Chapter 2: Designing and Planning Your NT Network Hardware

Abstract

This chapter shows you how to put together a LAN based on Windows NT Server. You will learn the ins and outs of cabling, network technologies, and network adapters.

Chapter 1 gave you an overview of network hardware and software—the various nuts and bolts that interconnect computers. This chapter tells you how to plunge your hands into that pile of bolts and put them together into a state-of-the-art LAN based on Windows NT Server 4.0.

If you’re going to use Windows NT Server as the basis of a new network, one of your first decisions will be what network hardware technology to employ. If you’ve already installed your network hardware or if someone else in your company is slated to do it, you can skip ahead to the section on setting up your network adapter and configuring Windows NT Server to work with it. You’ll need to understand your network adapter configuration before you can proceed with NT installation in Chapter 3.

Before you delve into this chapter, make sure that you have an understanding of network concepts and terminology. I assume that you’re already familiar with this material. You can use the glossary in Appendix B to fill in any fuzzy areas.

PLANNING THE WORK

There are several considerations to mull over when deciding how much of your network you’re going to install. Small networks are fairly easy to install, though putting cable through walls is hard to do. Stringing cables between offices through ceiling tiles is a risky business, and you need to comply with local fire codes. Making your own cables requires expensive tools, skill, patience, and luck. I recommend purchasing ready-made cable or having a professional create custom cable for you.

If you’re going to do it yourself, make sure to know all of the rules: building wiring codes, fire codes, maximum cable lengths, safety issues, cable tolerances, and so on. High-performance cable should always be installed by cable professionals. Make sure that what they do will be easy to maintain. Don’t scrimp on this.

If you’re planning a network that includes more than 40 computers, it’s wise to divide it into several segments connected by bridges or routers, discussed later in this chapter. Although you can learn the principles from books on enterprise networking, it’s often best to have your large network designed by a professional who’s made his or her mistakes on earlier clients.

If you do have a network cable professional do the job, watch, listen, learn, ask lots of questions, and take notes. In the future, you may have to crawl around and fix something quickly to support your company’s payroll check run. You’ll certainly also need to be able to handle small modifications to your network design, as well as deal intelligently with contractors to do major operations.

NETWORK TECHNOLOGIES

In the past few years, two network technologies—Ethernet and Token Ring—have emerged as the most popular. Older systems such as ARCnet and StarLAN have lost favor due to their limited data-carrying capacity. Of course, new technologies are always emerging, with some looking more promising than others as we move into the next century of enterprise computing.

In reality, Ethernet and Token Ring have a lot in common. Both are widely supported by hardware and software vendors and have been set up as international standards by the 802 committee of the IEEE (Institute of Electrical and Electronic Engineers). Both share the bandwidth of the network cabling by allowing only one node to transmit at a time. Either technology can be used to build a versatile and reliable network based on Windows NT Server.

New multimedia applications such as video conferencing have begun to push Ethernet’s 10Mbps capacity and Token Ring’s 16Mbps capacity past their limits. In such cases, it’s often the network cable component that restricts the speed of data throughput. Several new network technologies, including FDDI and Fast Ethernet schemes increase performance to 100Mbps. Ethernet switching technology and ATM give each pair of communicating nodes its own end-to-end connection to eliminate the data-flow restrictions caused by sharing the cable. Wireless network adapters (utilizing either radio or infrared communications links) have been entering the market. They can be useful for low-traffic situations where cabling is either impossible or impractical.

When you’re choosing a LAN technology, as in any other endeavor, you need to balance cost, reliability, and performance. LAN technologies differ on the type of cable they use, how fast they run, how easily they allow you to troubleshoot the network, and how well they connect to larger computers.

In discussions of network hardware technologies, the access method is often a matter of heated debate. Actually, this is one of the least important technological differences, at least in comparing today’s mainstream approaches. Most systems limit use of the network to one computer at a time. ATM is the exception to this rule.

In the following sections, I introduce the network cabling alternatives available to you and then discuss and compare the major network technologies from which you can choose. Once you’ve gone this far, I’ll discuss selecting and configuring your network adapter hardware.

NETWORK MEDIA TYPES

No, this section isn’t about television executives. Over the past 10 years, manufacturers have developed networks that can transmit data on just about any type of wiring, but generally they utilize one of five types of network media:

Coaxial cable (coax)
Shielded twisted-pair cable (STP)
Unshielded twisted-pair cable (UTP)
Fiber-optic cable (fiber)
Wireless transmission

I present and compare each of these types of network media in the next sections.

Coaxial Cable

Before 1990, nearly all LANs used coaxial cable, or coax, that is similar to television cable. Coax consists of two conductors centered on a common axis (hence, the name coaxial). The center conductor is a thick copper wire. An insulation layer separates the two conductors and keeps the copper wire centered for peak performance. The outer shield conductor, typically made of foil or copper braid, prevents EMI from leaking into or radiating from the cable. The outer jacket is typically plastic or Teflon. Figure 2-1 illustrates the coax structure.

Figure 2-1: Coaxial cable (coax) uses two conductors to transmit data electrically.

There are several flavors of coax. One of the most common types in very small networks is called thin Ethernet, or thinnet. Many network starter kit products use this type of thin, black cable, sometimes referred to as RG-58. Thicker cables (known as thick Ethernet, thicknet, or “frozen yellow garden hose”) are generally very expensive and inflexible but allow longer cable lengths and better performance. The industry trend has steadily been toward thinner coax.

Two kinds of coax are available for different types of office installations. PVC coax is flexible and can be routed in exposed office areas. PVC refers to the material (poly-vinyl chloride) used to construct the cable’s insulation. Plenum coax can be routed above false ceilings (called the plenum area), because it uses insulation that is fire-resistant and produces less poisonous fumes if it does burn. Plenum cable is less flexible and more expensive than PVC.

Coax has been around for more than twenty years, so it’s well-understood, reliable, and readily available. Installation is simple and cost-effective. The cable itself is inexpensive in its thinner forms, but thick versions can be quite expensive. Since it’s used to implement a bus topology, any cable break along the way can bring down the network. If mice (the rodents, not the pointing devices) get hungry enough, they’ll chew through coax cable. Although coax is highly resistant to EMI, it’s not invulnerable, so avoid it in electrically noisy environments. Few modern network adapters support coaxial cable exclusively, and increasing numbers of adapters that don’t support it at all are appearing on the market. Avoid coax-only support when selecting your network adapter.

Shielded Twisted-Pair Cable

Historically, the next type of cabling used after coax for network transmission was shielded twisted-pair, or STP. (No, this isn’t a brand of motor oil.) STP cable contains a pair of insulated copper wires that are twisted around each other. The twists cause the wires to alternate positions, which has the effect of reducing absorption or radiation of EMI. In general, the more twists per unit of length (for example, 10 twists per foot), the better the cable’s data transmission capabilities. A conducting shield, similar to that in coax cable, surrounds the twisted wires to further inhibit incoming or outgoing EMI. Most STP cables actually contain multiple twisted-pairs. In some cables, a common shield surrounds all pairs. In others, each pair has its own shield, to prevent interference between adjacent pairs. The outer jacket is typically made of plastic. Figure 2-2 illustrates the STP cable structure.

Figure 2-2: STP cable surrounds pairs of twisted wires with an outer shield conductor.

IBM Type 1 cable, used in Token Ring networks, is the only shielded twisted-pair cable in common use today. This flavor of STP contains two twisted-pairs surrounded by a common shield. A few versions of Ethernet adapters out there can use STP.

STP cable’s size, cost, and bandwidth, as well as its installation costs are about on par with thin coax cable. EMI resistance and durability of STP are also close to that of coax. Because of its EMI resistance, STP is a good choice for Token Ring installations in electrically noisy environments such as factories and warehouses.

Like coax, STP has grown steadily less popular over time, especially since IBM endorsed the use of unshielded twisted-pair (UTP) cable for Token Ring installations. STP is more expensive and difficult to install than UTP cable, discussed next.

Unshielded Twisted-Pair Cable

Unshielded twisted-pair cable, or UTP, is very similar to the cable used by telephone companies, although of much higher quality. Like STP, UTP uses a pair of insulated wires twisted around each other to reduce incoming and outgoing EMI. Unlike STP, there’s no shield conductor surrounding the wire pair. Lack of the shield reduces both the cost and physical bulk of the UTP cable but increases its susceptibility to EMI and its potential to radiate electrical noise. The bulk of the cable is reduced even further because UTP uses thinner wires than either coax or STP cables. Like STP, multiple pairs are often bundled together in a single cable. There are several types of connectors, but the most common is the RJ-45 connector, which looks similar to the modular plugs used in modern telephone connections. Figure 2-3 shows the UTP cable structure.

Figure 2-3: UTP cable consists of pairs of twisted wires with no outer shielding conductor.

Cross Reference: Although coax is used to wire physical bus topologies, UTP is typically used to wire physical star topologies. Thus, use of UTP requires wiring computers to hubs.

UTP cable varies greatly in quality and, therefore, in how well it can carry high-speed data. To avoid network nightmares, choose a more expensive, high-quality brand in the first place, and stay away from the temptation to mix your good cable with segments of unknown quality. In general, the rule is to never mix cable of different types or quality—even a single inferior segment can cause signal noise and data corruption.

Caution: Don’t even think about using flat (untwisted) telephone cable, commonly know as “silver satin,” to connect a computer network. Using just one six-foot length of this type of cable can prevent network data from passing through your entire network.

UTP is the least expensive of all types of cable, is relatively easy to install, is very flexible, and takes up very little space. Installation can be quite inexpensive, since many buildings in recent years have been prewired with high-quality UTP in anticipation of computer network and digital telephony needs. Because it’s used to implement a star physical topology, a break in one cable won’t bring down the network, in contrast to coax. On the down side, UTP is more easily damaged and is more susceptible to EMI than any other cable type. Moreover, it requires the additional cost of hubs to connect nodes to the network.

To help you determine if a certain type of UTP cable can effectively carry your network data, the EIA/TIA (Electronics Industries Association/Telecommunications Industry Association) has promulgated a UTP grading system. It categorizes UTP cable into five levels (sometimes called grades or categories).

Levels 1 and 2 should be used only for telephone voice and PBX applications, respectively. Most commercial telephone systems installed before 1990 use these lower-quality cables. Levels 3 and 4 can be used for either Token Ring or Ethernet LANs. Since 16Mbps Token Ring pushes the envelope of Level 3 performance, you should use at least Level 4 cable for 16Mbps Token Ring. Level 5 is the best grade of UTP cable. It can be used for Ethernet and 16Mbps Token Ring, as well as for the new Fast Ethernet running at a maximum rate of 100Mbps. Table 2-1 summarizes the maximum data rates of each of the UTP levels.

TABLE 2-1 UTP CABLE QUALITY LEVELS
*UTP Level*	*Maximum Data Rate*	*Description*
1	none	Voice grade cable that’s not usable for LAN data at all.
2	1Mbps	Usable for very low-speed transmissions (voice and PBX), but not usable at all for LAN data.
3	16Mbps	Usable for Ethernet (10Mbps, or 10Base-T) or Token Ring (4Mbps) LAN data.
4	20Mbps	Usable for Ethernet (10Mpbs, or 10Base-T) or Token Ring (4Mbps or 16Mbps).
5	100Mbps	Usable for Ethernet or Token Ring, as well as Fast Ethernet at up to 100Mbps.

Tip: If you’re installing a new network in your enterprise, opt for nothing less than level 5 UTP and install cable that contains extra pairs of wires. If you do, you’ll pave the way for easily upgrading later to the full-duplex switching network technologies that are appearing on the horizon.

Because your network cable infrastructure should be designed to last ten years or more, you’ll be much better off choosing level 5 UTP for any new network installation. The modest additional cost for the higher-quality cable will save you considerable time and labor later when you can confidently upgrade your network without installing new cable. Installing anything less than level 5 UTP today is a poor investment in the long run.

Note: Some people think that level 5 UTP cable can’t be used for 10Mbps Ethernet (called 10Base-T), which requires level 3 UTP. Actually, level 3 is only the minimum grade required—level 5 exceeds that minimum and works superbly with 10Base-T Ethernet systems. Keep in mind that level 5 can do anything levels 3 and 4 can do, and more.

Fiber-Optic Cable

Fiber-optic cable, or fiber, is the new kid on the network cable block, offering the best performance at the highest cost. It consists of a glass or plastic fiber core that transmits data using pulses of light. This nonelectrical approach makes fiber completely immune to EMI and corrosion. Surrounding the core, a glass coating (called cladding), is specifically designed to reflect signals back into the core to minimize signal loss. The outer jacket, typically made of Kevlar, can house one or more fibers. By bundling multiple fibers, you can fit many of them through the same space as a single copper cable. Many of today’s fiber-optic cables contain two fibers, one for transmitting and one for receiving. Fiber cable technology uses a laser diode as its light source, which accounts for the bulk of the cost of this network media. Figure 2-4 shows the fiber cable structure.

Figure 2-4: Fiber-optic cable uses light pulses to transmit data through a glass or plastic core.

Glass fiber cables can span miles without the need for repeaters, whereas plastic fiber is typically limited to the length of a football field. Data rates can potentially exceed one terabit per second, about 10,000 times faster than UTP cable. Installing fiber used to be very costly but can now be reasonably accomplished with the investment of a few hundred dollars in tools and a little training. The reduction in installation cost is primarily due to the recent development of simpler connectors. The cost of the cable itself is really no more expensive than thick coax. However, the real sticker shock hits for the transmission devices on each end. It can cost up to $1,000 to connect a node to a fiber network, although prices have been inching lower over time.

Because of fiber’s high cost, organizations tend to use a mixture of fiber and copper cabling. For example, at Microsoft, we used a fiber-optic cable backbone to connect the several buildings on the Microsoft corporate campus. Within each building, we used less expensive, level 5, UTP copper cabling.

There are several advantages to using fiber in portions of a network. If you have or expect to have high traffic in a segment of your network, fiber can be a good investment that you won’t outgrow for quite some time. Since fiber is completely immune to EMI from the environment, it’s ideal for electrically noisy environments. Conversely, fiber radiates no electrical noise, so it’s completely immune to electronic eavesdropping. Finally, fiber makes for good network connections between buildings. Since it isn’t electrical, it doesn’t attract lightning or cause data-corrupting ground loops, two problems often seen with copper cabling between buildings.

Cabling—The Final Analysis

Table 2-2 summarizes the comparison of various characteristics of the cable types discussed so far. This table should help you select the right cable for your organization.

TABLE 2-2 COMPARISON OF NETWORK CABLE CHARACTERISTICS
*Characteristic*	*Coax*	*STP*	*UTP*	*Fiber*
Data transmission bandwidth	High	Medium	Low to Medium	Highest
Sensitivity to EMI	Low	Low	High	None
Cable cost	Medium to High	Medium	Low	High
Installation cost	Medium	Medium	Low to Medium	Medium to High
Maintenance difficulty	Medium	Low	Low	High
Flexibility	Low to Medium	Medium	High	Low (no sharp turns)
Distance limitations	Medium to Long	Medium	Short	Very Long
Impact of damaged cable	High	Low	Low	Depends on topology
Security issues (eavesdropping)	Yes	Yes	Yes	No
Security issues (eavesdropping)	Yes	Yes	Yes	No
Cost of support electronics	Low	Medium	Medium	High
Adapter choices	Broad but shrinking	Limited	Broad	Limited
Physical cable bulk	Medium to Thick	Medium	Thin	Very Thin
Attenuation of signal	Medium	Medium	High	Low
Functionality for network backbones	Thicknet only	No	No	Ideal
Component standardization	High	Medium	High	Low

THE WIRELESS ALTERNATIVE

Wireless technology currently comes in two flavors—namely, optical and radio frequency. Optical networks, which use the same type of infrared signals found in your television remote control, can be implemented as node-to-node transmissions or broadcast transmissions that fill a room. The former requires that the communicating computers be within sight of each other and that the placement of transmitters and receivers avoids interruption by people walking by. Broadcast transmissions typically go to one or more centralized hub-like devices and are rebroadcast to other nodes in the same room. This approach reduces the need for an unbroken line of sight between nodes but also reduces reliability, since signals are weaker and more susceptible to changes in ambient lighting. Optical networks can be used outside to cross streets between buildings and so on. However, since you need to maintain a clean line of sight between transmitter and receiver, performance and reliability suffer from such simple things as fog, rain, hail, and pigeons.

The radio frequency approach to wireless networks has the advantages of being able to penetrate walls and not requiring nodes to be within sight of each other. Two major network technologies are available in this realm. The first is called spread spectrum, which was developed by the U.S. military for transmitting secure information. This approach uses unlicensed radio frequencies (including those used in alarm systems, garage door openers, and radio-controlled model trains, planes, and automobiles). It spreads the signals across a spectrum of frequencies, picking whichever ones have the least interference and sometimes transmitting different parts of a message over different frequencies to add more security.

The second approach to radio frequency networks is based on cellular phone technology. This method uses a licensed radio frequency of 18GHz and is offered primarily by Motorola, who holds the frequency licenses in most major cities. Nodes in this type of network are organized into groups, called cells, covering a limited area. Typically, several network nodes are wired to a special user module via cables. The user modules communicate by radio frequency to a hub that’s within the same room. Several hubs are often connected to each other via cables.

This approach sort of takes the “wireless” out of wireless networks. It’s wireless only between the user modules (to which the nodes are wired) and the hubs (which are wired together). Since 18GHz can’t penetrate walls, you’re faced with wires between rooms (that is, between hubs) with this technology as well.

If you have special node mobility requirements, wireless can be a viable solution. A good example of a wireless network is the Avis rental car return area at many airports. Their system allows you to check in your returned car right there in the parking lot. Employees walk around with hand-held PCs that communicate over a wireless network. Imagine if they each needed a cable strung back to the office.

Why Wireless?

Wireless is a very flexible solution for small physical areas with limited amounts of data communication. Because of the high cost and relatively low speeds associated with this technology, you should consider a wireless network only if there’s some concrete reason why cabling is impractical or impossible. (In some environments, concrete is the reason that you can’t string cables.) Otherwise, stick with cabling for now.

If you need to pick a wireless approach, I currently recommend using the spread spectrum technology. There’s reasonable hardware and software support for this approach right now, you don’t have to worry about line-of-sight issues (as you do with optical), it transmits through walls (unlike optical and 18GHz), and it actually is a wireless network (unlike the 18GHz approach).

Wireless technologies continue to improve in speed, reliability, and cost effectiveness. Although new standards are emerging in the wireless marketplace, some vendors are implementing higher-speed technologies at the expense of deviating from the new standards. Keep a lookout for breakthroughs in price and performance in this area.

In some buildings, you may be prohibited from installing cabling because of the age, historical, or religious significance of the structure. The need to pass your LAN across a public street may also bring wireless into the picture as a cost-effective alternative to leasing an expensive line from a telephone provider. Another factor to consider in deciding whether or not to go wireless is the real cost of moving computers in your organization. Wireless technology vendors tend to quote high costs for this activity, but you need to develop your own unbiased estimates for this cost.

IMPLEMENTING NETWORK TECHNOLOGIES

In this section, I discuss the most prevalent cabling options available for implementing specific network technologies. Here, you’ll learn the details of setting up Ethernet, Token Ring, Fast Ethernet, FDDI, and ATM networks.

Ethernet Cabling

The IEEE 802.3 standard describes most modern Ethernet implementations. Ethernet II, developed by DEC, Intel, and Xerox (sometimes referred to as DIX), is older and differs from the 802.3 standard. Ethernet II is still in common use today in TCP/IP and DECnet environments. I won’t differentiate between Ethernet and Ethernet II in this book, as the differences are minor. Regardless of the media or implementation, Ethernet’s maximum bandwidth is 10Mbps. It’s less than this with certain types of network media.

Of all the network technologies discussed in this chapter, Ethernet is the least expensive and most prevalent today. Many manufacturers build Ethernet ports into their computers, network adapters are available at commodity prices, and experienced technicians who understand Ethernet cabling and installation are easily available.

Thick Ethernet Thick Ethernet is also known as thicknet, 10Base5, and my favorite—chunkynet. Thicknet was the original Ethernet cabling approach, consisting of half-inch diameter coaxial cable. To connect to the network bus, you need a transceiver (transmitter-receiver) tapped into the cable. The most common type of transceiver in the 10Base5 world is something out of a Brahm Stoker novel, called a vampire tap. It clamps onto the cable, sinking metal fangs through the insulation to make contact with the conductors inside. Less popular types of access units exist, but they’re equally violent and less convenient—they require chopping the cable, forcing a special (N-series) connector onto the two ends, and connecting those ends to a special T-connector on the access unit. An AUI (attachment unit interface) or DIX (Digital-Intel-Xerox) connector is used to attach the node to the transceiver.

Thicknet is the most expensive and least convenient of the Ethernet cabling methods, but it does have a place in environments that require long segment lengths (for example, network backbone segments).

Thin Ethernet Thin Ethernet, also known as thinnet, cheapernet, and 10Base2, was developed to reduce the cost of cabling and eliminate the need for separate transceivers. The transceiver function was moved to the network adapter. To connect to the network bus, you simply use a T-connector to tap into the main network cable. If you use preconfigured cable with preinstalled connectors, no cable cutting or vampire bites are required.

Caution: Always connect the T-connector directly to the node’s network adapter. Although it may sometimes seem convenient to drop a short cable between the T-connector and the computer node, don’t do it. This procedure may provide convenient network access to the node that you’re cabling, but it can affect the main network cable enough to prevent other nodes from talking on the network.

Thinnet is much less expensive than thicknet, both in terms of cabling cost and support electronics. Its maximum segment lengths and the limit on the number of devices per segment is smaller. Since each node connection depends on three BNC connectors at the network adapter, any break in these connections can result in bringing the entire network segment down. Moreover, adding a node to a segment requires bringing down that segment during the connection process.

Violence on the Network

There’s some disagreement in the industry over what the BNC acronym really means. Scientist types say is stands for biconic (no, not bionic), based on its shape. Others say it stands for bayonet connector, based on the bayonet-like twisting action required to make the connection. The latter evokes a pretty gruesome image. Still other folks say BNC stands for bayonet nut connector. My advice? Never connect a nut with a sharp instrument like a bayonet. You never know what it might do.

Thinnet remains the low-cost leader, even against the more popular UTP alternative, since 10Base2 doesn’t require a hub. If per-node cost control is crucial and ease of reconfiguring isn’t important, I recommend 10Base2 only in very small networks (three to five nodes, at most) that coexist in one or two rooms. If you plan to grow the network beyond this (and most networks do grow), invest up front in UTP, discussed in the next section. Table 2-3 provides some details of 10Base2 characteristics and compares thinnet with other common Ethernet cabling alternatives.

TABLE 2-3 COMPARISON OF COMMON ETHERNET CABLING OPTIONS
*Characteristic*	*10Base5*	*10Base2*	*10Base-T*
Cable type	Thick Coax	Thin Coax	Level 3 UTP or higher
Cable diameter	0.5 in.	0.2 in.	0.2 in.
Minimum cable bend radius (flexibility)	10 in.	1 in.	0.5 in.
Nicknames	Thick Ethernet, thicknet	Thin Ethernet, thinnet, cheapernet	Twisted-Pair Ethernet, UTP
Relative cable cost	High	Medium	Low
Maximum cable length from transceiver to node	164 ft (50 m)	N/A	N/A
Maximum cable length from hub to node	N/A	N/A	330 ft (100 m)
Maximum segment length	1640 ft (500 m)	600 ft (185 m)	N/A
Maximum number of transceivers on a segment	100	30	N/A
Termination of segment	Each end of segment	Each end	N/A
Maximum number of repeaters between most distant nodes	4	2	N/A
Maximum network length	8200 ft (2500 m)	1800 ft (555 m)	N/A
Relative cost of support electronics	High	Medium	High
Optimal distance between node connections on the bus cable	Multiples of 8.2 ft (2.5 m)	Multiples of 1.64 ft (0.5 m)	N/A
Relative ease of reconfiguring	Low	Medium	High
Connector type	AUI or DIX	BNC	RJ-45

Twisted-Pair Ethernet Unlike 10Base2 or 10Base5, twisted-pair Ethernet, or 10Base-T, uses a physical star topology to connect nodes to hubs using level 3 UTP (or higher UTP grade) cable. Ethernet 10Base-T cables contain two twisted-pairs, one for transmitting and one for receiving. This allows the network to remain as a logical bus topology, just like 10Base2 and 10Base5. The cables often contain four pairs of wires, to allow for broken wires and future network expansion.

Hubs that handle between 4 and 128 nodes are available. They range in price and functionality from passive hubs to remotely manageable active hubs. As discussed earlier, the UTP hub approach is an extremely flexible alternative. You can add or remove nodes from the network without having to bring the network down. Reconfiguration and management of the physical star topology is infinitely simpler than the coax alternatives. Although 10Base-T cable is less expensive than coax, the cost of hubs can add from $20 to over $100 per node to the overall network cost. So, cost can’t be the primary factor in selecting 10Base-T. If your network will grow and will need to be reconfigured and managed, opt for 10Base-T.

Who’s on Base?

Where did the 10Base terms come from? These military-developed terms encode information about the cable medium. For example, 10Base5 refers to cable that uses baseband transmission to send data at 10Mbps with a maximum cable segment length of 500 meters. Can you guess what 10Base2 is? Did you say it’s the same as 10Base5, except the maximum segment length is 200 meters? Well, you’re close. The segment limit is 185 meters, but the standards folks rounded up. (10Base1.85 would have been too confusing, I suspect.) Of course, 10Base-T doesn’t tell you anything about the maximum segment length. So much for encoded information in these names.

Token Ring

Originally developed by IBM, Token Ring is now defined by the IEEE 802.5 standard. Unlike 802.3, the 802.5 standard doesn’t address cabling specifications. Thus, many in the industry have adopted IBM’s cabling designs. Shielded or unshielded twisted-pair cabling is typically used, although some fiber and coax varieties exist. Token Ring’s logical topology, a ring, is always implemented as a physical star topology. Nodes are connected to MAUs, and MAUs are connected to form an unbroken logical ring. The special connectors used to hook nodes to MAUs cause the ring to heal when the connector is removed, thus allowing the insertion and removal of nodes without disturbing the ring.

In general, you can hook up to 260 nodes and 33 MAUs to a single Token Ring. You can use repeaters to extend the physical size of a ring, although those repeaters also count as nodes. Token Ring capacity planning is a pretty involved process. IBM devotes a significant portion of their planning guides to calculating ring capacities and dimensions. Two speeds are available—4Mbps and 16Mbps. Most MAUs and many network adapters today support both rates, so there’s really no cost advantage to choosing the 4Mbps technology. Both are considered mature technologies.

Caution: Don’t try to run a mixture of 4Mbps and 16Mbps nodes on the same Token Ring network. The network won’t operate. Always choose and use a single speed for the entire ring. Mismatched speeds are the most common configuration problem on NT Token Ring networks today, especially with the availability of network adapters that can be configured through software to run at either rate.

Token Ring hasn’t gained as much acceptance as Ethernet because it costs significantly more per node to implement, is still considered proprietary IBM technology by many, and is more complex, requiring more technical knowledge to maintain. However, if you’re tying your network to IBM mainframe equipment, or you need a very deterministic approach to providing network access to all of your nodes, Token Ring is well worth considering.

FDDI

Originally designed for WANs, FDDI is now used in many LANs that demand high speed. FDDI, which has a 100Mbps bandwidth, was initially intended for fiber-optic cable, but some copper wire implementations have appeared as well. FDDI functions very much like Token Ring. It uses a logical ring topology and token passing for access control.

FDDI is often configured to use two rings, each shooting data in opposite directions. Nodes attached to one of the rings are called single-attached; nodes attached to both are dual-attached. The dual rings can have a diameter up to 100 kilometers and are able to heal themselves in certain situations where a link fails in one of the rings. This is illustrated in Figure 2-5.

Figure 2-5: FDDI networks can heal themselves if a link breaks between dual-attached notes.

High Fiber, Low Consistency

Pick up any four network hardware books, and you’ll see four different definitions of FDDI. There’s Fiber Distributed Data Interface, Fiber Data Distributed Interface, Fiber Digital Device Interface, and Fiber Data Distribution Interface. Note the common theme—they’re all fiber interfaces. If you want a lower-fiber diet, you’ll find implementations of FDDI over copper cabling (STP or level 5 UTP), sometimes called CDDI. CDDI has similar variations in definition, but you get the idea.

FDDI technology is ideal for high-speed network backbones and for connecting servers that stay up at all times. It isn’t happy in a desktop workstation environment where devices are turned on and off frequently. FDDI hasn’t gained as much support as Fast Ethernet (discussed next) because it costs significantly more per node to implement. Costs are drifting down, however. FDDI’s fault tolerance when implemented as a dual ring, along with its Token Ring-like deterministic access method, make it worth the additional cost in some situations.

Fast Ethernet

At a maximum of 10Mbps, Ethernet is running out of steam when faced with pushing huge amounts of video, voice, and imaging data across its wires. A number of attempts to standardize a 100Mbps Ethernet, called Fast Ethernet, are ongoing. Most of the proposals on the table reduce the maximum network size, raise the minimum cable grade to level 5, increase the number of twisted-pairs required in the cable, or some combination of these. If you have an existing network infrastructure, some of these requirements may impose high costs in replacing and upgrading cabling.

Several Fast Ethernet products have emerged, including network adapters, hubs, and cabling. In general, they cost significantly less than FDDI. However, since the standards haven’t been settled, you run the risk of investing in technology that might become obsolete once the standards are in place.

Switching Hubs

Several Ethernet hub products have emerged that provide a full 10Mbps bandwidth at each port, since they effectively make a temporary direct connection between two ports. By using two switching hubs together, you can increase the bandwidth between them by the number of ports in the hub. Some organizations use this approach to create a higher-bandwidth backbone between departments or buildings, for example. These hubs can extend the life of an existing Ethernet by increasing its bandwidth in critical paths within your network.

ATM

Put your credit card away. ATM is no cash machine. In fact, it will take your money. But you get what you pay for in high data bandwidth. Although very expensive now, many experts believe that ATM, or Asynchronous Transfer Mode, is the network technology of the twenty-first century. It’s designed to be very scalable and to handle continuous large volumes of traffic (combinations of data, voice, and video) for extended periods.

As a network administrator, you should educate yourself about ATM and whether it’s an alternative to consider. Since the ATM landscape is changing at a frenetic pace, I’ll present the fundamentals here. By the time you read this, there will undoubtedly be some new developments in the ATM marketplace.

ATM is a switching technology, rather than a shared media technology such as Ethernet or Token Ring. In fact, ATM is really closer to a telephone network than a computer network in its design. Any node can directly connect to any other node, without sharing communication bandwidth with other nodes—similar to telephone calls with no party lines. It can transmit data between nodes from 1Mbps to more than 1Gbps and at various speeds in between. Connections that need high speed can get it. Connections that don’t need high speed can transmit more slowly. Part of the connection negotiation process in ATM involves establishing end-to-end quality of service parameters—in other words, transmission speed and predictability of delays. This is quite different from other network technologies, which guarantee delivery of data, but never how fast or how predictably over time.

Transmitted data on ATM networks is performed using 53-byte packets called cells. The details of the physical layer aren’t specified by ATM, so data can be transmitted on various types of network media including twisted-pair, fiber, and coax. Current ATM products entering the mainstream don’t yet hit the 1Gbps mark. At the time of this writing, switches are typically offered for 25Mbps at the desktop, 155Mbps for workgroups, and up to 622Mbps for ATM backbones.

ATM is still an emerging technology, with standards for various aspects of it being set every few months. Thus, it’s still plagued with incompatibilities among different vendor implementations. This problem is offset somewhat because many of the ATM products are completely software-upgradable to meet the new standards as they arrive. Thus, many network administrators are already becoming card-carrying ATM users. (I suspect that even after this technology dominates the network industry in the next century, folks will still be making double entendre comments about ATM cards.)

LENGTHENING YOUR STRIDE

As you’ve seen, each network technology is subject to physical limitations imposed by the hardware and the environment in which the network operates. As data travels through copper network cabling, the signal weakens because of resistance in the cable and can become distorted because of external interference from the environment. This fact of life is known as attenuation. It limits the amount of distance allowable between nodes on the network and varies with the type of cable used.

If you need to have longer distances between nodes on the network than allowed by the type of cable and network technology used, you need to employ a device called a repeater. A repeater does exactly what its name implies. On one end, it receives an incoming signal and retransmits (or repeats) it out the other end. In the process, it boosts and cleans up the signal, making it as fresh and clean as the original.

WANs typically use many repeaters along the way. If you need a down-to-earth example, consider a communication satellite orbiting the planet. Its primary function is to act as a repeater. The satellite receives a signal sent from one place on earth, amplifies and cleans it, and then retransmits it to another earth location.

Numerous network hub devices today include built-in repeaters, so connecting to these hubs has the effect of boosting and cleaning the signal. In most network installations, the use of active hubs containing repeaters completely eliminates the need to use separate repeaters in the network. On the other hand, coax LANs (that don’t use hubs) typically require repeaters to extend the length of the bus cable.

BUILDING BRIDGES

What if you have two networks that use different types of cabling? Build a bridge. What if you have a network that’s overloaded with data traffic? Build a bridge. What if you want a more reliable and more secure network? Say it with me now. Build a bridge.

A bridge is a sort of smart repeater. At its most basic level, it performs the familiar function of a repeater by receiving, amplifying, and forwarding data on the network. However, a bridge does much more than this. It can link two network segments that use different types of cabling. This can be especially useful if different departments in your organization created independent networks, and you’ve inherited the task of hooking them together.

In addition, you can use a bridge to improve network performance. Insert a bridge in your network to break it into multiple segments. The bridge will isolate the segments from each other, except for data that must travel from a node on one segment to a node on the other segment. Local network traffic that used to be seen by the entire network is now only seen by the local segment.

For example, say, you use a bridge to separate your network into two equal-sized segments. Assuming the network traffic was evenly spread over the entire network, adding a bridge reduces the traffic on both segments by an average of 50 percent.

By segmenting your network this way and using bridges to link the segments, you also gain an additional measure of reliability and security. If one network segment goes down because of a broken cable or other equipment failure, the other segment is unaffected and can continue to operate. For larger networks, this can be extremely important. On the security front, you can design your network so that sensitive traffic is isolated to one segment. This limits your exposure to malicious folk with sniffers (the network analyzer kind, not the nose kind). Even if performance isn’t a problem on the network, you may opt in some cases to segment your network, just to achieve the reliability and security benefits.

SELECTING YOUR NETWORK ADAPTER

Once you’ve selected the network technology and cabling (or lack of cabling) approach to take for implementing your network, the next step is to select network adapters for your servers and clients.

In the following sections, I cover the issues that you need to consider when selecting a network adapter for your Windows NT Server computers and other computers on your network.

Network Compatibility

Is the adapter compatible with both your network cable infrastructure and your chosen network technology? A few short years ago, thick Ethernet connectors started disappearing from many network adapters, and more recently thin coax connectors have become less prevalent. Be sure that you know what connectors are provided on the network adapter.

Is the adapter listed on the NT Hardware Compatibility List, or is it 100 percent compatible with another adapter that’s listed? Although you may save a few dollars by purchasing unknown clone adapters, it’s definitely worth the small premium to populate your servers with well-known, reliable brands (such as Intel, 3Com, or SMC for Ethernet; IBM, Proteon, SMC, or Madge for Token Ring). Remember, you’re buying reliability for your server, a computer that must perform consistently for multiple users.

If you’re using Token Ring, make sure that the adapter matches the speed of your ring. Better yet, use an adapter that will operate at either 4Mbps or 16Mbps. Most modern adapters support both.

NT Compatibility

Does the adapter have an NDIS 3.0 device driver available, either included in the NT installation, included in Microsoft’s driver library, or provided by the adapter vendor? Check with the network adapter manufacturer and the NT Hardware Compatibility List for this information. (Visit www.microsoft.com/BackOffice/ntserver/hcl on the Web for an up-to-date list.)

Tip: Windows NT Server provides a number of certified network adapter drivers on the NT CD-ROM. Find them in \DRVLIB\NETCARD\X86 (or MIPS, ALPHA, or PPC, depending on what platform you’re running). Under this directory, change to the subdirectory whose name matches your adapter. Many of these drivers include a readme file with specific installation instructions. If there’s no readme file, just copy all of the files in the directory to the root directory of a floppy disk.

If you’re planning to install multiple network adapters in your computer, check with the adapter manufacturer to assure that their NT driver supports multiple adapters. Most do, but occasionally they don’t. It’s best to check before purchasing, to avoid surprises.

Tip: Within the NT development group at Microsoft, the most widely used adapter has been the Intel EtherExpress 16. It’s the adapter that Microsoft shipped with Windows for Workgroups, so they had lots of these available. Although this adapter is not inherently better than others, it’s probably had more air time running NT than any other adapter on the market.

Server Compatibility

Does the adapter match your computer’s bus (ISA, EISA, MCA, PCI, or VLB)? If you have PCI slots in your server, definitely use a PCI network adapter. If you have EISA slots in your server computer, opt for a network adapter that’s EISA rather than ISA. You’ll typically get better performance for a small incremental cost difference. Stay away from VLB network adapters. In theory they’re fast, but they typically tap lots of CPU bandwidth and aren’t transportable to more standard, non-VLB computers. In MCA computers, IBM adapters tend to be more reliable than others, in my experience.

If you’re using a non-Intel platform, it’s especially important to check ahead of time for driver availability, since some adapter manufacturers have chosen not to port their drivers to MIPS, Alpha, or PowerPC.

Will the adapter work in your computer without creating interrupt and memory conflicts? Be armed with the details of your computer’s configuration before you purchase a network adapter. Know which interrupts and I/O addresses are available in your server. I’ve run into several situations where I can’t use a network adapter because it provides very limited interrupt choices that all conflict with existing adapters.

Flexibility

Does the adapter allow for multiple types of cable, in case your approach to network cabling changes in the future? For example, if you’re using coax today, but you’re considering switching to UTP, check to see if the adapter provides both UTP and coax connections. This type of adapter is often called a combo. A few years ago, combos provided thicknet and thinnet on the same adapter. Later, combos more typically offered thinnet and UTP. Now, combos with 10Mbps UTP and 100Mbps UTP are available, and thinnet BNC connectors are beginning to disappear.

Since interrupts and I/O address ranges are precious commodities in the PC architecture, look for network adapters that support as many settings as possible. This level of flexibility will make installation of the adapter proceed more smoothly. Choosing an 8-bit ISA adapter isn’t only a slow option for a server, but it’s very inflexible because it only supports a small number of interrupts. If you need to use an ISA network adapter, opt for a 16-bit adapter that supports a wider range of IRQ settings.

Is the adapter easy to configure? Opt for a software-configurable adapter, if possible. You’ll save time as well as wear and tear on your computer by not having to pop the adapter in and out to view or change jumper settings.

Networking—It’s in There

Several NT-compatible computers are now appearing with a network adapter built right into the computer’s motherboard. Intel, DEC, AMD, and National Semiconductor are among the suppliers of these built-in network chips. You can recognize them in the Windows NT Hardware Compatibility List of network adapters by looking in the Bus Type column for the word “embedded.” For example, a number of MIPS processor systems have the National Semiconductor SONIC network adapter sitting on the motherboard.

Building in network capabilities is cost-effective for both the computer manufacturer and you. The incremental price of a motherboard with an embedded network adapter is often much less than the cost of a separate adapter card. Moreover, if you need every precious expansion slot in the computer for other interface cards, these motherboards will save you a slot that you’d have burned for the network adapter card. As another plus, some of these chips may offer better performance than a comparable adapter card.

Like everything else, however, this approach has some disadvantages. For one thing, what you gain in cost savings, you lose in flexibility. Having a replaceable adapter card is preferable if you plan to upgrade your networking capabilities as prices fall, speed increases, and new networking technologies emerge. This is especially important for servers, which are constantly expected to offer high performance. Another issue to consider is network adapter failure. These devices do occasionally croak, and built-in adapters can’t typically be replaced.

My advice is to stick with replaceable network adapter cards, if you can spare the motherboard slot. If you do get a computer with an adapter on the motherboard, though, be aware that you’ll probably be using a slot sometime in the future when you need to upgrade or replace your network adapter.

INSTALLING YOUR NETWORK ADAPTER

Since every network adapter is different, you can’t substitute this section for your network adapter documentation. However, there are several common issues and points that should assist you in successful installation of your network adapter to work with Windows NT Server.

Avoiding Conflicts

Before installing and configuring your network adapter (or any other adapter, for that matter), you’ll need to know what’s in your computer and what resources are used. Make a list of all devices, their interrupts, I/O addresses, memory addresses, and DMA channels. You’ll need this information to determine what resources are free for use by your network adapters. Most network adapters use one interrupt line, one to 16 I/O port addresses, and up to 64K of memory-mapped address space.

Here are some techniques to simplify gathering this information and avoiding conflicts:

Network adapter utility. Some network adapters come with software that detects resource conflicts and helps determine what’s safe for your network adapter to use.
BIOS configuration utility. Your motherboard’s built-in BIOS configuration utility can sometimes provide resource usage information on standard PC devices such as COM ports, parallel ports, and so on.
PCI 2.1 compliance. If your motherboard and additional components comply with the PCI 2.1 voluntary standard, the hardware itself can help resolve interrupt conflicts, sometimes without your intervention. However, you’ll still need to gather information on any ISA adapters using some of the other techniques listed here.
EISA configuration utility. If your computer is an EISA machine, you can use the EISA configuration utility to gather resource information on all EISA adapters. Unfortunately, you’ll need to resort to other techniques for gathering information on ISA adapters.
PS/2 reference disk. If your computer is a PS/2 or other MCA-based computer, you can use the reference disk that came with your computer to gather the resource information for all adapters in the system. These computers will also tell you if you’ve created a resource conflict after installing a new adapter.
Configuration under DOS. If all components are already installed and configured with their drivers to run under DOS, your chances are much higher that they’ll run under Windows NT Server. This isn’t a guarantee, as DOS only talks to one device at a time, so it’s more forgiving of some conflicts.
Microsoft’s MSD utility. If you’re running DOS 6.22 or higher or Windows for Workgroups on the computer, you can run the MSD utility that came with the operating system to scan your computer to determine what resources are in use.
Microsoft’s WINMSD utility. If you’re already running NT version 3.5 or higher, you can run the WINMSD utility that came with NT to scan your computer for a list of resources in use.
Other utilities. You can use one of several third-party software packages to scan your computer to determine what resources are in use. Checkit, Norton Utilities, QA Plus, Probe, and Manifest are examples of products designed to help in this area.

Caution: Don’t take the results of MSD, WINMSD, or similar products as gospel. They may sometimes report that interrupts and I/O ports are free when they really aren’t. This often happens if you’ve installed a device but haven’t loaded the driver for it yet. The operating system sometimes has no way of detecting the device unless its driver is loaded.

Ultimately, you’ll probably need to examine the jumpers physically or check the software configuration settings for each adapter in your computer. This is especially true of ISA adapters. Utilities won’t tell you the whole story.

Selecting IRQs Interrupt levels, or IRQs, are the most scarce device resource in your computer. There are only 16 of them, and all but three already have a standard assignment. You’ll spend more time juggling IRQs to avoid conflicts than anything else. Table 2-4 outlines the IRQs used in a PC environment. This table provides typical interrupt assignments and some advice on whether to use a particular IRQ.

TABLE 2-4 INTERRUPT ASSIGNMENTS AND USAGE ADVICE
*IRQ*	*IRQ Assignment*	*Advice*
0	System clock (15ms timer tick)	IRQ 0 is hardwired and is never available to you. Don’t use it.
1	Keyboard (8042 controller)	IRQ 1 is hardwired and is never available to you. Don’t use it.
2	8259 cascade from IRQ 9	IRQ 2, although apparently available to some adapters, actually isn’t. It’s triggered whenever IRQs 8 through 15 are triggered. Don’t use it. (Some old VGA adapters use IRQ 2. Disable this usage under NT, if possible.)
3	COM2 and COM4 serial ports	IRQ 3 is occupied if you have two or more serial ports installed. If you have only one serial port, this IRQ may be available for assignment to another device.
4	COM1 and COM3 serial ports	IRQ 4 is occupied if you have one or more serial ports installed. If you have no serial ports (a rare condition), this IRQ may be available for assignment.
5	LPT2 parallel port	IRQ 5 is occupied if you have more than one parallel port installed. If you have only one parallel port, this IRQ may be available for assignment. (A virtual LPT2 port via a connection to a network printer doesn’t consume this IRQ.)
6	Floppy disk controller	IRQ 6 is never available to you, unless your floppy controller allows setting the floppy disk IRQ to something other than 6.
7	LPT1 parallel port	IRQ 7 is occupied if you have one or more parallel ports installed. If you have no parallel ports (a rare condition), this IRQ may be available for assignment.
8	Real-time clock/calendar	IRQ 8 is hardwired and is never available to you. Don’t use it.
9	8259 cascade to IRQ 2	IRQ 9, although apparently available to some adapters, actually isn’t. It’s triggered whenever IRQs 8 through 15 are triggered and is electrically tied to IRQ 2. Don’t use it.
10	No typical assignment	IRQ 10 may be available for assignment.
11	No typical assignment	IRQ 11 may be available for assignment.
12	PS/2 mouse port or inport/bus mouse	IRQ 12 is typically occupied if you have a PS/2 mouse port installed. If you have a serial mouse instead, this IRQ may be available for assignment.
13	Math coprocessor error signal	IRQ 13 is never available in a Windows NT 4.0 computer, since 486 and higher processors all use it. Don’t use it.
14	Disk controller	IRQ 14 is used by the standard primary IDE and EIDE (and older ESDI and ST506) hard disk drive controllers. If you have only SCSI drives in your computer, this IRQ may be available for assignment.
15	No typical assignment	IRQ 15 may be available for assignment.

Tip: Because of the arrangement of interrupt controllers in the PC design, IRQs 8 through 15 have higher priority than IRQs 3 through 7. Of the potentially available IRQs, it’s most efficient to assign high-speed devices in the range 8 through 15, and lower-speed devices in the range 3 through 7. Network adapters and disk controllers belong in the high-priority range. However, if you’re running out of IRQs, you may not have the luxury of assigning based on device throughput.

The amount of attention that you need to pay to PCI adapters depends on your motherboard and BIOS design. Some designs take interrupt and address assignments completely out of your hands, dynamically making assignments to adapters when the computer boots. For example, the HP Vectra XU series takes this approach, and it’s sometimes difficult to determine which interrupt was assigned to which adapter. At the opposite end of the spectrum, some designs force you to assign specific interrupts to specific PCI slots statically using the BIOS configuration utility. Still other computers fall somewhere in the middle. They ask you to list interrupts available to PCI slots, and the BIOS makes dynamic assignments from this list at boot time. Although the first approach is the most automatic, I recommend the latter two approaches, as they provide you with more control over and knowledge of what resources are in use in your computer.

Note: Many newer computers, especially PCI-based computers, contain secondary IDE or EIDE adapters built into the motherboard. These devices typically use up another interrupt resource beyond IRQ 14 used by the primary IDE. Be sure you know which interrupts are consumed by your built-in IDE/EIDE adapters.

Selecting I/O Addresses Most network adapters require you to assign a range of I/O addresses to be used for I/O operations. Again, you need to assign a unique range to the adapter to avoid conflicting overlap with any other devices in the computer. Table 2-5 presents the typical I/O address ranges used in the PC, to assist in your quest for unused I/O address ranges.

TABLE 2-5 I/O ADDRESS ASSIGNMENTS IN A PC
*I/O Port Address Range*	*Typical I/O Address Assignment*
170–17F	Secondary IDE/EIDE controller
1F0–1FF	Primary IDE/EIDE controller
200–20F	Joystick/game port
230–23F	Microsoft bus mouse or inport mouse
278–27F	LPT2 parallel port
2E8–2EF	COM4 serial port
2F8–2FF	COM2 serial port
280–29F	Used as default by some network adapters
300–31F	Used as default by some network adapters
320–32F	PC/XT hard disk controller (also used on some PS/2 models)
330–33F	Used as default by some common SCSI host adapters
378–37F	LPT1 parallel port
3C0–3CF	EGA video mode
3D0–3DF	CGA video mode
3E8–3EF	COM3 serial port
3F0–3F7	Floppy disk controller
3F8–3FF	COM1 serial port
A20–A2F	Used as default by some network adapters

Tip: It isn’t always clear, even from the network adapter documentation, how many addresses an adapter actually consumes. If it isn’t documented, assume that the adapter eats up 16 I/O addresses, and you’ll be safe from overlap with other ranges.

Selecting Memory Addresses Many network adapters use memory-mapped buffering, so they need a unique address range in memory to use for reading and writing data. Yet again, you have the task of assigning a unique address range to your adapter to avoid overlap with any other devices using memory address ranges. Table 2-6 shows you which address ranges are already consumed in a typical PC environment.

TABLE 2-6 MEMORY-MAPPED ADDRESS ASSIGNMENTS IN A PC
*Memory Address Range*	*Typical Memory Address Assignment*
A0000–BFFFF	EGA video memory
A0000–C4000	VGA video memory
B0000–B1000	Monochrome video memory
B8000–C0000	CGA video memory
C8000–CBFFF	Hard disk BIOS
CC000–CDFFF	Used as default by some network adapters
D8000–DBFFF	Used as default by some network adapters
E0000–EFFFF	System BIOS expansion: PS/2 extended BIOS or Plug and Play BIOS
F0000–FFFFF	System BIOS (in ROM)

Selecting DMA Channels Some network adapters use DMA channels to bypass the CPU and communicate directly with system memory. DMA channels must be uniquely assigned in the computer as well. Standard PCs use only DMA 0 for DRAM refresh purposes and DMA 2 for the floppy disk drive. The remaining channels 1 and 3 through 7 are available for devices. You’ll need to check the settings and documentation for each of your other installed devices to determine which DMA channels have been assigned.

Selecting the Connector One of the most common errors in configuring combo network adapters is choosing the wrong connector. For example, one version of the Intel EtherExpress 16 adapter has both thicknet and thinnet (BNC) connectors. NT configuration of this adapter defaults to thicknet, so you’ll need to explicitly change it to thinnet if you’re using that type of cable. Before you install a combo adapter, know which type of connector you’ll be using to hook to the network and configure the adapter appropriately.

Selecting the Speed In the Token Ring world, the most common adapter configuration error is choosing a speed that doesn’t match the speed of the ring. Attaching a node running at the wrong speed can bring down the entire ring, so make sure that you configure the adapter with the speed that matches your ring. As mentioned before, most modern Token Ring adapters support both speeds. This is both a blessing (in terms of flexibility) and a curse (in terms of providing an easy way to hose your network).

Tip: Once you have all of the configuration information about the network adapter and the existing components in your computer, write it all down and keep it in a handy location near or on the computer. Many folks keep the configuration information in a plastic or paper pouch attached to the side of the PC. I don’t recommend putting this information in a file on the computer’s hard drive, since you may need the configuration when you’re unable to boot the computer.

Setting the Settings

If you have an older network adapter, you’ll likely need to configure it using jumpers or DIP switches, according to the instructions accompanying your network adapter. I won’t go into detail about how to accomplish this, except to advise you not to use a pencil to toggle DIP switches. The pencil lead not only makes a mess but also can fall into the switch, introducing a potential short circuit. Use a pen or paper clip end to set DIP switches. Most modern network adapters provide a software setup utility. A few have a combination of switches, jumpers, and software configuration.

Caution: Always use an antistatic wrist strap attached to a true ground when you’re handling, inserting, or removing adapters. Even small amounts of static electricity can destroy both your adapters and your computer. Make sure that you’re properly grounded before removing an adapter from its anti-static package.

If your computer is powered down but remains plugged in, you can attach the wrist strap to the metal PC case, since it’s grounded as long as the computer is plugged in. (Unplugging the computer removes its connection to ground and invites static electricity.) If you use this approach for grounding, make absolutely sure that you’ve turned off the computer before you begin surgery.

If you’re configuring an Intel x86 computer, install the adapter in your computer according to the manufacturer’s instructions. (If your computer is RISC-based, you may first need to install the adapter in an Intel computer to run the DOS-based configuration utility. Check with your adapter manufacturer to determine if you need to do this.) Don’t connect the adapter to your network at this stage.

If the adapter manufacturer provides a software configuration utility, it’s probably a DOS-based program. (Most of these configuration utilities won’t run properly under Windows NT, since they attempt to touch hardware directly. You’ll need to run the utility under DOS, unless otherwise specified by your adapter manufacturer.) If your computer already has DOS installed, boot to DOS and run the utility provided. If you don’t have DOS installed on the computer, use another computer to create a DOS boot floppy using the SYS command. Then, copy the network adapter configuration utility and all associated files to the DOS boot floppy, boot your computer using the floppy, and run the configuration utility.

After running the configuration utility, you may need to turn off your computer for a few moments before attempting to test the adapter. Follow the instructions provided by your adapter manufacturer.

Worksheet Entry

In Chapter 1, in Worksheet 1-3 (Windows NT Server Setup) on line 23, record the make and model of the network adapter installed in your computer. On line 26, record the configuration information for your network adapter, including IRQ, I/O address range, and any other network adapter configuration information. You’ll need all of this information during Windows NT Server installation in Chapter 3.

Testing the Adapter

If you’re going to install Windows NT Server as either a PDC or BDC, you’ll need a completely functioning network adapter to succeed during installation. Thus, it’s critical that you make sure your adapter is functioning before proceeding.

Some network adapters come with diagnostic utilities that you can use to test the adapter once you’ve installed it. The value of these utilities varies widely between manufacturers. Functionality ranges from simply reporting your configuration settings to actually sending data over the network to another computer.

Tip: If your diagnostic utility sends data over the network, it’s best to connect it to another computer that’s not on your production network, just in case the new network adapter isn’t configured correctly or is malfunctioning. If you don’t have this luxury, you can use another node on your production network. Just hang on tight to your rabbit’s foot while you run the test.

If your network adapter doesn’t come with a diagnostic utility, and you’re installing on an Intel x86 computer, you can follow the adapter manufacturer’s instructions to install DOS drivers to test the adapter. This may be a major task, so I recommend this last resort only if you’re going to install NT as a PDC or BDC and your adapter doesn’t come with it’s own test utility.

INSTALLING NT AFTER THE ADAPTER

Once the network adapter is physically installed in your computer, it’s properly configured, and you’ve tested it to make sure it works, you’re ready to install Windows NT Server. Chapter 3 will walk you through installation of the entire operating system, using the worksheet you completed in Chapters 1 and 2. In this section, I fill you in on what to expect during the portion of Windows NT Server installation that deals with your network adapter.

Assuming that you take my advice in Chapter 3 and allow NT to detect your network adapter, NT will present you with a dialog box requesting information on your network adapter settings. For example, if your network adapter is an Intel EtherExpress 16, you’ll see a dialog box like the one shown in Figure 2-6. It asks for the IRQ, I/O address, channel ready setting, and cable type. Since different network adapters require different configuration parameters, your dialog box will probably vary from this example.

Figure 2-6: NT requests configuration information about your network adapter.

In each box, just type or select the information that you recorded on your worksheet, then click OK. A handful of network adapters ask for additional information. Type or select the additional settings requested and click OK to continue. Be sure that your settings exactly match how the adapter has been configured, or your network adapter won’t operate correctly.

Tip: The list of available settings that NT presents during the configuration process doesn’t always match the available settings for your adapter. For example, say you have a clone network adapter that’s Novell NE2000 compatible. NT detects the adapter and recommends that you use the NE2000-compatible driver. When you do, you’re presented with a long list of possible interrupt settings, but the adapter may only accept a small subset of these settings.

Moreover, don’t assume that the default settings that NT shows you represent the current configuration of the adapter. If you elect to detect the adapter, NT detects the presence of the adapter but not necessarily its current configuration settings. Since NT doesn’t yet support Plug and Play, what you see may be a good guess at the settings, but they don’t necessarily reflect reality. Know what the settings should be ahead of time, and enter them as requested.

If NT doesn’t detect your network adapter, or if you elect not to have NT scan for adapters, you’ll need manually to choose from a list of adapters. Figure 2-7 illustrates this. Scroll through the list, highlight the appropriate network adapter using the mouse, and click OK.

Figure 2-7: Pick the appropriate network adapter manually, if you elect not to autodetect.

If your adapter isn’t on the list, you’ll need to supply a network adapter driver floppy disk. Use the disk that came with your adapter or the one you created from the NT driver library earlier in this chapter, in the section called “NT Compatibility.” Click Have Disk and follow the instructions displayed.

Once you configure your network adapter and other network components under Windows NT Server, NT will attempt to start the network on your computer. If it can’t, you’ll have the opportunity to review and correct your configuration settings.

That’s really all there is to it. Keep what you’ve learned in this section in mind when you go through the complete Windows NT Server installation process in Chapter 3. Remember, record all of your adapter configuration settings on the worksheet before you begin NT installation.

INSTALLING NT BEFORE THE ADAPTER

Perhaps you’re adding another network adapter to your existing Windows NT Server computer. Or you may have originally installed NT as a stand-alone server before you had a network adapter available. Or you may wish to replace the adapter with a more powerful one. In any of these situations, you can install a new network adapter after Windows NT Server is already installed on your computer.

Caution: To create a working Windows NT Server PDC, you must have a working network adapter in your computer before you install Windows NT Server on it. Before installing an NT BDC, you must have a working network adapter in the computer, and the existing PDC for the domain must already be up and running on the network. In these cases, you can’t install your first network adapter after installing Windows NT Server.

Here are the steps for installing a new network adapter in a computer that’s already running Windows NT Server (Figure 2-8):

Figure 2-8: Using the Control Panel, you can install another network adapter after installing Windows NT Server.

Log on to the server as an administrator.
Click Start Settings Control Panel.
In Control Panel, double-click the Network icon.
Click the Adapters tab. Then click Add.
Choose one of the listed adapters or click Have Disk, if you have a driver disk for your adapter.
Follow the instructions on-screen for configuration of the adapter.

MANAGING NETWORK BINDINGS

As an NT network administrator, you need to be aware of network bindings and how they affect your networked computers. For example, if you want to run two different protocols on two different network adapters in your server, you’ll need to manipulate network bindings. Moreover, to optimize network performance, you may want to change the order of bindings for the most efficient operation. In this section, I discuss what network bindings are and how you can manage them.

What Are Bindings?

Transport protocols are assigned to run on specific network adapters. When a protocol runs on a specific network adapter, the protocol and adapter are bound to each other. The act of creating this association between protocols and adapters is called binding.

The same terminology is also used to describe associations between higher-level network components (typically called services) and protocols. For example, on my Windows NT Server computer, the server component is bound to the NetBEUI protocol, which, in turn, is bound to the NE2000 network adapter. The chain of bindings linking the highest-level network component to the lowest is called the binding path.

Viewing Network Bindings

When you install Windows NT Server, Setup automatically creates an appropriate set of binding paths, linking the network components that you installed. By default, if you elected to install multiple protocols and multiple adapters, all of the installed protocols are bound to all of the network adapters in your computer. (If this configuration is what you intended, you need not worry about managing bindings.) Likewise, when you install a new protocol, it’s automatically bound to your network adapters.

To view your current network bindings, perform the following steps:

Log on to the server as an administrator. Click Start Settings Control Panel.
In Control Panel, double-click the Network icon. Click the Bindings tab.
Under Show bindings for, select the level of the network component whose bindings you want to view. Click the plus (+) sign to expand portions of the binding path. Your options are all adapters, all protocols, and all services. Depending on which one you select, you’ll be able to view the same information in three different ways. If you select all adapters, you’ll see all the network adapters, the protocols bound to those adapters, and the services bound to those protocols. I recommend using this view, since it provides a clear picture of the binding path for each adapter, as shown in Figure 2-9. If you select all services, you’ll see the same information in reverse. For each service, you’ll see the binding path to protocols and adapters, as illustrated by Figure 2-10. Selecting all protocols offers the least information, since it presents just the bindings between protocols and adapters. Services aren’t included, as you can see in Figure 2-11.

Figure 2-9: The adapters view shows the entire binding path for each network adapter.

Figure 2-10: The services view shows the entire binding path for each service.

Figure 2-11: The protocols view shows the bindings between each protocol and adapter.

Enabling and Disabling Bindings

Bindings can be enabled and disabled, based on your use of the network components installed on your computer. For example, you may want the TCP/IP protocol to run on only one of your network adapters.

To disable an existing binding, click the component that you want removed from the binding path, and click the Disable button. When you do, you’ll see an icon next to the component indicating that it’s disabled. Depending on the component that you selected, other components may be disabled as well.

For example, as shown in Figure 2-12, if you disable the NetBEUI protocol, all services bound to it are disabled. Since it’s the only protocol bound to the network adapter, the adapter is no longer bound to anything and is disabled also. If you had another adapter in the computer, with its own set of bindings, those bindings would be unaffected.

Figure 2-12: Disabling the only protocol binding on an adapter disables the entire binding path.

Enabling a binding is just as easy. Click the component you want enabled in the binding path, and click Enable. When you do so, the icons next to the affected components will change to reflect the enabled bindings.

Changing Binding Order

The order of bindings reflects a relative priority between network components. Components bound first have higher priority than components bound last. You can change the order of network bindings to optimize your computer’s use of the network. For example, if your server is running both NetBEUI and IPX/SPX protocols, but most computers on your network run just IPX/SPX, you may want to change the order of protocol bindings to services so that IPX/SPX is always tried first when establishing connections.

Figure 2-13 and Figure 2-14 illustrate this example. To move a protocol, select the protocol that you want to move, and click either Move Up or Move Down to change its relative position. Figure 2-13 shows the NetBEUI protocol selected. After clicking the Move Down button twice, the NetBEUI protocol becomes the last protocol bound to the server component, as shown in Figure 2-14.

Figure 2-13: Select the protocol that you want to move.

Figure 2-14: The NetBEUI protocol has been moved below the IPX/SPX protocol.

You can apply this same technique to other services. In this example, it’s wise to change the protocol binding order under the workstation service (the redirector) in the same way. If you have Windows NT Workstation clients on your network, it’s a good idea to optimize their protocol binding order as well.

SUMMARY

In this chapter, you’ve learned the ins and outs of cabling, network technologies, and network adapters. Whether or not you’re personally involved in designing and installing your network, you now have a better understanding of, and appreciation for, what’s out there. Perhaps I’ve presented some options and planning considerations that you haven’t considered before.

Note: Be sure to complete the network adapter information in the Chapter 1 worksheet before you proceed to Chapter 3. There, you’ll use the worksheet to help perform an actual install of Windows NT Server on your computer. Because you’ve prepared well for installation, you'll have a lot less to worry about during the setup process.

<< Previous Page Next Page >>

Microsoft Windows NT Server Administrator's Bible: Option Pack Edition