Selecting the right components for your network can make the difference between an easy installation and a difficult one. There are a great many network interface cards (NICs), hubs, cables, and other components on the market today, and it is not always true that more expensive equipment is better. This chapter will examine the criteria you should use when shopping for these components and give you some idea of where to go for the best prices.

Networking purchases often do not stop at NICs, hubs, and cables, though. While it is not necessary to purchase new PCs, software, and other components when you network your computers together, you should consider the type of network you are trying to build and the location where you are going to build it when you are making new purchases.

The installation of a NIC and the required networking software does not add a significant burden to the average PC, so it is safe to say that any computer that can run Windows 98 can be connected to a Windows 98 network, as long as it has a slot available for a NIC. However, if you do intend to purchase any new PCs, you might want to spend some time considering the roles that your individual computers can play on the network.

Although many of the organizational decisions about where you will store your applications and data on the network can wait until after you've installed the LAN, some can affect your PC purchasing decisions. The primary question involves whether you intend to designate one or more systems as file servers. On a Windows 98 network, a file server uses the same hardware and software as any other computer; the difference lies mainly on how you use the machine.

There are several benefits to designating one computer as the location where all of the data shared by your network users is stored, including the ability to create and administer network shares on a single machine and more easily protect the data with passwords and backups. These advantages are discussed more extensively in "The Single Server Scenario," found in Chapter 7, File and Printer Sharing.

However, a computer that is sharing its resources with multiple network users simultaneously might require hardware that is better suited to server applications. If you are purchasing a new PC for use as a Windows 98 file server, there are several choices you can make that will improve its performance in a network environment. These choices center primarily around the storage subsystem, since this is the area that will be impacted the hardest by network use.

For a computer that will be serving files to many different users, the best upgrade you can get is to switch from EIDE (Enhanced Integrated Drive Electronics) hard disk drives to SCSI (Small Computer Systems Interface). Most of the PCs sold today use EIDE drives, as they are fast, cheap, and available in high capacities. The introduction of the EIDE interface (replacing IDE) doubled the device capacity in a system from two to four, making it possible for an average system to contain up to two hard disks, as well as CD-ROM and Zip drives.

However, SCSI drives are inherently more efficient than EIDE in a multitasking environment, because the SCSI adapter can queue file requests and process them in a more intelligent manner without placing an additional burden on the system processor. SCSI is also better suited to the simultaneous use of multiple devices on the same bus. A tape drive or CD-ROM burner on the same SCSI bus as your hard disk drives will perform better when writing to tapes or CDs, because the data will be transmitted directly from the hard drive to the recording device, without having to pass through the system bus.

SCSI can also support more devices than EIDE: up to seven or fifteen, depending on the type of SCSI adapter you choose. You can purchase SCSI tape drives, CD-ROMs (both read-only and read/write), cartridge drives, scanners, and other devices, in addition to hard disk drives. Unlike an EIDE interface, on which you must configure the drives as masters and slaves, SCSI devices all function as peers on a specialized bus that is independent of the system bus.

Many computer manufacturers provide an alternative configuration for their systems that uses SCSI instead of EIDE drives and includes a SCSI host adapter, but you can also add a SCSI bus to an existing system without too much difficulty. To do this, you must purchase a SCSI host adapter card that plugs into the computer's expansion bus. Adapter cards are available for the PCI (Periperal Component Interconnect) and ISA (Industry Standard Architecture) buses, as well as older technologies like EISA (Extended Industry Standard Archiecture), but for a server you are better off selecting a card that uses the fastest bus available to you, which will usually be PCI.

For these reasons, SCSI drives are more expensive than their EIDE counterparts, and you also have to purchase a SCSI host adapter card for the system as well, which also adds to the price. However, if you choose to store all of your shared data on a single server, it can be worth the expense.

The original SCSI standard was designed to move data at up to 5 MB/sec over an 8-bit wide bus. The SCSI-2 standard increased the speed to 10 MB/sec, and there have been several newer standards introduced since then that call for wider buses and transfer rates up to 80 MB/sec. Most of these standards are backwards compatible, meaning that if you use a host adapter card that can support the most advanced standards (such as Fast and Wide SCSI and Ultra SCSI), you can connect virtually any SCSI device to it.

Adaptec, Inc. is the manufacturer of a large line of SCSI host adapters and related hardware that represent the industry standard. For a Windows 98 system functioning as a file server, you will most likely want a SCSI host adapter that supports both internal and external devices, and that includes its own BIOS. This way, you can use the adapter with your internal SCSI hard disks as well as external devices like tape drives.

The system BIOS in a PC supports EIDE, but not SCSI drives, so most SCSI adapters contain a supplemental BIOS that enables the system to use a SCSI hard drive as a boot device. However, there are some low-priced adapters that are designed for use only with an external device, like a scanner, that have no BIOS and no connector for internal devices. Be sure to check for these features when evaluating SCSI adapters.

A basic PCI adapter like the Adaptec AHA2940, with BIOS and both internal and external connectors, runs about $200. For $25 to $50 more, you can move up to the Adaptec AHA2040UW, which supports Ultra Wide SCSI, with transfer rates up to 40 MB/sec. Both of these adapters are PCI devices, suitable for use in a workstation or small server, and there are many other models by Adaptec and other manufacturers. When you spend the extra money for an adapter that supports high speed transfers, be sure that the hard drives you purchase also support the same standard, or you may find your expensive equipment running at basic SCSI-2 speed.

It is also important to make sure that the system functioning as a file server has sufficient memory installed to keep it running efficiently. Windows 98's virtual memory system swaps data from the memory array to a paging file on the hard drive as needed. When the system is short on memory it pages more frequently, and the additional burden on the disk storage subsystem can degrade the performance of its file serving tasks. With memory prices as low as they are, it would be well worth the extra cost to have at least 64 and as much as 128 megabytes or more of memory installed in the machine.

Purchasing a PC specifically for use as a network workstation raises more questions about what the computer does not need than what it needs. In a business environment, you may not require some of the components that have become standard equipment in the PCs sold for home and individual use. For example, when your computers are networked, there is often no need for each machine to have a CD-ROM drive. Legitimate business uses for CD-ROMs include the installation of new software and access to CD-based reference tools, both of which can be provided using drives that are shared over the network. Playing games and music CDs may be activities that you want to discourage in your office, and there's no better way to do this than to deny users the tools required for them.

In the same way, you should consider whether you want users to have multimedia sound capabilities. A collection of computers in the same room, all with speakers running, can generate enough noise to disturb the working environment and make an unprofessional impression on the phone.

Another question to consider is the amount of disk space required on each computer. If you plan to store all of your shared data on a Windows 98 file server, then you might be able to purchase workstations with smaller hard drives and save some money there. Typical PCs targeted at home users often have drives with capacities of six to ten gigabytes or more. Depending on the applications you intend to run on your network systems, you may find that two to four gigabytes is more than sufficient.

Of course, in a home or home office situation, you may want the capabilities that a computer equipped for multimedia provides. In this case, your concerns when purchasing PCs are as follows:

• That the computer has sufficient memory to accommodate the networking components.
• That the computer has at least one expansion slot free for a NIC.

Although the official system requirements for Windows 98 say that it will run with 16 megabytes of memory, you should consider 32 megabytes to be the minimum for a workstation. This is not so much due to the requirements of the networking modules, which are minimal, as it is to the general performance of the operating system. Windows 98 which responds better to a RAM increase than to virtually any other upgrade.

Most computer manufacturers specify how many expansion slots are included in their machines in their literature and on their web sites, but they rarely say how many of those slots are left free in the finished product. Motherboards today can contain integrated video, audio, SCSI, and even network adapters, so it can be difficult to predict how many slots will be free just by counting the peripherals included in the system. When you're buying new computers, whether workstations or a server, be sure that there is a slot free of the appropriate type for the NIC you want to use.

The fabric of an Ethernet network itself is composed of cables and the hubs and NICs to which they connect. At the very least, you will have to purchase a NIC for each computer you want to connect to the network and cables to connect them. If your network will consist of only two computers, you can connect them with a special type of cable called a crossover cable. However, for three of more computers, you must purchase a hub that will connect to each system's NIC.

Selecting the proper hardware for your network can be a daunting task. There are a huge number of products on the market, and manufacturers use a great deal of jargon to describe them. The objects of the exercise are as follows:

• To select products that work with your computers.
• To select products that work with each other.
• To select products that provide the desired level of service
• To purchase the products at the best possible price.

The following sections discuss the various types of NICs and hubs on the market and the criteria you should use when selecting components for your network.

There are only two practical choices when it comes to the speed at which you will run your Ethernet network: standard 10 Mbps Ethernet and 100 Mbps Fast Ethernet. (Gigabit Ethernet is too new and expensive a technology to be a practical alternative at this time.) In general, standard Ethernet NICs and hubs will be a bit cheaper than their Fast Ethernet counterparts, but usually not overwhelmingly so.

Virtually all of the Fast Ethernet NICs on the market are dual speed, running at either 10 or 100 Mbps. The Fast Ethernet specification defines an auto-negotiation function that enables NICs and hubs to exchange information about their capabilities with other devices on the network and configure themselves to run at the fastest speed they have in common. Thus, you can begin buying NICs that are capable of Fast Ethernet speeds even if you only intend to run your network at 10 Mbps right now. The difference in price between standard and Fast Ethernet NICs can be as little as $10.

If some of your computers have only standard Ethernet NICs, or if you do not yet want to spend the money for a Fast Ethernet hub, it's still a good idea to spend the few extra dollars on any new NICs you're buying for 10/100 devices.

Hubs are available in 10 Mbps, 100 Mbps, and dual-speed (10/100 Mbps) configurations. Dual-speed hubs like those from manufacturers like Asante and 3Com support both 10 and 100 Mbps nodes on the same network by receiving data at either 10 or 100 Mbps and transmitting it out the rest of the network at the speed appropriate for each node (calculated through auto-negotiation).

Like the dual-speed NICs, these hubs enable you to preserve your investment in networking hardware by providing a base for future upgrades. Even if you already have computers with 10 Mbps Ethernet cards installed, you can purchase a dual-speed hub and upgrade the NICs at your convenience. The hub will automatically detect the high-speed capabilities of each NIC as you install it and adjust itself accordingly.

Half of the battle that constitutes the NIC installation process is won by selecting an appropriate card both for your network and your computer. The basic type of card you will select is based on the protocol you run at the physical link layer (Ethernet, in this case) and the type of cabling you will use to form the network medium. Equally important, however, is selecting cards that are properly supported by your computers. After all, even the best Ethernet card on the market is useless if your computer doesn't have the right type of bus slot free in which to install it, or the system resources it requires to run.

All of the mathematical computations, that is, all of the actual computing in a PC, is performed by the microprocessor. Most of the other components in the machine, including the memory, the disk drives, and the expansion cards, are connected to the processor by an electrical conduit built into the system's motherboard known as a bus. Every NIC must connect to the system bus through an expansion slot in order for the data transmitted and received over the network to reach the processor and the other components.

There are two major bus types used in desktop systems today, PCI (Peripheral Component Interconnect) and ISA (Integrated Systems Architecture). In addition to these two is the PC Card (formerly known as the PCMCIA, or Personal Computer Memory Card International Association) standard used by virtually all portable systems, such as laptops and notebooks. While it is not necessary to understand the inner workings of these bus types in order to buy a NIC, you should understand the effect of your selection on the installation process.

ISA is an older bus architecture that has been around since the original IBM PC was released in 1981, while PCI is newer, having been introduced in 1992. The main differences between the two are speed and ease of configuration. PCI is inherently faster than ISA, both because it transfers data at a higher speed and uses a wider bus. The performance capabilities of the two bus types as found in a modern computer are compared in Table 4-1.

Bus speed is not as great an issue in the selection of a network card as you may think. As you can see in the table, even the venerable ISA bus has a rated throughput of over 64 Mbps, while a standard Ethernet network runs at only 10 Mbps. This bus throughput statistic is theoretical, and may be more that twice the actual realized figure, but even so the bus operates at least three times faster than a traditional Ethernet network.

If you decide to install a Fast Ethernet network, running at 100 Mbps, the ISA bus may be a bottleneck, but only if you are regularly transferring very large amounts of data. In most cases, ISA provides enough speed to support normal networking activities, but a PCI NIC is certainly a viable, albeit more costly, alternative. Above all, make sure that you have a free bus slot of the appropriate type available in each machine you plan to connect to the network. A computer that already has a lot of peripheral devices installed in it such as sound cards, internal modems, DVD drives (which usually include a separate MPEG decoder board), and SCSI adapters may be severely short on bus slots, so it is important look inside your computers to see what slots are available.

Before you even purchase a NIC for a computer, you should check to see what resources the machine has free. The first step is to determine which (if any) slots are unoccupied. Simply looking at the slot covers on the back of the computer is not good enough for this purpose. Some of the expansion cards in today's PCs (such as the MPEG decoder cards used with DVD drives, for example) do not have external connectors, and may be occupying a slot that looks empty from the outside of the machine. You also cannot tell whether a free slot uses ISA, PCI, or any other bus by looking at the outside of the box.

You have to open up your computer to see what slots are free and, in many instances, opening the case is the most difficult part of the entire NIC installation process. Many computer cases now use thumbscrews that eliminate the need for a screwdriver, but you still often have to wrestle with the cover a bit before it comes off. Make sure the system is powered down and unplugged before you open the case, and be sure not to damage any of the computer's internal components with the edges or corners of the cover. It is also a good idea to wear a grounding strap whenever you work inside a computer, or be sure to touch a metal ground before you handle any of the internal components.

Desktop computers use various types of internal architectures that determine the placement of the expansion slots. Larger systems have the slots on a single motherboard that also holds the processor, the memory, and other components. So-called "slimline" systems often have a riser card perpendicular to the motherboard that contain the slots. These systems are not as tall as standard desktops, because the installed expansion cards are stacked parallel to the motherboard instead of being mounted perpendicular to it.

Most PCs today have both ISA and PCI slots, but how many of each type and whether they are occupied or not is dependent on the computer and the components installed in it. When you look inside your computer, you should see a row (or maybe two) of slots, arranged next to each other so that the ends of the expansion cards protrude through the computer case, either horizontally or vertically. You can always tell an ISA from a PCI slot by examining the row of connectors within the slot. The ISA card connector is broken into two sections, one roughly twice the length of the other. PCI slots have a single unbroken line of connectors, and are roughly two-thirds the overall length of an ISA slot. The basic configuration of these two slot types is shown in Figure 4-1.

If you find that a computer has no free slots for a network card, you are not necessarily out of luck. Although a network interface must be an internal device, not all other peripherals have to be also. You may be able to free up a slot by substituting an external modem for an internal one, or by using a tape drive, cartridge drive (such as a Zip), or scanner that connects to a parallel port instead of to a SCSI adapter card.

If you have both ISA and PCI slots available in a system, your decision between the two should be based on the prices of the respective NICs and the amount of data that will be travelling around your network. If, for example, you are building a Fast Ethernet network for use by a graphic design firm that regularly deals with enormous digital image files or works with streaming video, then PCI NICs are preferable, even if they are more expensive. For normal business networking, such as file and printer sharing and Internet access, ISA NICs are sufficient.

With the general trend towards higher speed networks and the desire for simplified administration, PCI adapters have become increasingly popular in recent years. In fact, you may find that PCI adapters for Fast Ethernet networks are more readily obtainable than ISA cards and are likely to be cheaper due to sale prices and rebates. Thus, assuming that the computers you intend to network have the required slots free, PCI NICs are recommended over ISA cards for their superior performance, simplified installation and administration, and possibly their lower price.

As important to your NIC purchasing decision as the bus type the card uses is the matter of hardware configuration. Every expansion card in a computer requires certain system hardware resources that enable the other components in the computer to access its functions. These resources can include the following:

• Interrupt request lines (IRQs) - Integers from 0 to 15 identifying hardware lines that devices use to send interrupt signals to the system processor.
• I/O port addresses - Unique hexadecimal numbers that identify specific peripheral devices installed in the system.
• Memory addresses - Hexadecimal numbers that identify locations in the upper memory area allocated to specific hardware devices.

At the very least, every NIC will require an IRQ, and may also require one or both of the other resources.

The hardware resources within a computer are finite. For example, a PC has sixteen IRQs, most of which are occupied by the system's standard equipment. When you install a new device into the system, such as an expansion card, it will nearly always require one of the available IRQs. If you configure a device to use an IRQ that is already assigned to another device, a conflict will result that at the very least will cause one or both devices to cease functioning properly. In some cases, the conflict could cause the entire computer to crash. The same holds true for the other types of hardware resources used by peripheral devices.

The other problem with assigning hardware resources to new devices is that the devices themselves may only support the use of certain resource values. Thus, you may find that your new NIC supports only IRQs 3, 5, and 10 and your computer has none of those IRQs free. In a case like this, if the card is going to work, you (or the operating system) must configure one of the other devices to use a different IRQ in order to free up 3, 5 or 10.

In this respect, the advantage of cards that use the PCI bus is that the system can utilize Plug and Play to configure the appropriate hardware resources automatically. ISA cards, on the other hand, may require that you manually select the hardware resources that the card should use by adjusting small jumpers or flipping DIP switches on the board, or by running a configuration utility supplied by the NIC's manufacturer.

Problems may arise with respect to hardware resources when you have a computer that is already heavily loaded with peripheral devices. It is entirely possible for a computer to have no available IRQs or other resources because of the hardware already installed in it. It is also possible for there to be no way to reconfigure the existing hardware in such a way as to free up the resources needed by the new device.

Before you purchase NICs, you should determine what hardware resources (if any) are available in each of your computers and make sure that the cards you select support those free resources.

In most cases, if you are not using old equipment and if there is an expansion bus slot free in the computer, there will be sufficient resources of the right types available to run a network interface card. This is because the NICs on the market today typically support a wide range of hardware resource values or use PCI and Plug-and-Play to configure the resources automatically. Years ago, it was easy to buy a NIC that could run using only two or three possible IRQs, for example. Today's NICs often support eight or more IRQs.

When you are evaluating NICs (and particularly when you are considering cards that are either very old or very cheap), it is a good idea to find out which types of hardware resources the cards require and what values for those resources the card supports, so that you can determine if they will run in your computers. If the NIC manufacturer's literature or web site does not provide this information (and most should), then you may be able to find it by running the software utility supplied with most of the cards sold today, that you can use to configure the card's hardware resource settings. This utility is usually bundled with the software drivers for the NIC and is probably available for download from the manufacturer's web or FTP site.

In addition to determining the hardware resource settings supported by a NIC, you must also examine your computers to discover whether or not the required resources are available. Even if your systems will automatically configure the cards using Plug-and-Play, knowing what resources are required and which are available can be helpful. Should you be installing ISA cards or should Plug-and-Play fail to function properly, you will need this information to manually configure the card.

Windows 98 includes a Device Manager application that enables you to view and modify the configuration parameters for every hardware component in the system. To open the Device Manager, launch the System Control Panel and select the Device Manager tab to display an expandable of the computer's components. You can use this list to examine the properties of any device in the machine.

When you select the Computer icon at the top of the display and click the Properties button, you see the Computer Properties dialog box shown in Figure 4-2.

The radio buttons at the top of the View Resources page enable you to display all the system hardware resources of the following types:

• Interrupt request (IRQ)
• Input/Output (I/O port address)
• Direct Memory Access (DMA)
• Memory address

Armed with this information you can manually install your NIC. If you need to specify an IRQ because Windows requests it or you need to set a jumper on an ISA card, you can now determine which IRQ values should work by comparing those supported by the card with those available in the computer. If the Device Manager shows all of the resources of a particular type as allocated, you may not be able to successfully install a NIC in that machine without disabling or removing another device in the system.

Plug-and-play (PnP) is a standard developed by a consortium of hardware and software manufacturers including Intel, Microsoft, and others that is designed to simplify the process of upgrading a PC. When it functions properly, the Plug-and-Play standard completely eliminates all concerns about hardware resources when installing an expansion card or other peripheral device. Computers that support PnP should automatically detect new PnP devices in the machine during the boot sequence, automatically configure the new hardware, and install the device driver required for the new component to communicate with the rest of the computer. Nearly every NIC on the market today and almost every computer purchased since 1996 supports PnP -- check the cards that you intend to purchase to be absolutely sure.

When you install a new NIC into your computer, PnP works by detecting the configuration requirements of the card and the available hardware resources in the computer. If the card requires an IRQ, for example, the system locates an unused IRQ in the computer and configures both the card and the device driver to use it. If necessary, PnP can dynamically adjust the hardware resources of other devices in the system in order to provide support for all of the installed components.

When PnP works, it works very well. You simply insert a NIC into a bus slot and reboot the machine. During the startup sequence, Windows 98 displays a message box informing you that it has detected a new device, identified it, and is installing the required driver. If the operating system distribution files are not stored on your hard disk, you will have to insert the Windows 98 CD into the CD-ROM drive, but the rest of the process is completely automatic.

Unfortunately, when PnP fails to properly detect, install, or configure a new device, the problem can be more difficult to correct than it would be with a non-PnP system. Because Plug-and-Play essentially reconfigures all of the hardware in the computer during each reboot, you may find that an IRQ that was available before the NIC was installed is occupied by another device after you insert the new card.

Despite occasional difficulties with specific devices, however, Plug-and-Play often works quite well, particularly with network interface cards, and it is strongly recommended that you purchase new cards that support the PnP standard. (Actually, it would probably be difficult to find cards currently in production that do not support the PnP standard.)

Since Ethernet supports several different types of cable, NICs for Ethernet networks are available with various combinations of cable connectors. Virtually all of the NICs in production today have RJ-45 connectors for twisted pair cable, since this is by far the most popular network medium currently in use. Many NICs, though, also have other cable connectors, supporting either thin Ethernet (with a BNC connector), thick Ethernet (with an AUI connector), or both.

Thin Ethernet and thick Ethernet networks only support a maximum speed of 10 Mbps, so you are less likely to see dual speed NICs with multiple connectors. However, some companies do make them, to facilitate upgrades from thin or thick Ethernet at 10 Mbps to 100BaseT.

The inclusion of multiple connectors can add substantially to the cost of a NIC, in some cases almost doubling the price. In the case of a small Windows 98 network, you will probably be using unshielded twisted pair cable and should never need the BNC or AUI connectors.

Few computer stores maintain a large selection of NICs, and you might find that the only NICs a store has in stock are combo cards supporting two or three cable types. This type of card is beneficial to the retailer because it provides maximum compatibility while minimizing inventory, storage, and display expenses. Don't pay for hardware you don't need; purchase NICs with RJ-45 connectors only, even if you have to special order them. The difference in price can be substantial.

Some NIC manufacturers produce separate lines of cards for servers and workstations, which may cause some confusion to the first-time buyer. As far as basic NIC functions are concerned, there is no difference between the card in a server and that in a workstation. Some manufacturers categorize their products using these terms in order to recommend them for specific applications. Gigabit Ethernet cards, for example, might be classified as server NICs because they are most likely to be used in a server. This does not mean that the card would not function in a workstation, though.

Other manufacturers, however, produce NICs with features specifically designed to optimize server performance. For example, 3Com markets a Fast Ethernet card intended for servers that includes a load balancing feature that enables you to insert multiple NICs into a single computer and have them work together as a single, high-bandwidth, virtual NIC. Products like these are naturally more expensive than standard workstation NICs and are intended for servers on large, busy networks. Even if you designate one of your Windows 98 systems as a file server, you do not need features like these in a small network. A standard NIC intended for use in a desktop or workstation system will be perfectly adequate.

Whenever possible, it is a good idea to select a manufacturer and model of network card that is compatible with all of the computers on your network. This not for compatibility reasons (since all Ethernet cards should be able to communicate with each other), but rather for practicality and price. Familiarizing yourself with the properties, drivers, and limitations of one NIC is easier than working with several different ones.

In addition, most of the major NIC manufacturers sell their cards in multiple unit packs. By saving on packaging and documentation, manufacturers can sell these packs at a substantial discount. Prices for NICs (like those for all computer hardware) change constantly, but to illustrate the amount of money you can save by buying all of your NICs at once, the following table lists one vendor's prices for the 3Com Fast Etherlink 10/100 PCI card in several multiple unit packs:

Thus, you can see that if you have plans to expand your network in the future, it might be wise to purchase a large number of network cards at one time, even if you don't have use for all of then yet. For even more savings, you might be able to organize a community purchase with other network administrators in your area, such as those you might meet in a users' group meeting or other computer organization.

To minimize the chance of having installation difficulties, it is recommended that you stick to one of the well-known NIC manufacturers, like 3Com or Intel. A basic 10 Mbps 3Com Etherlink III ISA card with an RJ-45 connector only can be had for as little as $60, with greater discounts for multiple unit packs. A great many of the technical support problems that occur during NIC installations are due to the use of old, unsupported cards or no-name bargain models. Spending several hours of your valuable time troubleshooting NIC installations is hardly worth the few dollars you can save by purchasing cheap cards.

The hubs used on small Ethernet networks are relatively simple, electrical devices that receive signals through any one of their ports, amplify them, and transmit them out through all the other ports. This way, every packet generated by a workstation on the network is transmitted to every other workstation. It is up to the network cards in the individual computers to read the header information in each packet and determine whether it should send it up the protocol stack or discard it because it is intended for another system.

Other, more advanced types of hubs operate at a more intelligently by reading the incoming packet headers themselves and forwarding each packet only out of the port connected to the computer that is the packet's destination. These can be called smart, intelligent, or switching hubs, because they minimize the amount of extraneous traffic sent over the network. Hubs like these are absolutely not needed on a small Windows 98 network. Even a dozen computers working at full capacity all the time can be adequately serviced by a Fast Ethernet network without the need for this type of device.

Hubs that function purely as electrical devices are sometimes called dumb or repeating hubs. A repeater is a device that amplifies an electrical network signal, so that it can travel a longer distance without attenuating. Attenuation is the tendency of a signal to weaken in strength as it travels along a cable or other medium. Because of the signal's tendency to attenuate, Ethernet networks using twisted pair cable are limited to a maximum cable segment length of 100 meters (328 feet). However, since the hub functions as a repeater, you can conceivably connect two computers that are 200 meters apart, as long as the hub is exactly in the middle.

Because they work only with the electrical signals traveling over the network, and do not have to interpret them, standard hubs are relatively inexpensive devices. The primary factors that determine their cost are the number of ports they have and the transmission speed they support.

You can buy hubs that support various numbers of workstations, usually in multiples of four. Four-, eight-, twelve-, and sixteen-port hubs are common, with prices rising accordingly with the number of ports. When you're shopping for a hub, it's a good idea to plan for a reasonable amount of short term network expansion, but you need not anticipate your long-term needs too accurately, because you can always purchase another hub at a later time and connect it to your existing hub to expand your network.

One port on every hub (usually the highest numbered one) uses different wiring connections to make it an uplink port. You use the uplink port to connect one hub to another, in order to expand the network. On some hubs, this port is switchable, so that you can use it to connect either to another hub or a workstation. On hubs where the uplink port is not switchable, the only way you can use this port to connect to a workstation is to use a special cable called a crossover cable.

Be aware that some hub manufacturers include the uplink port in the advertised number of ports, while some do not. If you have eight computers to connect, be sure to check whether the 8-port hub you intend to buy has eight ports total or eight ports plus an uplink port.

Another option that can add greatly to the price of a hub is the ability to manage the ports using SNMP (the Simple Network Management Protocol). A managed hub includes an agent that supplies status information about each port to a centralized management console. This capability can be invaluable on a large network that uses a great many hubs, but for a small Windows 98 network, it is completely unnecessary. To give you an idea of how this feature can affect the cost, one manufacturer's suggested retail price for an 8-port, unmanaged, 10BaseT hub is $145, while the equivalent 8-port, managed hub is $415.

The Ethernet transmission speed that the hub supports is another determining factor in the price of the unit. For the network to function, both the NICs and the hubs must support the same speed. Ethernet hubs are available in standard speed 10Mbps, Fast Ethernet (100 Mbps), and dual-speed (10/100 Mbps) configurations. Unlike Fast Ethernet NICs, which are virtually all dual-speed, hubs that support both 10BaseT and 100BaseT are substantially more expensive than those supporting only 100 Mbps.

For comparison purposes, Table 4-3 shows the manufacturer's suggested retail prices for three equivalent 3Com OfficeConnect 8-port hubs in 10BaseT, 100BaseTX, and dual-speed configurations, as well as the street prices at several online dealers.

Even though these street prices are for one particular hub manufacturer only, and will almost certainly be different by the time you read this, they illustrate that the huge jump in the suggested retail prices of the 10 Mbps and 100 Mbps hubs is not necessarily proportionate in the street prices. The prices also demonstrate that the dual-speed hub is substantially more expensive then one supporting Fast Ethernet only. As a result, purchasing dual-speed hubs is not recommended unless you intend to upgrade a network from 10BaseT to 100BaseT.

If you are installing a new network and are undecided between regular and Fast Ethernet, the difference in hardware prices as they stand today comes to approximately $10 per NIC and $100 extra for the hub. If you have no intention of ever upgrading to Fast Ethernet, then a 10BaseT network will save you a bit of money. However, a purchasing plan intended to run the network at 10 Mbps now and facilitate a later upgrade to 100 Mbps will save you perhaps $100 in the cost of standard Ethernet NICs (which you must later replace with Fast Ethernet models), but cost you an extra $170 for a dual-speed hub that will not need to be replaced later. Clearly, the extra cost for the hub and the additional work and expense required to upgrade the NICs makes this approach impractical. Considering that you are getting ten times the speed from 100BaseT and will have a network that should be able to serve you for several years to come, Fast Ethernet is worth the additional cost.

Most of the Fast Ethernet hubs on the market support the 100BaseTX standard, which requires the use of Category 5 UTP cable throughout the network. If you will be constructing a Fast Ethernet network using an existing Category 3 UTP cable installation, you must use 100BaseT4, which can run over Category 3 cable. NICs and hubs supporting 100BaseT4 are relatively rare, in comparison to 100BaseTX equipment, and significantly more expensive.

Several of the major network hardware manufacturers market lines of hubs and NICs that are intended for use in homes and small offices. Typically, these products offer good performance at a good price, but they are usually 100BaseTX components. To get 100BaseT4 support, you must look to the products that are intended more for the corporate market. This means that you are more likely to have to pay for features you don't need, like additional BNC or AUI connectors. In fact, for a small network, it may even be more practical, both in terms of economy and convenience, to install new Category 5 cable so that you can purchase the much more inexpensive and readily available 100BaseTX equipment.

Another feature to look for when purchasing Fast Ethernet hubs is the device's repeating capabilities. The Fast Ethernet standards define two types of repeaters, called Class I and Class II. The standard dictates that the repeater must be marked with the appropriate Roman numeral inside a circle, to identify its capabilities.

Class I repeaters work by translating incoming signals to a digital format and then translating them again into the appropriate format for each outgoing port. This enables a Class I repeater to connect cable segments that use different signaling schemes, such as 100BaseTX and 100BaseT4. This type of repeater can have longer timing delays, so the cabling guidelines are more strict. There can only be one Class I repeater between any two workstations on a network.

Class II repeaters have no translation capabilities and immediately send all incoming data out through the device's other ports without modification. As a result, you cannot connect cable segments that use different signaling schemes with a Class II repeater. However, since the device's timing delays are shorter, it is permissible to have up to two Class II repeaters between two workstations.

What this means for a small Fast Ethernet network is that if you purchase a hub that is a Class I repeater, then you can't expand your network by chaining two (or more) hubs together, because this will result in a situation where a workstation connected to one hub will have to go through two repeaters to get to a workstation connected to the other hub. If you purchase hubs that are Class II repeaters, then you can chain two hubs together, but you can't connect 100BaseTX and 100BaseT4 workstations to the same hub.

Since the network you will be building will almost certainly use the same type of cable for all of your connections, there is no need for the translation capabilities provided by Class I hubs. The ability to expand the network by adding another hub is much more valuable to the small network administrator, so it is recommended that you purchase Fast Ethernet hubs that are Class II repeaters. Most of the hubs intended for home and small business use are Class II repeaters.

The vast majority of the local area networks installed today use Category 5 unshielded twisted pair (UTP) cable to form the fabric of the network. UTP cable is easy to install, easy to conceal, and economical. Because telephone systems also use twisted pair, the cable installation for large office networks is often subcontracted to the same company installing the phone cables. When it comes to purchasing new cable, the question of the cable's category rarely comes into play, because virtually all UTP cabling today is Category 5.

The only time you might encounter a lesser grade of cable is when you are working with an existing installation that was intended for use with telephones. The phone systems in many offices use Category 3 (voice grade) cable that is quite sufficient for a 10BaseT network, but cannot run 100BaseTX. Fast Ethernet on Category 3 cable requires 100BaseT4 NICs and hubs.

When purchasing the cables you need to build a new Ethernet network, the primary decision you must make is whether to purchase prefabricated cables or build them yourself. Prefabricated Category 5 cables come in a variety of lengths, usually up to 50 feet, as well as different colors, and are readily available from most large computer stores. Cables like these are ready for use right out of the bag. Simply plug one end into a hub port and the other end into a computer's NIC, and it's ready to go.

Building cables yourself means buying a spool of bulk cable, cutting off a piece, and connecting RJ-45 jacks to both ends. You will save money on the cable itself, but you will also need the RJ-45 connectors, a crimping tool used to attach the connectors to the cable, and a reasonable amount of skill that only comes from practice. For a small network, building your own cables is usually impractical, for a number of reasons. First, the cost of the crimping tool will probably offset the money you saved by purchasing the cable in bulk, and second, by the time you start to get the hang of crimping the connectors onto the cable properly, you'll be finished making all the cables you need.

There are circumstances in which building your own cables for a small network makes sense, such as when you already have experience working with twisted pair cable in this way, you know someone who will lend you a crimper, and you plan to be making a lot of cables in the future. It can also be sensible to crimp your own connectors when you will be running the cable through walls, ceilings, or crawlspaces. By pulling the cable first and then attaching the connectors, you can drill smaller holes, resulting in a less intrusive installation.

In most cases, however, it is far more convenient to purchase prefabricated cables. As far as the expense is concerned, prefabricated cables certainly cost a good deal more than their component parts, but they are still not terrible expensive. A top-quality fifty-foot cable should cost no more than $15.00, and you should be able to get discounts for multiple units.

Unshielded twisted pair cable is rated according to specifications defined in the EIA/TIA (Electronic Industry Association/Telecommunications Industry Association) 568 standard in 1992. The standard calls for five categories of UTP cable, as shown in table 4-4.

Since the ratification of the standard, a company called Anixter, Inc., that was instrumental in the development of the cable ratings that later became the EIA/TIA categories has extended their original specifications. Anixter uses the term levels instead of categories, and has refined its level 5 specification and added two new levels, as shown in table 4-5.

The maximum frequency rating of the level 5 specification was doubled to bring it into compliance with the international standards document ISO 11801 and to bring the available data bandwidth measurement up to 1.2 GB/sec, making the cable suitable for Gigabit Ethernet applications. This specification has also been called enhanced Category 5, high-end Category 5 or Category 5+ by some manufacturers, although these names have not been ratified by the EIA/TIA.

When a revised Category 5 rating is ratified, it will probably be called Category 5E (for enhanced), but the new standard may not be equivalent to the level 6 rating. There is also a category 6 rating currently under review by the EIA/TIA that provides bandwidth that is double that of the existing Category 5 rating. Category 6, therefore, is not the same as Anixter's level 6 rating. A category 7 rating that provides bandwidth up to 750 MHz, in the early stages of development, also differs from Anixter's level 7.

The level 6 and 7 specifications exist to provide what Anixter refers to as headroom. The history of the computer industry is full of obsolete products and standards that no one at the time thought would ever be exceeded. When Fast Ethernet was introduced in 1995, increasing LAN performance tenfold, who would have believed that Ethernet speeds of 1000 Mbps would be possible within five years? Level 6 and 7 cables are intended to provide consumers with a cabling infrastructure that will remain useful throughout at least the next few technological innovations.

What this means to you when purchasing cables is that virtually all of the prefabricated UTP cables marketed today conform at least to the EIA/TIA Category 5 rating. However, some of these Category 5 cables are also rated to 350 MHz and conform to the Anixter Level 6 standard as well. Level 6 cables are usually more expensive than Category 5, but only slightly; you might have pay a few extra dollars on the average cable. While 100 MHz Category 5 cabling is rated as sufficient for 100BaseTX communications, for a few extra dollars you can be assured that your network is ready for whatever higher speed technology might come along in the next few years.

Once you have decided on what category or level of cable you intend to use, there are a few other properties you should consider in order to make an effective purchase. One of these is the length of your cables. Both the regular and Fast Ethernet specifications allow you to use cable segments up to 100 meters in length, but prefabricated cables that long are rare. The usual maximum length that you will find in stores, catalogs, and web sites is 50 feet, although a few vendors offer 100-foot lengths.

Sometimes the available lengths of a particular make of cable are dictated by its construction. For example, some manufacturers make both solid and stranded cables of the same rating. The actual conductive material is a solid cable is composed of a single strand of copper that provides excellent transmission properties. However, the single strand causes the cable to be relatively fragile; you should not uses solid cables in locations where they will be repeatedly flexed or twisted. The conductor in a stranded cable uses multiple copper strands that provide increased flexibility, but at the cost of additional signal loss due to attenuation. For this reason, stranded cables are typically available only in shorter lengths and should be limited to no longer than 20 feet.

The fire laws and building codes in your area may also influence the cable you use for your network. The bulk cable used for internal installations is usually available with either a standard PVC sheath or one that uses Teflon or some other substance that does not produce toxic gases when it burns, as PVC does. This type of cable is called plenum cable because it is suitable for use in building spaces called plenums that provide room air ventilation. If you plan on installing your network cable in the walls of an office building, and especially if you have to run cable between floors, you must be aware of the local regulations governing the products used in these spaces and purchase the right kind of cable. Plenum cable can cost up to twice as much as standard PVC cable, but you have no choice but to use it if the regulations require it.

Another feature that can cause price differences is the inclusion of a protective boot at each end of the cable. The lowest priced cables typically have no boot; the RJ-45 connectors are simply crimped onto the ends of the cable. As you move up the price scale, you'll find cables with molded plastic boots (sometimes called snagproof cables) that cover the body of the connector and attach to the cable itself. The boot prevents the squared edges on the connector from catching on walls and furniture as you are pulling the cable and protects the connector from being pulled or twisted off. The price difference for cables with molded boots is usually minimal; if your network will use exposed cables or if you ever need to move computers around after the installation, then snagproof cables are worth the expense.

The final, and least technical, factor you might want to consider is the color of your cables. This is particularly true for a home network. If you have white walls and light-colored carpets, you probably don't want a black cable snaking along the baseboard. UTP cables are available in a variety of colors, but you will almost never find much of a color selection in a store. A good catalog or online supplier should be able to provide any cable you want in at least five or six colors.

It may seem from the preceding sections as though purchasing the hardware needed to construct your network is an extremely complicated exercise. In practice, the process can actually be very simple indeed. Most of the larger computer stores these days stock at least a minimal selection of Ethernet hardware, and you can very likely walk in and pick up a few NICs, some prefabricated cables, and a hub that will work together perfectly well. You won't get the best prices and probably not the best possible performance from your network, but there is a good deal to be said for convenience.

This book is not about pushing the limits of networking technology; it is more concerned with providing users with basic networking services as easily as possible. However, it would be wrong to direct you towards the cheapest and most convenient networking solution when you might find it insufficient for your current or future needs. For that reason, my tendency is to recommend that you spend a bit more money to get quality equipment that will provide excellent performance for years to come.

For the basic network type, Fast Ethernet is the way to go. The difference in cost between standard Ethernet NICs and hubs and the Fast Ethernet variants is small enough to make any investment in 10 Mbps technology impractical. For network interface cards, you should choose dual-speed cards that use the PCI bus and 100BaseTX network connections. For the hub, select an unmanaged model with Class II repeating that supports 100BaseTX only (as dual-speed will not be needed). The hub should provide enough ports to support your current computers (and network print servers, if you intend to use them), plus two or three more for possible growth. For cables, you should go for the best quality you can find. Level 6 (or enhanced Category 5) cables with molded boots don't cost that much more than the bargain basement variety. They will also last longer and stand up to foot and desk traffic better.

As mentioned earlier, purchasing hubs and NICs from one of the major manufacturers is usually a good idea. These companies usually provide excellent documentation with which you can evaluate their products and stand behind them with better warranties than no-name companies that might disappear tomorrow. In addition, several of the large network hardware manufacturers have begun to take an interest in the home and small business networking market by releasing products intended specifically for these markets. For example, the 3Com Office Connect line of NICs and hubs are designed to work well together, and are even available in network-in-a-box bundles that include NICs, cables, and hub.

When deciding where you should make your purchases, consider that computer stores offer convenience, but not many keep more than a few networking products in stock. If you go to the trouble of evaluating the features of various NICs and hubs on the market, you should purchase the exact models you want, and this usually means going through a mail order firm or web site. As long as you protect yourself from unscrupulous vendors by making your purchases with a credit card, these methods are quite safe, usually offer a large selection at highly competitive prices, and may save you money on sales tax as well (depending on the location of the company).

Apart from the PCs themselves, NIC, cables, and a hub are all the hardware you need to build a local area network. However, there are other optional components that you might want to purchase, as well as network considerations when you are evaluating peripherals.

Any printer that works on a standalone system will also work on a network, but if you are purchasing a new printer, then you might want to consider some of the features that make them more suitable for network use. As a general rule, connecting a printer directly to the network cable is more convenient than connecting it to a workstation, and there are many printers on the market with built-in network interfaces that eliminate the need for a separate print server device.

For small networks, there are laser printers designed for use by workgroups that provide features that make printing easier for multiple users. Some of these features are as follows:

Speed

Duty Cycle

Paper trays

Toner Cartridges

Power Consumption

Expandability

Ink jet printers provide a low-cost alternative to laser printers. Ink jets cannot produce the excellent resolution of a laser, but they do often provide color printing capability. Few ink jet printers are designed for use on networks, but they can be shared just like any other printer with a Windows 98 driver.

External print server devices enable you to connect a printer directly to your network instead of to a computer. This prevents a computer from having to function as a print server, a function that can detract from its performance as a workstation. Hewlett Packard manufacturers a line of JetDirect print servers that are industry standard. Chapter 7, File and Printer Sharing, discusses the process of installing an external print server and sharing the printer with network users.

Some printers contain a slot that you can use to install a print server on an expansion card. For a printer without such a slot, you can use an external device that connects to the network hub with a UTP cable and to the printer with a standard parallel printer cable.

JetDirect print servers range from a simple one-printer device for 10BaseT networks to more advanced models that support multiple printers and other network types, such as Fast Ethernet. As with the NICs intended for PCs, Fast Ethernet print servers typically are dual-speed devices that enable you to upgrade your network from 10 to 100 Mbps Ethernet at any time.

In recent years, several manufacturers have released various types of hybrid peripheral devices marketed for the home and small office user. These machines typically combine printer, scanner, fax, and copier functions into one box, saving both money and desktop space. Unfortunately, however, sometimes the idea is better than the execution. Low-end devices if this type tend to combine a mediocre ink-jet printer, a low resolution scanner (suitable for faxing only), and a slow copier with a rather high cost per page. Some higher-end models improve on all of these functions, providing color scanning, printing, and copying as well.

In most cases, however, ink-jet printing does not produce sufficiently professional-looking output for business correspondence. Since there are many low-priced fax machines and scanners on the market, buying separate devices plus a laser printer might be a more satisfactory solution.

One other combination device worth mentioning is called a mopier (derived from MOP, or multiple original printing). A mopier is essentially a laser printer with a network interface and the print speed, paper handling, and collation features of a copier. Users can send a print job directly to the device over the network and have it product multiple original documents, just as if they printed the document and copied it using two separate machines. If you are planning to buy both a copier and a laser printer for your office, then a mopier could be a solution that saves money not only in the price of the hardware, but in user productivity as well.

Voltage spikes resulting from lightning storms, power failures, and faulty wiring can profoundly damage computers and other electronic devices, so it's a good idea to protect them with surge protectors. There are a great many different makes of surge protectors on the market, but a lot of them are little more than overpriced extension cords.

A proper surge protector uses metal-oxide varistors (MOVs) to shunt any voltages above 200v to the ground, and has a status light that indicates whether or not the unit is functioning properly. A surge of very high voltage can burn out the MOVs, and the only way to know if this has occurred is for the surge protector to let you know. A good surge protector also should conform to the UL1449 standard published by Underwriters Laboratories, which defines the properties that provide adequate protection to a PC.

Another good feature found on some surge protectors is a jack to protect a telephone line as well. A voltage surge can travel through a phone line and damage your modem or the computer itself.

You will usually have to spend a bit more to get quality surge protectors for your systems. They can cost $30 to $40 dollars, while the bargain basement models are often less than $20.

An uninterruptible power supplies (or UPS) is a device that uses a battery to maintain power to your computer in the event of a power failure. The UPS is not designed to provide an alternate power source for an extended period of time, but only long enough for the computer to be powered down in the proper way.

A UPS is usually not needed for a typical workstation, but they can be beneficial if you store vital data in a form that could be damaged by an uncontrolled system shutdown. If you have a computer that you use as a file or database server, a UPS can protect your data from damage.