communications traffic without the loss of any data is the basic function of a media access control mechanism like CSMA/CD.

Full duplex Ethernet, on the other hand, functions more in the way that a telephone does, allowing both ends of a link segment to transmit and receive simultaneously, theoretically doubling the overall throughput of the link. For this reason, the entire Ethernet media access control system can be dispensed with when a full duplex link is established. In order to establish such a link, only two stations can be present in the collision domain. Like a party line telephone system, chaos would ensue if more than two parties were all speaking at the same time. Therefore, full duplex Ethernet is usually used to connect two packet-switched ports on remote hubs. The elimination of the media access protocol also removes the need for any concerns about signal propagation across the link, so very long distances may be spanned by 10BaseF or 100BaseFX links. The only limitation would be imposed by signal loss due to attenuation which, in fiber-optic connections, is minimal. Links of this type can span up to two kilometers or more and form an excellent means of connecting remote buildings on a campus network.

It must be noted that the full duplex Ethernet has not been standardized by the IEEE or by any other standards body. Individual hardware vendors are responsible for creating and marketing the concept. There may, therefore, be significant variations among different vendors in the rules that for establishing such links. Compatibility of hardware made by different manufacturers is also not guaranteed. In addition, you will find that a full duplex link will generally not deliver the doubled throughput that theory dictates should result. This is because some of the higher layer network protocols in the OSI model also rely on what are essentially half duplex communication techniques. They cannot, therefore, make full use of the capabilities furnished by the data link layer, and the overall increase in throughput may be limited to somewhere between 25% and 50% over that of half duplex Ethernet. The cost of implementing full duplex into existing adapter and hub designs, however, is minimal, adding no more than 5% to the cost of the hardware. Therefore, even the moderate gain in throughput provided may be well worth the cost involved.

Auto-Negotiation

As on 10BaseT networks, 100BaseT utilizes a link pulse to continually test the efficacy of each network connection, but the fast link pulse (FLP) signals generated by 100BaseT adapters are utilized for another function as well. Unlike the normal link pulse (NLP) signals generated by 10BaseT, which simply signal that a proper connection exists, FLP signals are used by 100BaseT stations to advertise their communications abilites.

At the very least, an indication of the greatest possible communications speed is furnished by the FLP, but additional information may be provided as well, such as the ability of the station to establish a full duplex Ethernet connection and other data useful for network management. This information can be used by the two stations at either end of a link segment to auto-negotiate the fastest possible link supported by both stations. Although an optional feature, according to the 100BaseT standard, auto-negotiation is a popular option, considering the large number of hubs and adapters coming to market that can support both 10 and 100 Mbps speeds. Several different approaches to the inclusion of additional functionality into the FLP exist, however. One of these that has received a good deal of attention is called NWay. Developed by National Semiconductor, NWay must reside in both the adapter and the hub for the full auto-negotiation capabilities to be utilized. Many vendors are considering it for inclusion in their products, but until a standard for this technology is realized, either by an official governing body or simply by vox populi, these must be considered to be proprietary techniques and evaluated as such.

Since auto-negotiation is optional, there is more control provided over the generation of the link pulse signals than with 10BaseT. Settings are usually made available at each device to allow the pulses to be generated automatically when the device is powered up, or it may be implemented manually. Fast link pulses are designed to co-exist with the normal link pulses so that negotiation may take place with existing 10BaseT hardware as well. A traditional 10BaseT hub with no knowledge of auto-negotiation, when connected to an Ethernet adapter capable of operation at both 10 and 100 Mbps, will cause a link to be established at the slower speed and normal 10BaseT operations to continue without incompatibilities. This allows network managers to implement an upgrade program in any manner they choose. At this point in time, anyone with intentions of upgrading to 100BaseT should begin purchasing dual-speed Ethernet adapters for any new systems being installed. This way the replacement of a 10BaseT hub with a 100BaseT model can be performed at any future time desired, and the appropriately equipped workstations will shift to the higher speed connection as soon as the new equipment is detected. As with 10BaseT systems, the pulses are only generated during network idle periods and have no effect on overall network traffic.

The auto-negotiation feature, when it is enabled, determines the highest common set of capabilities provided by both stations on a link segment, according to following list of priorities, and then creates a connection using the highest priority protocol of which both sides are capable:

1. 100BaseTX Full Duplex
2. 100BaseT4
3. 100BaseTX
4. 10BaseT Full Duplex
5. 10BaseT

Notice that although 100BaseT4 and 100BaseTX are both capable of the same transmission speed, 100BaseT4 is given the higher priority. This is because it is capable of supporting a wider array of media types than 100BaseTX. A segment with hardware at both ends that supports both transmission types will default to 100BaseT4, rather than 100BaseTX, unless explicitly instructed otherwise.

When auto-negotiating hubs are used that are of the multiport repeater type, it must be noted, however, that since only a single signal is generated for use on all of the device's ports, the highest common speed of all of the devices connected to the hub will be used. In other words, a hub with ports connected to eleven DTEs with 100 Mbps network adapters and one DTE with a standard 10BaseT adapter will run all of the stations at 10 Mbps. A packet-switching hub is, of course, not subject to this limitation. Since each of its ports amounts to what is essentially a separate network, individual speed negotiations will take place for every port.

100VG AnyLAN

The primary source of competition to the 802.3u Fast Ethernet standard in the battle of the 100 Mbps networking specifications is known as 100Voice Grade AnyLAN, as defined in 801.12 IEEE standard. Championed by Hewlett Packard and AT&T, as well as several other companies, it is, as the name implies, a networking standard that, like 100BaseT, provides 100 Mbps throughput but is specifically designed to take advantage of the existing voice grade Category 3 wiring that is already installed at so many network sites. Like 100BaseT4, the lower grade of cabling requires the use of four wire pairs instead of two, but beyond this, 100VG AnyLAN is radically different from 100BaseT.

First of all, 100VG AnyLAN cannot, by any means, be called an Ethernet network. In fact, it is a new protocol that is unique to the networking world and this, if anything, is its greatest drawback. All of the investments in time and money made on Ethernet or token-ring training along with management and troubleshooting tools for these environments are lost when you convert to 100VG AnyLAN. In addition, the standard is based on the assumption that the greatest single investment made in a network is in the cable installation. The basic philosophy of 100VG AnyLAN is to use a network's existing cable plant, including the existing RJ-45 jacks and cross connectors, but all other components, including hubs and adapters, must be replaced.

The cutthroat competition over these competing 100 Mbps standards is the result of users' clamor for a convenient and economical upgrade strategy for their networks. After all, FDDI and CDDI networks providing the same throughput have been available for years, but the expense and labor involved in converting to a network that runs such technology to the desktop has remained the prohibitive factor preventing its widespread acceptance. Both 100BaseT and 100VG AnyLAN provide more reasonable upgrade capabilities than FDDI and CDDI, providing the means for a gradual conversion spread out over as long a period of time as desired. Individual workstations can be upgraded to 100VG AnyLAN as the user's need arises for, as with 100BaseT, there are combination adapters available that provide plugs for both 10BaseT and AnyLAN.

In addition, the same Ethernet packet format as 10BaseT is used by 100VG AnyLAN, allowing hubs for the two network types to co-exist on the same network. Although the Ethernet frame type is being supported first, there are also plans for 100VG AnyLAN hubs supporting the 802.5 frame type used by token-ring networks to be made available, as well as units supporting both packet types. The signaling scheme and the media access control protocol used by 100VG AnyLAN, however, are different from those used by any other network.

As a general rule, the overall similarity of the 100BaseT hardware to its 10BaseT counterparts will allow compatible equipment to be developed and produced more quickly and less expensively than that for 100VG AnyLAN. A combination 10BaseT/100VG AnyLAN NIC, for example, actually amounts to the components of two separate adapters on one card, while a 10/100BaseT NIC can utilize some of the same components for both functions to keep costs down. The same holds true for 100BaseT hubs and bridges, which are little more complex than 10BaseT models with the same capabilities. Also, a great many more vendors are currently producing 100BaseT hardware than 100VG AnyLAN, and many more systems manufacturers have declared their preference for its use than have advocated the other standard, giving it an immediate price advantage in the marketplace and a superior collection of testimonials.

These are all very young products, however, and it is difficult to predict the direction the pendulum will swing. Some hardware manufacturers are planning to produce equipment for both network types, refusing to take a definitive stand for either one over the other. Others are attempting to combine the functionality for both networks in single devices, allowing the administrator to choose one of the two network types depending on the needs of the individual user. I dare say, though, that one of these network types will prove to be the dominant interim solution, as network administrators everywhere lick their chops in anticipation of ATM, which nearly everyone agrees will eventually come to dominate the networking industry, at some point one, two, five, or ten years down the road, depending on whose opinion you believe.

On the other side of the argument, however, is the fact that, despite the unavailability of long-term, real-world performance data, early reports indicate that 100VG AnyLAN generally provides a greater increase in network throughput than 100BaseT does. This is primarily because 100BaseT is subject to the same latency problems and tendency towards diminished performance under high-traffic conditions that normal Ethernet is. The technology that 100VG AnyLAN is based on provides nearly the entire potential throughput of the segment to each transmission.

Obviously, choosing one of the two standards is a complex decision, which must balance the need for maximum throughput versus a more solid, competitive, and economical market for the required hardware and factor in the need for staff training in order to support this new protocol. The following section will examine how 100VG AnyLAN provides this allegedly superior level of performance, and provide background information to aid in the decision-making process between the two competing standards.

Quartet Signaling

A 10BaseT network utilizes two wire pairs for its communications. One is used to transmit, and the other for collision detection. The 100BaseT4 standard uses four pairs of wires, with the extra two pairs usable for communications in either direction. 100VG AnyLAN also uses four wire pairs, but it utilizes a technique called quartet signaling that allows it to transmit over all four wire pairs simultaneously. The encoding scheme used, called 5B/6B NRZ, allows the number of bits transmitted per cycle to be two-and-a-half-times greater than that of 10BaseT networks. Multipled by the four pairs of wires used to transmit, this results in a tenfold overall increase in transmission speed, using only a slightly higher frequency than 10BaseT, thus allowing the use of voice grade cabling.

Demand Priority

The basic reason why all four wire pairs can be used to transmit simultaneously is that 100VG AnyLAN eliminates the need for a collision detection mechanism such as that found on Ethernet networks. The media access control method utilized by 100VG AnyLAN is called demand priority, and while it is radically different from the CSMA/CD method used by Ethernet, it makes a good deal of sense for the environment that it's used in.

As we have seen, the 10BaseT network standard is an adaptation of a protocol that was originally designed for a bus topology composed primarily of mixing segments, on which multiple nodes must contend for the same bandwidth. Networks wired in a star topology, however, are composed primarily of link segments. While the 802.3 standard was ingeniously adapted to the star configuration by designating the link segments for connection of the hub to the node, and the mixing segments for the interconnection the hubs, the primary sources of possible media contention are the network workstations. When the workstations are connected to a hub using link segments, negotiation for media access need only be conducted between two different entities (while on a mixing segment, up to 30 entities can be contending for the same bandwidth). 100VG AnyLAN takes advantage of the star topology by having intelligence within the hub control access to the network medium.

Demand priority calls for individual network nodes to request permission from the hub to transmit a packet. If the network is not being used, the hub permits the transmission, receives the packet, and directs it to the proper outgoing destination port. Unlike Ethernet, where every packet is seen by all of the nodes within a given collision domain, only the transmitting and receiving stations, along with the intervening hubs, ever see a particular AnyLAN packet, thus providing an added measure of security unavailable from traditional Ethernet, token-ring, or FDDI networks.

Since arbitration is provided by the hub, priorities for certain data types can also be established, allowing particular applications to be allotted an uninterrupted flow of bandwidth, if desired. For real-time multimedia applications such as videoconferencing, where careful flow control is required, this can be a crucial factor to good performance. As with token ring, there are no collisions on a 100VG AnyLAN network that is running properly. There are no delays, therefore, caused by packet retransmission's and no cause for network performance to decrease as traffic increases.

Integration with 10BaseT

100VG AnyLAN can also be integrated into a 10BaseT segment through the use of bridges that buffer the higher speed transmissions, feeding them to the slower medium at the proper rate. This technique can also be used to attach 100VG networks to an existing backbone. No packet translation of any kind is necessary, which avoids any delays that would normally be incurred by this process and allows the necessary bridging circuitry to be easily incorporated within the hub if desired.

Overall, 100VG AnyLAN requires a higher degree of commitment from the network administrator than 100BaseT does. The hardware is much less reliant on tried-and-true technology and the innovative nature of the standard implies a greater risk as the marketplace determines whether the concept will continue to be a viable one. For the many administrators who are considering these 100 Mbps technologies as interim solutions for their networks, it would be understandable for them to be reticent to expend the time, effort, and expense to adapt to a new network type that would only be phased out within a few years. Current indications point to 100BaseT as being far more widely accepted by the industry than 100VG AnyLAN, but there are major industry players advocating both systems, and both or neither could come to dominate the high speed networking world over the next few years.

The Workstation Bus

It should be noted that, for any network offering 100 Mbps performance levels, the ISA bus will generally be insufficient to support the needs of the network interface. 100BaseT network adapters are currently available only for EISA and PCI buses, and testing of various cards for both bus types made by the same manufacturers yields very little performance difference between the two. It should therefore not be necessary to upgrade from an EISA to a PCI machine simply to take full advantage of 100BaseT.

Adapters for the VESA Local Bus are not being produced, primarily because vendors have achieved the best performance levels from adapter designs using the bus mastering capabilities supported by the EISA and PCI buses to prevent network data from having to be moved on and off of the card in order to be manipulated by the system processor. Avoiding any additional burden on the processor also helps to increase overall system performance. High performance 100BaseT cards may also offer SRAM FIFO (high-speed static RAM, first in, first out) caching, coprocessors and dedicated chips providing increased performance for the adapter's media access control functions.

While ISA cards for 100VG AnyLAN do exist, their generally poor performance levels seem to indicate that the system bus is a probably the location of a greater bottleneck than any caused by the network. Obviously, the benefits and drawbacks of all of the available buses are as applicable to LAN adapters as they are to SCSI or video cards. Consult chapter 5, "The Server Platform," for in-depth coverage of the attributes of the different bus types.

Assessing User Need

Of course, as with any network-related upgrade, the question arises as to the real need for 100 Mbps to the desktop. Depending on the other hardware involved, the operating systems used by the servers and workstations, and the size and quantity of the files transmitted over the net, the overall increase in productivity provided by this type of network upgrade may prove to be negligible. This technology excels primarily in the sustained transfer of large files over the network medium. For applications that generate large amounts of network traffic, such as scientific, engineering, prepress and software development environments, this may be a boon, and for networks that have been continually expanded in size and traffic levels without an increase in throughput, a significant performance bottleneck may be removed.

Most general use business networks, however, are relatively empty, and the thoughtful LAN administrator faced with a slow network must be careful to determine exactly what is causing the slowdown before committing to a costly upgrade program. From a practical standpoint, Ethernet traffic problems can probably be more efficiently and economically addressed with the addition of Ethernet switches and a proper evaluation and reorganization of the network plan. A wholesale replacement of all hubs and adapters is probably not necessary just to have a properly functioning business network environment.

As to the new multimedia data types that are threatening to overwhelm traditional networks, if an administrator were to honestly ask whether their users really had a productive need for full motion video to the desktop, the answer would probably be no. Just because a new technology becomes available, doesn't mean that we should all go out and search for some way to put it to use. In fact, even full motion video can be adequately delivered to the desktop over the network, when 10 Mbps of dedicated bandwidth is supplied. When the networking industry marketing machine goes into a feeding frenzy over a new technology like Fast Ethernet, it can be difficult to find a clear path through the carnage in order to see whether the new product actually make things better than they were before. This is a question that every LAN adminstrator must answer individually for every network that they are responsible for.

Token Ring

Barring the new networking technologies now gaining widespread attention in the marketplace, the traditional alternative to Ethernet has been the token-ring network. Originally developed by IBM, who still remains its primary champion, token-ring networks can deliver data at 16 Mbps using a media access control mechanism that is radically different from the CSMA/CD scheme used by Ethernet.

The IEEE 802.5 standard defines a token-ring network. The standard was deliberately developed to be an alternative to other 802.x media access control specifications, all of which utilize the same logical link control protocol defined in the 802.2 standard. Unlike the bus and star topologies utilized by Ethernet networks, token ring, as the name implies, organizes its connections in a logical ring topology so that packets can be passed from node to node in an endless rotation.

It is called a logical ring topology because the network is actually wired according to the same sort of star arrangement as a 10BaseT network. The ring exists only within the hubs to which all of the nodes are attached (see fig. 7.10). Usually known as multistation attachment units, MSAUs (sometimes improperly called MAUs), or simply wiring centers in the 802.5 document, token-ring concentrators provide more functionality than the multiport repeaters used for Ethernets. A token-ring MSAU monitors the existence of each node attached to its ports. A packet originating at any station is passed to the MSAU, which passes it in turn to the next station in the ring. That station returns it to the MSAU, which continues passing the packet around the ring until each attached node has received it. The packet is then removed from the ring by the node where it originated from.

For this system to work, the MSAU must be constantly aware of the operation of each attached node. Should a packet be sent to a non-functioning workstation, it will not be returned to the MSAU and therefore cannot be sent along on its way. Therefore, MSAUs continuously monitor the activity of all of the attached workstations. A node that is switched off or malfunctions is immediately removed from the ring by the bypass relays in the MSAU, and no further packets are sent to it until it signals its readiness to continue.

The original token-ring MSAUs developed by IBM actually used a mechanical device to control access to each port. Before attaching the cable that was connected to a workstation, the network administrator had to initialize the port to be used with a keying device that entered that port into the ring. In addition, the actual network medium and the connectors used were all of proprietary IBM design. The cabling was quite thick (a good deal thicker than the 50-ohm coaxial used for thin Ethernet), and the connectors large and unwieldy. Cables were also sold only in prepackaged form, and were available in a limited assortment of lengths. Since there was no competition at the time, the prices of all of the hardware components were gratuitously inflated. The original token-ring networks were also designed to run at a maximum speed of 4 Mbps.

It should be clear that, if these conditions had not changed considerably, token ring would have gone the way of StarLAN and ARCnet on the slow road to oblivion. However, the modern token-ring network is considerably more advanced than this, and now can provide 16 Mbps of throughput in what is, arguably, a manner that is better suited to high traffic networks than Ethernet. First of all, the preferred network medium for token ring is now type 1 shielded twisted-pair cabling (STP) that is similar to the UTP used in 10BaseT networks, except for additional insulation surrounding the twisted strands, thus providing increased resistance to EMI as well as a higher cost for the medium and its installation. The connectors used may be of the familiar RJ-45 telephone type, but DB-9 connectors (such as are used for the serial ports on PCs) can also be used. At the MSAU, self-shorting IBM data connectors are usually used.

The 802.5 document states that up to 250 stations can operate on the same ring, but it defines no specifics for the type of cable to be used, and real-world performance figures depend highly on the type of cable and the round-trip lengths of the segments extending from the MSAU to the connected node (called the lobe length). Token ring can even be run over conventional UTP cable, although the number of attached nodes and the lobe lengths will be further limited by the lesser capabilities of the medium. The IBM Token Ring specifications allow up to 260 lobes on an STP segment and up to 72 on a UTP segment. The following list describes the cable types, as defined by IBM, which have come into general use in the network industry.

• Type 1 Two pairs of 22 AWG cable with foil wrapped around each pair and a second foil sheath around both pairs, along with a ground. The most commonly used STP cable for token-ring networks.
• Type 2 Same as type 1 with an additional four pairs of voice grade cable attached to the outer sheath. Designed for single cable delivery of both telephone and data services.
• Type 3 Four pairs of unshielded 22 or 24 AWG cable, twisted at least twice per foot. The IBM classification for the UTP cabling used in many 10BaseT networks.
• Type 4 There is no Type 4.
• Type 5 Fiber-optic cable.
• Type 6 Two pairs of stranded, shielded 26 AWG cable used for patch or jumper cables.
• Type 7 One pair of stranded, 26 AWG cable.
• Type 8 Two pairs of 26 AWG cable run in parallel (that is, with no twisting). Used for under-carpet installations.
• Type 9 Two pairs of shielded, 26 AWG cable. Used for data, despite decreased transmission capabilities due to narrower guage.

Generally speaking, the maximum allowable lobe length for cable types 1 and 2 is 200 meters, for type 3: 120 meters, for type 6: 45 meters, and type 5 can be up to 1,000 meters in length. The term lobe length is defined as the round-trip signaling distance between a workstation and a MSAU. A 100 meter cable connection therefore yields a lobe length of 200 meters. No more than three segments (joined by repeaters) can be connected in series, and there can be up to 33 MSAUs on a single network. If these last two statements seem contradictory, it is because when multiple token-ring MSAUs are interconnected to form a single ring, they do not comprise separate segments, as would the use of multiple hubs in an Ethernet network. Also, all of the nodes on a particular network must run at the same speed. Dual speed token-ring NICs and MSAUs, which run at either 4 or 16 Mbps, are common. However, all of the ports on a single MSAU must run at the same speed and can be connected to another MSAU running at a different speed only by a bridge or a router so that two separate network domains are established.

MSAUs are also considerably more advanced than they were in the early days of token ring, with models available that provide similar features to the higher-end Ethernet concentrators. In addition, the current move of the industry towards switching over bridging or repeating also applies to token ring. Complex switches are even available that allow the connection and routing of data over multiple network types from the same device. Token ring, Ethernet, Fast Ethernet, FDDI, and even ATM networks can all be connected to the same device, and a packet entering through any port is routed directly out the port where the destination address is located, after being translated to the proper signalling type for the destination network.

Multiple MSAUs can also be interconnected to form a single ring. In a token-ring environment, a ring would be the equivalent of a single collision domain or network in Ethernet parlance. Since collisions are not a normal occurrence on token-ring networks, the term collision domain is not valid, but a ring consists of a group of DTEs interconnected so that the same network segment is shared by all. Every MSAU has special ports, labeled Ring In (RI) and Ring Out (RO), which are designated for connection to other MSAUs, allowing you to create large rings of up to 250 nodes.

Media Access Control in Token-Ring Networks

As we have seen, the real key to the functionality of any network is how multiple stations can communicate using a shared network medium. Unlike Ethernet, in which each DTE essentially executes an independent instance of the accepted MAC protocol for its own use, the MSAU arbitrates token ring's media access.

Essentially, media access in a token-ring network is controlled by the passing of a specialized packet, or token, from one node to the next around the ring. Only one token can be present on the ring at any one time and that token contains a monitor setting bit that designates it as a "busy" or "free" token. Only a node in possession of a free token may transmit a frame.

When a workstation is ready to transmit a packet, it waits until a free token is sent to it by the preceding node in the ring. Once it has been received, the transmitting node appends its packet to the token, changes the monitor setting to busy, and sends it on its way to the next node in the ring. Each node in turn then receives the packet and passes it along, thus functioning in normal repeat mode, the functional equivalent to a unidirectional repeater.

Whether the packet is destined for use by that workstation or not, the packet is passed to the next node. Having traversed the entire ring, the packet then arrives back at the node that originated it. This node then reads the packet and compares it with what it had previously transmitted, as a check for data corruption. After passing this test, the originating workstation then removes the packet from the ring, generates a new free token and sends it to the next node, where the process begins again.

Some token-ring networks also support a feature called early token release (ETR), in which the sending node generates a free token immediately after it finishes transmitting its packet. Thus, the packet is sent without a busy token included, but with a free token following immediately after. The next station in the ring receives the data packet and the token, and may then pass on the first data packet, transmit a data packet of its own, and then transmit another free token. In this manner, more than one packet may be traveling around the ring at any one time, but there will still be only one token. This eliminates the waiting periods incurred as tokens and packets are passed from station to station.

Thus, in theory, a collision should never occur on a token-ring network. Packets may be transmitted at the maximum rate allowed by the MAC protocol with no degradation of network performance. This is why many people consider token ring to be a superior type of network for heavy traffic environments. As an Ethernet network becomes busier, a larger number of collisions occur, forcing a greater number of retransmissions, and therefore, delays. On a token-ring network, although there is a greater amount of overhead traffic generated by maintenance functions, the maximum possible delay incurred before a given station can transmit is the period that the node must wait for a free token to be passed to it. The greater the traffic on the network, the longer this delay will be, but there is no additional traffic generated by the retransmission of packets damaged by collision, allowing the network to utilize virtually its entire bandwidth for authentic non-redundant traffic.

Token-ring stations are also capable of utilizing different access priority levels so that specific stations can be configured to be more likely to receive a free token that it can transmit with. The later discussion of the 802.5 frame types in this chapter covers how these priorities are exercised. There are also automatic mechanisms in a token-ring network that provide the means for recognizing and localizing error conditions on the network.

When any station on a ring detects a problem, such as a break in the ring, it begins a process called beaconing that helps isolate the exact location of the problem. Beacon frames are sent out over the network, which define a failure domain. The failure domain consists of the station detecting the failure and its nearest active upstream neighbor (NAUN). If there are any stations located between these two, they must, by definition, be inactive, and are designated as the locations of the failure. An auto-reconfiguration process then begins; active stations within the failure domain activate diagnostic routines in the hope of bypassing the offending nodes, allowing communications to continue. Depending on the cause of the problem, the network may ultimately be halted, or it may continue to operate by removing problem stations from the ring.

As a means of monitoring and maintaining the network, one node on the ring acts as an active monitor. This station functions as the instigator for most of the ring control and maintenance procedures conducted by the network. Since all stations are capable of generating a token, for example, there must be one station that generates the first token, in order to start the process. This is one of the functions of the active monitor. It also initiates the neighbor notification process—each node on the network learns the identity of it nearest active upstream and downstream neighbors, provides timing services for the network, checks for the existence of packets circulating continuously around the ring, as well as performing other maintenance functions.

Any station may become the active monitor through a process called token claiming that is initiated whenever any station (or standby monitor, or SM) on the network fails to detect the existence of a frame or an active monitor (through the receipt of an active monitor present, or AMP MAC frame) within a designated amount of time. Token claiming consists of each SM sending out specialized frames based on address values. The first SM to receive three of its own frames back is designated the active monitor (AM). In this manner, the active monitor constantly checks the network, and the other nodes constantly check the active monitor to ensure that the network access mechanisms are always functioning properly.

There are many other functions defined in the network management protocol (NMP) defined in the 802.5 standard document, some of which may be performed by the active monitor or by other stations on the network, which may or may not be wholly dedicated to such a purpose. These include the Ring Parameter Server (RPS), which monitors the addresses of all nodes entering and leaving the ring; the Ring Error Monitor (REM), which tracks the occurrence and frequency of errors occurring on the ring; the LAN Bridge Server (LBS), which monitors the activities of all bridges connected to the network; and the Configuration Report Server (CRS), which gathers performance and configuration from other nodes on the network. All of the information generated by these functions can be sent to a node that has been specifically designated as the network management node by the running of software designed to compile, track, and analyze all of this data and adjust the network's performance characteristics accordingly.

The 802.5 Frame Format

Unlike Ethernet, which uses one basic frame type for all of its functions, the IEEE Token Ring standard defines several basic frame formats (see fig. 7.11), which are used for many different functions: a data/command frame, a token frame, and an abort sequence frame. The data/command frame is a single frame type that can be used both for transfer of LLC data to upper level protocols and for MAC information used to implement one of dozens of ring maintenance control procedures. Only the data frame contains information that is destined for use by protocols higher up in the OSI model. All of the other frame configurations are used solely for maintaining and controlling the ring.

Figure 7.11 The fields that comprise the IEEE 802.5 data/control frame.

A token frame, three bytes long, consists of only the Start Delimiter, Access Control, and End Delimiter fields, just as previously defined. The abort sequence frame, used to clear the ring when a premature end to the transmission of a packet is detected, consists only of the Start and End Delimiters, as previously defined. These two frames are used only for control and maintenance of the 802.5 protocol.

The Downside to Token Ring

The primary drawback to a token-ring network is the additional expense incurred by the higher prices for virtually every hardware component required for its construction. Throughout its history, token ring has been dominated by IBM, which has functioned as the trendsetter for the technology far more than standards bodies like the IEEE have. Throughout its history, it has usually been IBM that was first to release innovations in token-ring technology, such as the increased 16 Mbps transmission rate, only to have them assimilated into the published standards at a later time. Indeed, the 802.5 document is very brief (less than 100 pages) when compared with the 802.3 standard. There are also fewer vendors and therefore less competition in the token-ring hardware market than there are in that of Ethernet.

Token-ring adapters can cost two or three times more than Ethernet adapters, with similar markups applied to MSAUs and other ancillary hardware. Token ring also offers fewer convenient throughput upgrade paths than Ethernet does. Migration to a 100 Mbps technology will require the wholesale replacement of virtually the entire network, except for the cabling itself. For these reasons, Token ring has remained second to Ethernet in popularity, with approximately 10 million nodes installed worldwide, but its proponents are earnest and quite vocal, and its capabilities as an efficient system for business networking, incontestable.

ARCnet

Although it is hardly ever used in new installations these days, the Attached Resource Computer Network (ARCnet) is another networking standard for the physical and data link layers of the OSI model. Introduced by the Datapoint Corporation in 1977, SMC has been the primary ARCnet vendor since 1983. Running at 2.5 Mbps, ARCnet is the slowest network of those considered in this chapter. This is one of the primary causes of its unpopularity because, otherwise, ARCnet is capable of providing the same basic network services as Ethernet and token ring at far lower costs and with a great deal of physical layer flexibility.

ARCnet can be wired in a bus topology, using RG-62/U coaxial cable and BNC connectors (also known as high impedance ARCnet), or in a star topology, using UTP or IBM Type 1 cabling with RJ-45 or D-shell connectors (also known as low impedance ARCnet). Hybrid networks of mixed bus and star topologies (also known as a tree topology) can also be assembled, consisting of nodes daisy-chained with twisted pair cable connected to a hub that connects to other hubs using coaxial cable. ARCnet is very forgiving in this respect. As with the other network types, care must be taken to properly terminate all segments, using a 93-ohm resistor pack for coaxial buses and a 105-ohm resistor for twisted pair. (Note that the 92-ohm resistor pack differs from the 50-ohm terminators used by thinnet, although they may be virtually identical in appearance). Even fiber-optic cable can be used with ARCnet.

The connectors used for ARCnet are standard BNC connectors for coaxial cable. Twisted pair can utilize RJ-45 connectors or the standard D-shell connectors used by the serial ports on PCs. Connection boxes called active links are used to connect high-impedance cable segments, and baluns are available for providing an interface between coaxial and twisted-pair cable types.

Three types of ARCnet hubs are available. Active hubs, containing anywhere from 8 to 64 ports, have a power supply and function as a repeater as well as a wiring nexus. Passive hubs, which have only four ports, use no power and function simply as signal splitters. Intelligent hubs are also available, which are capable of monitoring the status of their links.

A high-impedance (coaxial) ARCnet network must use only active hubs. Segments connecting two stations can span up to 305 meters, while segments connecting hubs can extend 610 meters. Up to eight nodes can be connected in series without an intervening hub, and there must be at least one meter of cable between nodes. A low-impedance ARCnet network can use both active and passive hubs. A segment connecting an active hub and a node or two active hubs can span up to 610 meters, while a segment connecting a node or an active hub to a passive hub can be no more than 30 meters. Passive hubs can only be located between active hubs and nodes. Two passive hubs can never be directly connected to each other. High and low impedance network segments can also be mixed on the same network, provided that the limitations for each are observed. Up to 10 nodes can be connected in series when UTP cable is used.

The maximum limitations for any ARCnet network are 255 nodes (active hubs count as nodes) and a total cable length of 6,000 meters. Maximum segment lengths may vary depending on the type of cable used but a maximum of 11 dB of signal attentuation over the entire segment at 5 MHz is all that is allowable. Two connected nodes must also have a signal propagation delay of no more than 31 microseconds.

Unlike most network types, the node addresses of ARCnet networks must be manually set (from 1 to 255) on the NICs through the use of jumper switches. Address conflicts are, therefore, a distinct possibility, resolvable only by manual examination of all of the NICs. The adapter with the lowest numerical node address automatically becomes the network's controller, similar in basic function to the active monitor on a token-ring netwok.

Like token ring, ARCnet uses a token-based media access mechanism. A token is generated by the controller and sent to each station in turn, giving them the opportunity to grab the token and transmit. ARCnet, however, uses a far less efficient signaling scheme to arbitrate the token passing. Once a token is grabbed, a query and an acknowledgment must be exchanged by the sending and receiving stations before the actual data frame can be transmitted. It is not until the transmitted frame is received and acknowledged by the destination that the token can be released by the sender to the next station.

This is another major drawback that has contributed to the virtual disappearance of ARCnet in the business networking world. Its 2.5 Mbps transmission rate, which is slow enough already, is further reduced by the large amount of signaling overhead required for normal communications (three bits of overhead per byte transmitted). Other exchanges of control information between sources and destinations are also required that contain no data and provide additional overhead.

There have also been known to be compatibility problems with some upper layer protocols, such as NetWare's IPX, due to the small frame size used by ARCnet. No more than 508 bytes of data can be included in an ARCnet frame, and the standard datagram size used by IPX is 576 bytes. An extra layer of translation, called the fragmentation layer, had to be devised to allow NetWare traffic to run on ARCnet. This extra layer breaks IPX packets into two smaller packets that are capable of being sent within the ARCnet frame's data field, and then reassembles them at the destination, adding still another level of overhead, of course.

For a network that is used for absolutely nothing more than file and printer sharing, and a minimal amount of these, ARCnet may be marginally suitable, and would certainly be far less expensive than any of the other major network types. For use in any business that plans on being in operation more than two years down the road, however, ARCnet is a shortsighted solution that will probably disappear from use completely within a few years and is not recommended under any conditions.

FDDI

Since its introduction in 1986, the Fiber Distributed Data Interface (FDDI) has come to be the accepted standard for high-speed network backbones and connections to high performance workstations. Running at 100 Mbps, it remains to be seen how the new high-speed technologies will affect the use of FDDI for these purposes. Both Fast Ethernet and 100VG AnyLAN offer the same speed with considerably lower upgrade and installation costs, even providing fiber-optic standards for connections over long distances and between buildings. If, once ATM becomes standardized, it proves to be half as popular as it seems that it will be, them FDDI's days could well be numbered.

The FDDI standard was created by the ANSI X3T9.5 committee. The document describes a network laid out in a dual ring topology, using the same token passing media access control mechanism that token ring uses, except that early token release is always used, instead of being optional. The dual ring provides two independent loops with traffic traveling in opposite directions, in order to provide fault tolerance for the network (see fig. 7.12).

Under normal conditions, only one of the two rings actively carries traffic. When a break or other disturbance in the primary ring is detected, relay stations on either side of the break begin to redirect traffic onto the secondary ring. Stations connected to both rings have two transceivers and are designated as dual-attachment (DAS) or Class A stations; single-attachment (SAS) or Class B stations are connected to only one ring, have only one transceiver, and therefore cannot benefit from this fault tolerance. A FDDI ring can contain up to 1,000 stations, with a cable length of no more than 200 kilometers. The use of Class A stations, however, effectively halves these limitations, as they count for two connections each and the dual rings double the length of the cable.

Figure 7.12 This is a basic FDDI double-ring network.

No more than two kilometers of cable can be laid without an intervening station or repeater. Obviously, these so-called limitations provide for a larger and longer network than any of the other protocols considered thus far.

The rings by which a FDDI network is organized may be actual ones, in that the stations are wired directly to one another or a concentrator may be used, as in a token-ring network, to provide a virtual ring to what is physically a star topology physical installation. A concentrator provides an easier mechanism for automatically removing a malfunctioning station from the ring but also provides a single point of failure for the entire network. The cable called for by the standard is graded index multimode fiber with a core diameter of 62.5 micrometers. Other types of single mode and multimode cable have been used successfully, however, as well as standard Type 1 STP and Category 5 UTP, although these are limited to a distance of 100 meters or less between connections.

The Physical Layer

In a FDDI network, the physical layer is divided into two sublayers, the physical medium dependent layer (PMD) and the physical layer (PHY). The PMD defines the optical characteristics of the physical layer, including photodetectors and optical power sources, as well as the transceivers, medium interface connectors (MICs) and cabling, as with other network types. The power source must be able to send a signal of 25 microwatts, and the detector must be able to read a signal of as little as two microwatts. The MIC, or FDDI connector, was designed by ANSI especially for this standard, and it has come to be used for other fiber-optic media standards as well. It is designed to provide the best possible connection to avoid signal loss and is keyed to prevent incorrect component combinations from being connected together. Other, less expensive, connector types have also been used for some FDDI networks, although their use has not been standardized. Be sure to check on the type of connectors used by all FDDI hardware that you intend to purchase, as interoperability may be a problem with anything other than official FDDI connectors.

The PHY layer functions as the medium-independent intermediary between the PMD layer and the MAC layer above. As the first electronic layer, it is implemented (along with the MAC layer) by the chipset in the FDDI network adapter and is responsible for the encoding and decoding of data into the light pulses transmitted over the medium. The signaling scheme used by FDDI networks is quite different and more efficient than the Manchester and Differential Manchester schemes used by Ethernet and token ring. Called NRZI 4B/5B encoding, this method provides a signaling efficiency rate of 80%, as opposed to the 50% rate of the other network types. This means that an Ethernet network pushed from 10 to 100 Mbps would have to utilize a 200 MHz signal, while only 125 MHz is needed by FDDI, to provide the same throughput.

FDDI-I and FDDI-II

FDDI also supports a more flexible system of assigning bandwidth according to priorities than token ring does. Available bandwidth can be split into synchronous and asynchronous traffic. Synchronous bandwidth is a section of the 100 Mbps that is designated for use by traffic that requires a continuous data stream, such as real-time voice or video. The remaining bandwidth is devoted to asynchronous traffic, which can be dynamically assigned, according to an eight level system of priorities administered by the station management (SMT) protocol that is part of the FDDI specification.

The original FDDI standard supported asynchronous communications and, while it did have a synchronous mode definition, it did not provide the degree of flow control that was needed for applications such as real-time video. Thus, the FDDI-II standard was created to define what is officially known as hybrid ring control (HRC) FDDI. The basic difference in the standards was the addition of a mechanism, called a hybrid multiplexer (HMUX) that allowed both packet-switched (from the original MAC layer) and circuit-switched data to be processed by the same PHY layer. The circuit-switched data, which can be defined as a real-time data stream such as voice or video is provided by an isochronous media access control (IMAC) mechanism called a circuit switching multiplexer (CMUX).

It is essentially the IMAC and the HMUX that make up the hybrid ring control element of the FDDI-II standard. Other changes made to the document at this time included the addition of alternative fiber media types, including single mode fiber-optic cabling. The hybrid mode capabilities of a FDDI-II are optional. The network can be run in basic mode that differs little from the original standard.

The FDDI Frame Type

FDDI utilizes a token passing MAC mechanism that is very similar to that used by standard Token Ring networks. Two basic frame types, a token frame and a data/command frame, are defined, with the fields and their functions basically similar to those defined in the 802.5 standard. FDDI even has a Station Management (SMT) protocol that is very similar in function to Token Ring's NMT, providing ring management and frame control to the network.

The Downside to FDDI

Fiber-optic cable is very difficult and expensive to install, and while users may be tempted to try to perform a coaxial or twisted pair physical layer installation themselves, they should not even consider doing fiber without expert help. These factors are major contributors to the limited but stable market that FDDI seems to have established for itself. It has found its niche in the networking world and it fulfills it admirably, but the fast-rising young newcomers, like Fast Ethernet and 100 VG AnyLAN are a distinct threat to its continued use. When the same speed and segment lengths can be realized with less expense, retraining, and maintenance costs, no persuasive reason can be found for installing new FDDI backbones. Unless these newer network types fail utterly, and this is doubtful, the FDDI standard may lapse from general use entirely before the end of the decade.

Cable Installation

Installing network cabling of any type is not something that can be properly learned from a book. While it may be relatively easy and inexpensive to connect a handful of PCs into a workgroup network using prepackaged cables, creating the physical fabric of a large network with results that are both functional and businesslike in appearance is a task better left to professionals. Unfortunately, it can sometimes be difficult to discern the professionals from the hacks with no training that simply hang out a consultant's shingle and purport to be networking experts.

The physical layer is often treated separately from other types of network training. CNE certification may include some basic training concerning the different types of cabling and the guidelines for the various network types, but it does not in any way cover such tasks as the crimping of connectors onto bulk cable, which are the most crucial parts of a physical layer installation. For 10BaseT installations, companies that install telephone systems certainly have the knowledge to properly pull and connect the cabling, but you should be certain that they are familiar with the special requirements for a data grade installation.

It should be quite simple to find a contractor who is capable of installing the cable properly. The problem usually lies in finding one who will do it for a good price. In any case, you should require a contractor to furnish a complete diagram of the proposed cable layout, including the locations of all connectors involved, making sure that the distances are within the specifications for the network type. Depending on your estimate of the contractor's expertise, you may or may not have to inspect their work closely to be sure that it's done properly. Since most cabling jobs will be hidden within the walls and ceilings of the site, the time to find problems is while the work is being done and not afterwards.

Summary

The physical and data link layers are the fundamental building blocks of a modern LAN. None of the higher level functions will be able to proceed normally if the foundation that they are built on is unstable. An informed and intelligent decision as to the proper network type to use for the needs of a particular organization can set the standard for the way that the network is to be built and they way that it will be run. Technology and purchasing decisions made well at the outset can ensure that a network installed today can later be adapted to whatever needs may arise.

In coordination with the material covered in chapters 5, "The Server Platform," and 6, "The Workstation Platform," nearly all of the essential hardware needed to construct a basic LAN is discussed. Other sections of this book will cover the many different products and services that can be used on the network, but all of them are dependent on this basic infrastructure. If the foundation is not solid, then the tower cannot stand for long; time and effort expended on network fundamentals will never be wasted.