Chapter 1: Designing and Preparing Your NT Server Computer

Abstract

This chapter shows you how to design your Windows NT server. You will walk through all the decisions required prior to installation of the operating system and you will complete the Windows NT Server Setup Worksheet.

---

 

Making critical decisions is tough. Making critical decisions quickly during software installation can be asking for trouble. You often end up making wrong choices because you don’t have all of the information you need and then you have to live with the consequences. Sometimes this means taking twice as much time by having to “do it right the second time.”

The best way to approach Windows NT Server 4.0 installation is to gather and understand all of the information you need up front. This approach allows you to make the important decisions before even beginning the installation procedure. You’ll avoid making mistakes that may require you to reinstall the software multiple times. Even if you’re like me and just launch the install program and fly by the seat of your pants, this preplanning process can spare you potential headaches when installing NT. You may grumble now, but you’ll thank me later.

In this chapter, I walk you through several aspects of designing, upgrading, and preparing your server computer to allow smooth installation and operation of Windows NT Server 4.0. In addition, I walk you through most of the important decisions you’ll be required to make before installing the operating system. At key milestones along the way, you’ll record those decisions on the worksheet at the end of this chapter. You’ll need this completed worksheet in Chapter 3 when you perform the hands-on installation of Windows NT Server on your computer.

Once again, I can’t emphasize enough the importance of this planning phase prior to Windows NT Server installation. Don’t skip it. The time, frustration, and job you save may be your own. In addition, the Windows NT Server Setup Worksheet that you complete in this chapter will be a vital addition to your network notebook.

DESIGNING YOUR SERVER HARDWARE

If you have the luxury and challenge of starting from scratch to specify the requirements of your server hardware, you’ll need to make a number of decisions before you start spending your company’s money. In this section, I offer advice on the NT tradeoffs associated with each hardware component of the server. I also recommend what works best with Windows NT Server in the real world.

If you’re inheriting existing server hardware and you’re upgrading it to run Windows NT Server, read on. The information presented here will be valuable in arriving at your hardware upgrade strategy.

Setting Server Hardware Goals

As with any major purchasing decision, it’s critical that you first understand and focus on the goals that you want to achieve with your Windows NT Server hardware. It also helps to prioritize those goals, in case you have to draw a line somewhere when you run out of hardware budget. Here’s a list of typical server hardware goals:

 

• Maximize performance
• Maximize reliability
• Maximize expandability
• Obtain fast, reliable support
• Minimize cost (overall, or per user)

These are all excellent objectives. You probably have others specific to your own organization and situation.

 

Hardware Compatibility

The Windows NT hardware compatibility testing program was devised to assure that NT would run on a wide variety of hardware configurations and to confirm the quality of third-party NT device drivers. Microsoft provides hardware manufacturers with a compatibility test kit that allows them to test their own hardware and drivers running on Windows NT. In addition, Microsoft performs a large chunk of hardware compatibility testing in their own labs. The result of this broad testing effort is a lengthy hardware compatibility list, offering you a wide range of hardware choices. You can find an up-to-date copy by visiting www.microsoft.com/BackOffice/ntserver/hcl on the World Wide Web.

The NT Hardware Compatibility List isn’t a comprehensive rundown of all hardware products that work with Windows NT. It’s simply the list of all computers and devices that have been formally tested with NT. Some manufacturers choose not to test their products formally for compatibility with NT, but they work great with it, nonetheless. (Microsoft makes no distinction between NT Server and NT Workstation, when it comes to hardware compatibility.)

For example, the clone computer I’m working on right now has never been on the NT compatibility list, but I’ve been successfully running various versions of NT on it since 1991. Likewise, the network adapter in this computer isn’t on the compatibility list, but it’s fully compatible with one that’s listed. It operates without a hitch, using the network adapter driver that Microsoft tested with a different, but fully compatible, adapter.

Selecting Your Processor

In this section, I discuss the factors to consider when deciding which processor to use with Windows NT Server 4.0. This choice will have an effect on several other downstream decisions that you make about your server hardware.

Choosing Your CPU Architecture

Windows NT Server will run on Intel x86, DEC Alpha, MIPS R4x00, and PowerPC platforms. Performance benchmark tests show many of these systems to be great performers, and each platform offers a wide range of processor speeds and prices. In terms of reliability, the CPU architectures are pretty much equal.

Performance

Whether RISC processors will buy you much performance on a network server is debatable and highly dependent on your use of the server. If your server is only going to share files and printers (in addition to NT Server’s logon validation responsibilities), the differences in performance between platforms is small. High-end Intel CPUs will do just as well as RISC processors.

Will you be running server applications, such as SQL server, SMS, SNA, or Microsoft Exchange server? If your server will be occupied with CPU-intensive activities by running the server side of a client/server application, some of the faster RISC processors can make a difference in the performance of your server. However, since each manufacturer (including Intel) is continuously releasing new versions of its CPUs, the landscape will vary from month to month. Different CPU platforms have been leapfrogging each other to gain the performance lead.

Sitting on the Benchmark

I’ve purposely steered clear of providing you with actual CPU performance benchmark results for several reasons. First, by the time you read this, a whole new set of CPU products will be available from most of these manufacturers. Second, each manufacturer tends to quote different industry benchmarks for their processors, making meaningful comparisons between architectures difficult. Some even use different benchmarks to rate processors within the same product line. Finally, different manufacturers’ benchmark tests are performed on a wide variety of hardware configurations, making the results even more difficult to interpret and compare. Your best bet is to look for the latest independent test results in PC periodicals. These tests are typically performed on virtually identical hardware configurations, using standard performance benchmarks in a controlled environment. Although benchmarks cannot fully emulate the real world of your enterprise network, they do provide a basis for comparison.

My advice is to educate yourself on the latest CPU performance benchmark results. This knowledge will let you determine which processor is best for your particular requirements. Be sure to focus on benchmarks that best match how you intend to use your server. For example, if your NT computer is going to act as an application server, don’t focus on interactive desktop-application benchmarks.

Software Compatibility

Many Win32 software developers are focusing primarily on the Intel platform, even though support of RISC platforms is just a recompile and testing cycle away. You may find that the important NT Server applications that you want to run are available only on Intel, and perhaps on a subset of the other NT-supported CPU platforms. Almost every vendor provides at least an Intel version of their Win32 applications, so this may be a factor in your processor decision.

If you’re going to run 16-bit x86 applications regularly on your NT Server, you’re better off with an Intel processor. Although RISC-based platforms will run most DOS and Win16 applications, they do so through software emulation. An Intel-based computer will execute these applications faster than RISC-based computers. Consider obtaining and running 32-bit versions of your applications, if they’re available. Running 16-bit applications on your NT Server isn’t the best use of its CPU power.

Platform Support

You can safely bet that Windows NT will support Intel x86-compatible processors far into the future. The worst thing that will happen in the Intel arena is dropped support for older CPUs, as the processing requirements of the operating system increase and the popularity of older CPUs decreases. You’ve already seen this happen with Microsoft’s abandonment of the 386 in NT 4.0. Someday, when the child or grandchild of today’s Pentium Pro becomes the standard desktop CPU, you may see diminished support for the 486 processor (and increased sales of Pentium OverDrive upgrades to breathe a few more months of life into these computers).

You can also bet that Windows NT will support other RISC processors in the future, as they prove themselves NT-worthy, both technologically and economically. Even though NT is portable, adding support for a new processor architecture represents a huge investment for both Microsoft and the CPU manufacturer. There’s the initial investment of porting and testing NT on the new platform, requiring dozens of people and scores of computers based on the new CPU. Then there’s the ongoing commitment to building and testing all future versions of the operating system and flavors of computers based on the new CPU architecture. So, the decision to move NT to a new platform isn’t taken lightly at Microsoft. Likewise, if a platform just refuses to succeed in the marketplace, there’s always the possibility that NT support might be dropped for that platform.

The Alpha CPU architecture appears to be around for the long haul. DEC is producing a continuous stream of increasingly faster Alpha CPUs and systems, including SMP computers. Several firms offer MIPS and PowerPC-based Windows NT computers, but neither platform has really taken off like gangbusters yet. Although these computers running UNIX or Apple operating systems (in the case of PowerPC) seem to be plentiful today, systems running NT are slightly harder to come by. If you’re taking the road to RISC, my recommendation is that you choose your hardware vendors carefully and include quality and timeliness of service and support in your purchasing decision.

In the following sections, I provide some additional information on each processor architecture, to aid in your CPU selection.

Intel Processor Roundup

Intel offers a wide range of x86-compatible CPUs, with more arriving every few months. Table 1-1 paints the current Intel processor landscape. I’ve focused on only those CPUs that Windows NT Server 4.0 supports.

TABLE 1-1   THE NT-COMPATIBLE INTEL PROCESSORS
Intel CPU Family	Internal Processor Speed	System Bus Speed	Level 1 Cache Size	Level 2 Cache on Chip	System Bus Width
i486 DX	33MHz	33MHz	8K	none	32 bits
i486 DX	50MHz	50MHz	8K	none	32 bits
IntelDX2 (486)	50MHz	25MHz	8K	none	32 bits
IntelDX2 (486)	66MHz	33MHz	8K	none	32 bits
IntelDX4 (486)	75MHz	25MHz	16K	none	32 bits
IntelDX4 (486)	100MHz	33MHz	16K	none	32 bits
Pentium	60MHz	60MHz	8K(I)+8K(D)	none	64 bits
Pentium	66MHz	66MHz	8K(I)+8K(D)	none	64 bits
Pentium	75MHz	50MHz	8K(I)+8K(D)	none	64 bits
Pentium	90MHz	60MHz	8K(I)+8K(D)	none	64 bits
Pentium	100MHz	66MHz	8K(I)+8K(D)	none	64 bits
Pentium	120MHz	60MHz	8K(I)+8K(D)	none	64 bits
Pentium	133MHz	66MHz	8K(I)+8K(D)	none	64 bits
Pentium	150MHz	60MHz	8K(I)+8K(D)	none	64 bits
Pentium	166MHz	66MHz	8K(I)+8K(D)	none	64 bits
Pentium	200MHz	66MHz	8K(I)+8K(D)	none	64 bits
Pentium Pro	150MHz	66MHz	8K(I)+8K(D)	256K	64 bits
Pentium Pro	166MHz	66MHz	8K(I)+8K(D)	512K	64 bits
Pentium Pro	180MHz	66MHz	8K(I)+8K(D)	256K	64 bits
Pentium Pro	200MHz	66MHz	8K(I)+8K(D)	256K	64 bits
Pentium Pro	200MHz	66MHz	8K(I)+8K(D)	512K	64 bits

I’ve listed both the internal CPU speed (the rate at which instructions are executed) and the system bus speed (the rate at which memory and peripherals are accessed). Both are important to the overall performance of the computer. Where they differ for a given processor, Intel has employed a clock multiplier to run the CPU internally at a higher rate than the rest of the computer. This approach improves performance significantly for CPU-bound applications (and for Windows NT itself), but the improvement becomes negligible if your application is memory- or I/O-intensive.

In Table 1-1, (I) refers to the instruction cache size, and (D) refers to the data cache. Both types of cache are discussed later in this section.

The 486 CPU

The minimum Intel processor for Windows NT 4.0 is a 486. Microsoft recommends a minimum of a 33MHz 486 CPU for Windows NT Server. Frankly, the 486/33 crawls when running NT, so I recommend at least a 486/66 (DX2) or 486/100 (DX4).

Ye Olde Intel CPUs

The Intel 386 is dead, at least from Windows NT’s perspective. Don’t even attempt to install Windows NT Server 4.0 on a 386-based computer. Setup won’t let you do it. Older revisions of Intel processors have sometimes been the source of headaches under Windows NT. Some of the early 386 CPUs contained problems that were fixed in later revisions of the chip. The 386 B1-step revision wasn’t supported at all by Windows NT. Now that the 386 isn’t supported, this is no longer an issue. The 486 C-step revision has a bug that affects Windows NT, but Setup recognizes this chip and installs a special HAL (hal486c.dll) to work around it. There’s a slight performance cost when running in this configuration, but the tradeoff is reliable operation.

The 486 DX CPU is essentially a 386 processor combined with a 387 math coprocessor on a single chip. In addition, the 486 adds an on-chip (primary) cache, which significantly speeds memory access in many situations. The cache stores data and instructions used recently, as well as nearby data and instructions that are likely to be needed soon. Having this information in the cache minimizes the time that the 486 has to wait for slower memory-access cycles. Most, but not all, 486 motherboards also provide a larger external (secondary) cache, which increases the chances that the processor won’t have to access slow memory.

Note: Avoid the 486 SX, which is a 486 DX minus the math coprocessor. Since NT uses the floating point unit, and since your applications may also make use of it, you’re better off starting with the DX version. If you start with an SX and decide later to add a coprocessor, you’ll spend more in the long run.

Relative to the Pentium and Pentium Pro processors discussed in the next sections, the 486 is slow. We recommend that you strongly consider a Pentium-based server, if you have a choice. The cost difference between 486 and Pentium computers is getting pretty small, relative to the significant increases in performance of the Pentium. In fact, 486 computers are getting harder to find everyday.

The Pentium CPU

The PC industry expected the successor to the 486 to be called the 586, but when Intel discovered that it couldn’t copyright numeric names for its processors, it dubbed the new product the Pentium. As Table 1-1 shows, the Pentium introduces a 64-bit data bus to the Intel processor line. The Pentium can execute up to two instructions per clock cycle, using two instruction pipelines (a feature traditionally found in RISC processors).

As shown by the (I) and (D) notations in Table 1-1, the Pentium splits its primary cache into two equal-size sections, allocated separately for instructions and data. Like the 486, most Pentium motherboards provide an external secondary cache to improve performance.

Ye Olde Pentiums

Can there be such a thing as an old Pentium? Shy away from the 60MHz and 66MHz Pentium processors, even at bargain basement prices. Because these CPUs used a higher voltage than more recent types, they had significant heat dissipation problems. Some computer manufacturers worked around the heat problem successfully; others didn’t. The result is that some computers based on these chips behave erratically. This characteristic certainly won’t meet your server reliability goal. Early versions of the Pentium processor included, at no extra cost, the infamous Floating Point Division Error. This bug in the Pentium chip affects floating point division, remainders, and transcendental functions such as sine, cosine, and meditation (for those metaphysical calculations). If you rely on any of these functions, your results may be affected by this bug. Windows NT provides a command line utility called PENTNT. (Type PENTNT /? at an NT command prompt for details.) It allows you to detect whether you have a Pentium CPU with the floating point problem. In addition, it provides you with the ability to bypass the chip’s hardware floating point calculations and use a software emulator instead. As you might expect, software emulation is significantly slower. Since NT itself uses some floating point functions, taking the software emulation road can slow down not only your applications but the operating system as well. A better approach to solving this problem is to contact your computer vendor or Intel for a replacement Pentium CPU (which they should give you at no cost). Use software emulation as a stop-gap measure only.

Pentiums are very powerful and much faster than 486 CPUs. Always opt for the highest motherboard bus speed available, currently 66MHz. This will minimize the time that your CPU will wait for memory accesses and will allow you to use the fast 133, 166, and 200MHz Pentium processors. Also, try to get the largest external cache possible on the Pentium motherboard.

The Pentium Pro CPU

With the Pentium Pro CPU, Intel has gone out on a limb, at least from the perspective of many industry pundits. Back in 1991, when Intel began designing the Pro, they thought that we’d be living in a completely 32-bit world by now. The Pentium Pro is optimized to run 32-bit operating systems and applications. When fed 16-bit software, it actually performs more slowly than regular Pentium processors. Thus, Pentium Pro computers make lousy Windows 3.x and Windows 95 systems (since both contain lots of 16-bit code), but they make ideal Windows NT computers. At least, as long as you’re running 32-bit applications.

The Pentium Pro can execute up to three instructions per clock cycle, as opposed to the Pentium’s two per cycle. Unlike the Pentium, the Pro can actually reorder instructions, executing those that are ready while putting others on hold.

Like the Pentium, the Pro splits the cache into two 8K chunks, one for instructions and one for data. In addition, the Pentium Pro CPUs are the first Intel processors to include an on-chip secondary cache. If you look at Intel’s standard picture of the Pro, you’ll see two separate silicon chips in the large CPU package. One of those chips is the secondary cache.

Tip: Although the external data bus of the Pentium Pro is still 64 bits, some vendors are using DIMMs, or dual in-line memory modules, instead of SIMMs to achieve an effective data path that’s 128 bits wide. The HP Vectra XU 6/150 was one of the first Pro-based computers to use this technique. Computers using this approach are blowing away the competition in performance benchmark tests. Look for this feature if you’re on the market for screaming performance. Of course, it will cost you more. In this case, though, you do get what you pay for.

The Pentium Pro is a very new kid on the Intel block. At the time of this writing, nearly all existing Pro computers use the 150MHz chip, and 200MHz systems are just now starting to trickle into the market. One of Intel’s Pentium Pro data sheets still says the CPU “may contain design defects or errors.” We haven’t yet seen what bugs in this processor might turn up after several months of use in the real world.

Intel Processor Upgrades

If all you’re after is higher CPU speed, Intel offers several processor replacements that you can use to upgrade existing hardware today or future hardware tomorrow. If you have the appropriate socket available to accept the upgraded processor (Intel calls them OverDrive processors), you can increase the internal speed of your CPU with only minor surgery. Table 1-2 outlines the available upgrade paths using Intel products.

TABLE 1-2   INTEL PROCESSOR UPGRADE PATHS
From Your Existing CPU	Running at	Upgrade Your CPU to	Running at	System Bus Speed Remains	Socket Type Required
i486 DX	25MHz	IntelDX2	50MHz	25MHz	1, 2, or 3
i486 DX	33MHz	IntelDX2	66MHz	33MHz	1, 2, or 3
i486 DX	25MHz	IntelDX4	75MHz	25MHz	1, 2, or 3
i486 DX	33MHz	IntelDX4	100MHz	33MHz	1, 2, or 3
i486 DX	66MHz	IntelDX4	100MHz	33MHz	1, 2, or 3
i486 DX	25MHz	Pentium	63MHz	25MHz	2 or 3
IntelDX2	50MHz	Pentium	63MHz	25MHz	2 or 3
i486 DX	33MHz	Pentium	83MHz	33MHz	2 or 3
IntelDX2	66MHz	Pentium	83MHz	33MHz	2 or 3
Pentium	60MHz	Pentium	120MHz	60MHz	4
Pentium	75MHz	Pentium	125MHz	50MHz	5 or 7
Pentium	66MHz	Pentium	133MHz	66MHz	4
Pentium	90MHz	Pentium	150MHz	60MHz	5 or 7
Pentium	100MHz	Pentium	166MHz	66MHz	5 or 7
Pentium	120MHz	Pentium	180MHz	60MHz	7
Pentium	150MHz	Pentium	180MHz	60MHz	7
Pentium	133MHz	Pentium	200MHz	66MHz	7
Pentium	166MHz	Pentium	200MHz	66MHz	7

Socket requirements are very strict. Each new processor technology has a different number and layout of pins, so your motherboard has to be equipped with the appropriate upgrade socket to handle the new processor. Check your computer documentation to determine what type of socket you have before you invest in one of these upgrades.

Caution: If you’re thinking of upgrading a multiprocessor computer using an OverDrive processor, think again. Intel warns that Pentium OverDrive processors won’t work properly in computers with more than one processor.

Intel currently has no plans to offer any OverDrive processor upgrades that will allow you to move from a Pentium to a Pentium Pro. Your only upgrade path from a Pentium is to a faster Pentium. Otherwise, you’ll need a new motherboard to jump into the Pentium Pro world.

Several of Intel’s competitors, such as AMD and Cyrix, have introduced x86-compatible processors that offer increased performance and, in some cases, lower cost. Some of these CPUs are designed as upgrades to existing computers; others are provided as standard equipment in off-the-shelf servers. I’ve seen many installations achieve excellent results with these third-party devices.

Caution: The NT software doesn’t recognize some third-party upgrade processors as Intel counterparts. Make sure that your processor is recognized as at least a 486 by Windows NT, or you won’t be able to install NT on it. Check the NT Hardware Compatibility List and the CPU manufacturer for compatibility information. Be sure to specify Windows NT 4.0 when you ask, since it won’t install on anything that looks like less than a 486.

In general, we suggest steering clear of 386-to-486 CPU upgrades for NT computers. Although the processor itself runs dramatically faster than the 386 it replaces, the 386 computer likely has a slow bus speed, little or no external cache, and sometimes a BIOS that can’t handle the higher-speed processor. Use these devices only as stop-gap measures until your new server hardware arrives.

MIPS PROCESSOR ROUNDUP

Windows NT was ported to the MIPS CPU architecture just after the port to the Intel x86 was complete. A number of MIPS processors have emerged since then, and several others are in the works. Prior to NT, MIPS’s big claim to fame has been its use as the core of powerful Silicon Graphics workstations. MIPS had a very strong start in the NT arena, but some of its recent offerings have been more expensive than and not as fast as Pentium Pros.

Table 1-3 shows the MIPS processor landscape as it stands at the time of this writing. A few of the faster processors are just about to be released. Like the Pentium and Pentium Pro, these processors have a 64-bit data path. System bus speeds vary up to 100MHz for some processors. Like the Pentium Pro, all MIPS processors include separate data and instruction on-chip primary caches, which are indicated in the table by (D) and (I). In most cases, the size of the primary cache is larger than that found on Intel’s CPU offerings.

TABLE 1-3   THE MIPS PROCESSOR LINEUP
MIPS RISC CPU	Internal Processor Speed	Level 1 Cache Size	Level 2 Cache Controller
R4000	100MHz	8K(I)+8K(D)	external
R4200	80MHz	16K(I)+8K(D)	external
R4400	200MHz	16K(I)+16K(D)	up to 4MB
R4400	250MHz	16K(I)+16K(D)	up to 4MB
R4400	150MHz	16K(I)+16K(D)	up to 4MB
R4600	133MHz	16K(I)+16K(D)	up to 4MB
R4700	133MHz	16K(I)+16K(D)	up to 4MB
R4700	175MHz	16K(I)+16K(D)	up to 4MB
R5000	180MHz	32K(I)+32K(D)	up to 2MB
R8000	75MHz	16K(I)+16K(D)	up to 16MB
R8000	90MHz	16K(I)+16K(D)	up to 16MB
R10000	200MHz	32K(I)+32K(D)	up to 16MB

You may notice that MIPS processor clock speeds and cache sizes tend to be higher than those of Intel. This phenomena is due, in part, to RISC’s requirement that the CPU execute more instructions to accomplish the same work. Also, because of the uniform size of RISC instructions, they tend to take up more space in memory than their Intel counterparts.

Note: You must have a MIPS R4000 version 2.0 or later in order to run Windows NT Server. Check with your computer manufacturer to ensure that you have the appropriate version, and make sure to indicate that you’re planning to run NT on it.

All but the R4000 and R4200 include an on-chip secondary cache controller. This isn’t the same as having an on-chip secondary cache such as the Pentium Pro. The MIPS processors have the cache controller built into the chip, but the secondary cache memory still resides outside the chip, on the motherboard.

The MIPS R4x00 family includes PC, SC, and MC versions. The PC flavor is designed for low-cost desktop computers and supports no secondary cache. I recommend staying away from this version, for performance reasons. The SC is designed for high-performance, single-processor computers with lots of secondary cache on the motherboard, and MC is designed for multiprocessor computers with large secondary caches. (Because RISC instructions take more memory, the size of the secondary cache becomes increasingly important on MIPS and other RISC processors.) The SC and MC MIPS flavors, coupled with large secondary caches, are good choices for NT servers.

Depending on the processor model, MIPS CPUs can execute either three or four instructions per cycle. R10000 can execute up to four instructions per cycle. Like the Pentium Pro, it can shuffle the order of instruction execution.

The MIPS CPU is available from a handful of semiconductor vendors. MIPS-based computers that can run Windows NT Server are offered by a dozen or so system vendors including NEC, NeTpower, Siemens-Nixdorf, and Silicon Graphics. A few vendors recently announced that they’re pulling out of the MIPS market in favor of manufacturing Intel SMP computers.

Alpha Processor Roundup

Once the MIPS port was completed, a team at DEC started working with Microsoft on a port of Windows NT to DEC’s Alpha processor architecture. Alpha CPUs have become quite popular, especially in the high-end workstation market. DEC has successfully released an increasingly powerful series of products, including multiprocessor server computers. Of all the RISC platforms supported by NT, most application developers work on porting to the Alpha as soon as they complete their Intel version.

Table 1-4 shows the Alpha processor lineup as it stands at the time of this writing. Alpha CPUs can execute up to four instructions per clock cycle. They have a 64-bit data path, although some can optionally support a 128-bit bus. Like the MIPS and Pentium Pro, all Alpha processors include separate data and instruction on-chip primary caches, shown in Table 1-4 by (D) and (I). The clock rates are phenomenal, but keep in mind that, as a RISC processor, the Alpha must execute more instructions to accomplish the same work.

TABLE 1-4   THE ALPHA PROCESSOR LINE
Alpha AXP CPU	Internal Processor Speed	Level 1 Cache Size	Level 2 Cache Controller
21064A	200MHz	16K(I)+16K(D)	up to 16MB
21064A	233MHz	16K(I)+16K(D)	up to 16MB
21064A	275MHz	16K(I)+16K(D)	up to 16MB
21066A	100MHz	8K(I)+8K(D)	up to 16MB
21066A	166MHz	8K(I)+8K(D)	up to 16MB
21066A	233MHz	8K(I)+8K(D)	up to 16MB
21164	266MHz	8K(I)+8K(D)	96K on chip
21164	300MHz	8K(I)+8K(D)	96K on chip
21164	333MHz	8K(I)+8K(D)	96K on chip
21164	366MHz	8K(I)+8K(D)	96K on chip
21164	400MHz	8K(I)+8K(D)	96K on chip

Like most MIPS CPUs, all Alpha processors include an on-chip secondary cache controller. The secondary cache memory either resides outside the chip, on the motherboard, or on-chip in the case of the 21164 Alpha processor.

PowerPC Processor Roundup

The PowerPC is the youngest of the NT-compatible CPU siblings. Official support for the PowerPC first emerged with Windows NT 3.51. Microsoft worked closely with both Motorola and IBM to make this happen. (Yes, cooperative efforts between Microsoft and IBM are still possible.) Except in the Apple arena, the PowerPC hasn’t really taken off, primarily because it hasn’t offered spectacular price/performance advantages over the other NT platforms. Table 1-5 presents the PowerPC processors available at the time of this writing.

TABLE 1-5   THE POWERPC PROCESSOR WORLD
PowerPC CPU	Internal Processor Speed	Level 1 Cache Size	Level 2 Cache Controller
601	66MHz	32K	external
603	66MHz	8K(I)+8K(D)	external
603	80MHz	8K(I)+8K(D)	external
603e	100MHz	16K(I)+16K(D)	external
603e	120MHz	16K(I)+16K(D)	external
603e	133MHz	16K(I)+16K(D)	external
603p	150MHz-200MHz	16K(I)+16K(D)	external
604	100MHz-200MHz	16K(I)+16K(D)	external
604e	150MHz-200MHz	16K(I)+16K(D)	external
620	133MHz	32K(I)+32K(D)	up to 128MB

The 601 and 603 series can execute up to three instructions per clock cycle. These chips can use a 32-bit or a 64-bit data bus. The 604 series can execute up to four instructions per clock cycle and have a 64-bit data bus. The 620 introduces an on-chip cache controller to handle up to 128MB of external secondary cache. In addition, the 620 increases the data bus width to 128 bits.

How Many Processors Do You Need?

Adding CPUs to multiprocessor computers is a great way to increase your server capacity. Contrary to what you might think, though, adding a second processor doesn’t double your throughput. Coordination among CPUs requires some additional overhead. Results of increasing the number of processors vary from one model to the next. Some Intel computers show an 80 percent throughput increase with the installation of a second processor, with additional CPUs providing less incremental performance improvement. If your server is CPU-bound and is running multiple applications or multithreaded applications, you should consider selecting an SMP computer.

By the time you read this, SMP computer manufacturers should be using a new version of the Pentium Pro processor (called A1 stepping) that solves the so-called Wob Clobber bug. This bug delayed shipment of several four-processor Pentium Pro computers. Also, Intel is now building four-processor Pentium Pro motherboards for use by other computer manufacturers. This should yield lower-cost, more standardized SMP computers. Multiprocessor computers based on MIPS, Alpha, and PowerPC CPUs are available today, with more on the way. NT’s Hardware Compatibility List provides examples of SMP computers based on each CPU platform.

The price of SMP computers is higher, even if you initially install only one CPU. However, the overall cost is typically much less expensive than purchasing separate server computers to achieve the same performance increase. Some of these products include redundant power supplies, hardware RAID drives, hot-swappable disk drives, and other components useful in a mission-critical environment. These features tend to increase their cost as well.

Selecting Your Motherboard

In this section, I discuss the issues to consider in choosing or upgrading your server’s motherboard. Your decisions will drive and, in some ways, limit the choices that you make later.

Which Bus Should You Ride?

Several different PC bus architectures have been developed since the introduction of the first IBM PC in 1983. Newer technologies typically offer increased throughput. Some motherboards today combine these technologies to provide a cost-effective approach to building high-speed servers. In this section, I briefly discuss each of these bus technologies.

The ISA Bus

The standard 8-bit PC adapter slot provided in the original IBM PC, later expanded to 16 bits in the IBM PC/AT, has become known as the ISA (Industry Standard Architecture) bus. It’s been widely adopted across the industry in PCs, so there are thousands of different adapters designed for it.

Relative to other PC bus architectures available today, the ISA is slow because of its limited speed and narrow data path. My recommendation is to steer clear of computers that have only ISA slots. Transferring data at an effective rate between 1.5MB and 5MB per second, your disk, network, and video performance will be abysmal in these machines. ISA is adequate for some client computers, but it definitely doesn’t cut the mustard for servers, unless your network is very small.

The VESA Local Bus

The VESA Local Bus, or VLB, was developed as a high-speed supplement to the ISA bus. This type of bus was designed to give video adapters direct access to the CPU. Up to two 32-bit VLB slots are included on an otherwise ISA motherboard to provide data transfer rates up to a speedy 132MB per second. VLB is typically designed into 486-based motherboards, because of the difficulty of adapting it to the Pentium processor due to its higher speed and 64-bit data bus requirement.

VLB was developed originally for video acceleration. Although nonvideo adapters have been developed, there are few of them. Since a maximum of two VLB slots are available on an otherwise slow ISA motherboard, high-speed expansion options are limited. I don’t recommend using a VLB/ISA motherboard in your server, unless your network is very small (say, ten nodes maximum) with low traffic.

Note: A successor to VLB, called VESA II, has appeared in an attempt to provide a 64-bit solution for Pentium-based computers. However, it has gained little support in the industry because of the overwhelming popularity of PCI, discussed later in this chapter. I recommend avoiding VESA II motherboards, since they have very limited adapter choices and an uncertain future.

The MCA Bus

IBM introduced the Micro Channel Architecture (MCA) with the release of their PS/2 computer. It was a completely new bus design, allowing both 16-bit and 32-bit data paths to transfer information up to 40MB per second. In addition, MCA adapters have the ability to take control of the bus and perform their own data transfers without any CPU intervention. This technique, known as bus mastering, relieves the processor of significant I/O burdens. MCA is completely incompatible with existing ISA adapters. You can only put MCA adapters in an MCA computer.

Because of its respectable speed and bus mastering capability, MCA can be suitable as the basis for a server computer. Keep in mind, though, that choices of adapters are much more limited than for other architectures, since relatively few hardware vendors have jumped on the MCA bandwagon. (IBM decided to make the MCA design proprietary, and few manufacturers decided to pay IBM the requisite licensing fees to use it.) Also, many existing MCA motherboards have slow clock speeds and little or no cache. So, keep these factors in mind before deciding to run with MCA.

From an administrative viewpoint, MCA is a bit more difficult to configure and maintain, since it requires running a special configuration program for each adapter that you install. You must keep track of configuration floppy disks for each of your MCA computers and devices.

The EISA Bus

Since most hardware manufacturers were unwilling to pay IBM for the rights to design MCA computers and adapters, several of them got together and defined the Extended Industry Standard Architecture, called EISA. Like MCA, this bus provides a 32-bit data path and bus mastering and requires running special configuration programs for each adapter. Unlike MCA, EISA offers full compatibility with ISA adapters and can even coexist with them in the same computer. Unfortunately, by maintaining ISA compatibility, EISA’s speed is constrained to 32MB per second.

EISA is certainly preferable to ISA in terms of throughput, and the breadth of adapter choices is orders of magnitude greater than MCA. EISA’s ability to support ISA adapters allows you to migrate from older, slower adapters to newer, faster ones as your network grows. If you have a choice between EISA and ISA, go with EISA. However, for high disk and network throughput in your server, look for PCI capability, which I discuss next.

The PCI Bus

At the time of this writing, Intel’s Peripheral Computer Interface, or PCI, is the state-of-the-art in high-speed PC buses. It’s designed with the Pentium in mind, providing a 64-bit data path. Like EISA, it supports bus mastering. However, PCI can operate at the speed of the system bus up to 66MHz, yielding a theoretical data transfer rate of up to 264MBps. In practice, today’s PCI bus designs (version 2.0 and 2.1) peak at 132MBps. Several manufacturers are starting to support a faster design based on the PCI 3.0 specification, but you’ll need both new motherboards and new PCI adapters to achieve the advertised speed of 264MBps. (Older PCI adapters will work in the new PCI 3.0 motherboards, but they’ll run at the slower rate.)

Most available PCI motherboards contain three PCI slots, with the remainder being either ISA or EISA slots. A few motherboards are now emerging with four and six PCI slots, which makes them more relevant products for servers. Because of the high speed and relatively low cost (due to availability of a standard PCI interface chip set), a wide range of adapter choices are already available, with more appearing daily.

Caution: Choose established motherboard and adapter manufacturers for your PCI products and avoid older revisions of PCI equipment. It has taken time for the dust to settle on this standard. Many flawed hardware designs and driver bugs have wreaked compatibility havoc in the PCI world. Focus on the NT Hardware Compatibility List, and contact each adapter vendor to determine compatibility among their adapters, your motherboard, and NT.

For Windows NT Server, I recommend a PCI/EISA motherboard with as many PCI slots as possible. PCI/ISA motherboards are a second choice. You’ll want to install disk adapters and network adapters in the PCI slots. If you plan on disk duplexing and network bridging, your need for PCI slots will quickly grow.

How Much Cache Do You Need?

The rule of thumb here is Bigger Is Better. The more primary and secondary cache that you have, the better your server’s performance will be. Get a processor with as much internal cache as possible, and never choose a motherboard that doesn’t provide room for an ample secondary cache.

Worksheet Entry

In Worksheet 1-3 (Windows NT Server Setup) on line 6, record the make and model of your computer, along with the type of bus that the computer has (ISA, EISA, MCA, or PCI).

WORKSHEET 1-3   WINDOWS NT SERVER SETUP
Decision #	Your Decision
1	Circle your CD-ROM drive type.	SCSI, IDE/EIDE, or OTHER
2	If your CD-ROM drives are SCSI, which one has the lowest SCSI ID number (make/model)?
3	List your SCSI adapters, non-SCSI CD-ROMs, and disk arrays.
4	Do you have storage devices not compatible with have NT (that require an NT driver disk)?	YES or NO. If YES, the driver disk available.
5	Are you upgrading from an older version of Windows NT Server?	YES or NO
6	Enter your computer type (make, model, bus type).
7	Display type (for Setup only).	Auto Detect
8	Enter your keyboard make and model.
9	Enter your keyboard layout.
10	Enter your pointing device (mouse) make and model.
11	Are you upgrading from Windows 3.x, Windows for Workgroups, or Windows 95?	YES or NO
12	List your plans to create/delete disk partitions. On which partition will you install NT system files?
13	Which file system do you want on the partition where the NT operating system files reside?	FAT or NTFS
14	Under what directory name do you want the NT system files installed?
15	Enter your name.
16	Enter your organization name (optional).
17	Circle the licensing option that you want to use.	Per Server or Per Seat
18	Enter the computer name for this server.
19	Circle the role of this server on the network.	PDC, BDC, Stand-Alone Server
20	Have your domain administrator password available.	For security reasons, don’t write the password here.
21	Choose to install optional components, unless you already have a non-Microsoft e-mail client.	YES or NO
22	Choose to install the Microsoft Exchange Client if you have or plan to install Exchange Server.	YES or NO
23	List your network adapters. See Chapter 2 for details.
24	Circle the network protocols that you want to install on this server. See Chapter 5 for details on installing TCP/IP.	NetBEUI, IPX/SPX, TCP/IP
25	Do you want to install RAS? See Chapter 5 for details.	YES or NO
26	Enter your network adapter configuration information. See Chapter 2 for details.
27	Enter the domain name.
28	Enter your time zone.
29	Enter your video adapter make and model.
30	Enter your video adapter chip set information.
31	Enter the number of screen colors.
32	Enter your desired video resolution.

Caution: Many vendors advertise internal CPU cache to make it sound like additional external cache. Don’t be fooled. Make sure that you know both how much external cache is actually supplied on the motherboard and how much total cache the motherboard design will accept. Today Pentium motherboards typically accept a maximum of 512K, and a handful allow 1MB of cache.

There are many opinions about how much secondary cache is enough. Although there’s no pat answer for all computer configurations, Table 1-6 provides some guidelines for Windows NT computers.

TABLE 1-6   GUIDELINES FOR SECONDARY CACHE SIZES
System RAM Size	Recommended Secondary Cache Size
<16MB	Doesn’t apply to NT
16MB	256K
24MB	256K
32MB	512K
64MB	512K
>64MB	512K–1MB

Because installation of secondary cache memory can be tricky, I recommend that you opt for as much external cache as you can afford up front. Today, most motherboards designed to be used as servers ship with 256KB or 512KB of external cache installed. Compaq was the first to offer a 1MB external cache in its server product line. (This is one of the perks of designing your own motherboards.)

Selecting Your Memory

The single most significant way to improve performance on a Windows NT server computer is to add more RAM. The more RAM you have, the more NT can keep cached in memory. Windows NT takes full advantage of all memory available to it. The more it has, the happier it is.

How Much Memory Do You Need?

You can listen to Microsoft, who recommends a minimum of 12MB for Intel x86 computers and 16MB for RISC computers. You can listen to me when I tell you that the more RAM you have, the better off you are. But how much memory do you really need in your Windows NT server computer? Worksheet 1-1 will help you get to a ballpark figure. You can use this worksheet for estimating the memory for each of your NT Server installations. I’ve included RAM requirements for Microsoft’s BackOffice applications, and you can add the memory requirements of other server applications that you plan to run.

WORKSHEET 1-1   WINDOWS NT SERVER MEMORY ESTIMATE
Line	Memory Requirement	Minimum	Recommended	Estimate
1	Number of concurrent users of this server
2	Is line 1 under 25?	16MB	16–32MB
3	Is line 1 between 25 and 250?	32MB	32–64MB
4	Is line 1 over 250?	64MB	64–128MB
5	Average size of open data files on the server, per user (in megabytes)
6	Multiply line 1 by line 5
7	Number of concurrent applications being run from the server
8	Average size of application executables being run from the server (in megabytes)
9	Multiply line 7 by line 8
10	Running Microsoft Internet Information Server?	add 0MB	add 16–48MB
11	Running Microsoft SNA Server?	add 0MB	add 8MB
12	Running Microsoft Systems Management Server?	add 16MB	add 16–24MB
13	Running Microsoft Exchange Server?	add 8MB	add 16MB
14	Running Microsoft SQL Server?	add 0MB	add 8MB
15	Subtotal (add lines 2–4, 6, and 9–14)
16	Is server a RISC-based computer? of line 15	add 15% of line 15	add 15–25%
TOTAL ESTIMATED RAM NEEDED (add lines 15 and 16)

Caution: Don’t put more than 16MB of memory on an ISA bus computer running Windows NT. Having more than 16MB can actually reduce performance dramatically, since NT has to perform special DMA buffering operations to memory above the 16MB line. If you’re going to need more than 16MB of RAM on your server (and you will), opt for a system bus other than ISA.

What Type of Memory Is Best?

Most computers today support standard DRAM (Dynamic RAM) in the form of SIMMs (single in-line memory modules) that plug into the motherboard. DRAM is typically fine for your NT server computer. Just be sure to use the appropriate speed recommended for your specific computer model.

EDO (Enhanced Data Output) RAM provides faster data throughput that some folks claim eliminates the need for a secondary cache. Indeed, in computers with no secondary cache, EDO RAM consistently outperforms DRAM. However, a cacheless EDO or DRAM computer is slower than a computer with a secondary cache. And a cacheless computer with EDO is typically slower than a computer using DRAM with a secondary cache (depending on the size of the cache). So take EDO’s claim to bring about a “cacheless society” with a grain of salt.

As I mentioned earlier in this chapter, I strongly recommend having a secondary cache. In computers with a secondary cache, EDO RAM provides about a 5 percent performance improvement over DRAM. At the time of this writing, EDO RAM is significantly more expensive than DRAM. It’s probably not worth the price premium to squeeze out that extra 5 percent. Once EDO costs come down and are more in line with DRAM, though, EDO will be certainly worth considering. However, keep in mind that some computers based on the Pentium Pro won’t be able to use EDO RAM, since early versions of the Pentium Pro’s PCI chip set don’t support it.

ECC (Error Checking and Correcting) RAM can detect and correct single-bit errors in RAM. ECC RAM is starting to show up in high-end servers and workstations. Since a RAM parity error can bring Windows NT Server down, ECC provides an added level of robustness by catching and correcting single-bit errors before NT ever sees them. At the time of this writing, ECC RAM costs about 30 percent more than conventional DRAM.

Selecting Your Storage Devices

In this section, I discuss mass storage devices for your Windows NT server computer.

Floppy Drives

On an Intel-based computer, in order to install Windows NT Server 4.0, you’ll need a high-density, 3.5-inch floppy drive. (There are ways to get around the floppy drive requirement, but you won’t be able to create an Emergency Repair disk—which you’ll need.) You’ll also need to make sure the computer is configured to boot from this floppy drive. On RISC-based computers, there’s no such thing as a boot floppy drive, but since these computers are ARC-compliant, they include a floppy drive anyway. (ARC stands for Advanced RISC Computing, a standard developed to ensure compatibility of new RISC-based computers with operating system software. The ARC standard specifies PC-like hardware and BIOS features required by RISC computers that are designed to run Windows NT.)

Hard Disks

In this section, I show you how to estimate the amount of disk space that you’ll require on your NT Server computer, as well as your choices of disk hardware.

How Much Disk Space Do You Need?

I’ve provided you with another worksheet to complete, this time helping you to estimate how much free disk space you’ll need on your server computer. Fill out Worksheet 1-2 for each of the NT servers that you’re planning to install. I’ve included disk space requirements for Microsoft’s BackOffice applications, and you can add the space requirements of other server applications that you plan to run.

WORKSHEET 1-2   WINDOWS NT SERVER DISK SPACE ESTIMATE
Line	Disk Space Requirement	Minimum	Recommended	Estimate
1	Default NT system space requirement	250MB		250MB
2	Server computer memory (in megabytes)
3	Add line 2 and 162MB
4	Copy line 1 or line 3, whichever is larger
5	Average space required by an application installed on the server
6	Number of applications being run from the server
7	Multiply line 5 by line 6
8	Average space allocated to each user on this server
9	Number of users storing information on this server
10	Margin for error/growth			1.15
11	Multiply lines 8, 9, and 10
12	Running Microsoft Internet Information Server?	add 50MB	add 100MB
13	Running Microsoft SNA Server?	add 30MB	add 30MB
14	Running Microsoft Systems Management Server?	add 100MB	add 100MB
15	Running Microsoft Exchange Server?	add 250MB	add 500MB
16	Running Microsoft SQL Server?	add 45MB	add 100MB
17	Subtotal (add lines 4, 7, and 11–16)
18	Is server a RISC-based computer? on line 17	add 25% of line 17	add 25–30%
TOTAL ESTIMATED DISK SPACE NEEDED (add lines 17 and 18)

Of course, not all of this space needs to be on one disk drive or partition. Some network administrators choose to place the material on four separate partitions, using the following scheme:

• Operating system files
• User applications that are run over the network
• Server applications that run on the server
• User data

Tip: Have you looked at the list of NT-compatible hard disk drives and been shocked at the small number of devices there? Standard hard disk drives don’t need their own device drivers. They typically work just fine on NT if they’re attached to the computer via an NT-compatible adapter. Nonetheless, a handful of drive manufacturers have opted to certify their SCSI, IDE, and PCMCIA drives as NT-compatible. As a result, the number of hard disk drives on the compatibility list is frighteningly small. These few manufacturers saw marketing value in having their devices listed, even though there’s no technical reason why these or any other hard disk drives shouldn’t work on NT.

So, don’t believe that the NT compatibility list is the definitive rundown of hard disks that you can use. As long as a hard disk drive meets the IDE, SCSI, or PCMCIA standard, it should have no problem running on Windows NT.

The Case Against IDE

IDE, which stands for Integrated Drive Electronics, is currently the most popular type of adapter/drive combination shipped in today’s PCs. IDE owes its existence to predecessor PC hard disk drive technologies, including ST506 (used in the IBM XT) and ESDI (used in the IBM AT). Two drives, one acting as master and one acting as slave, can connect to one IDE adapter. Because of this relationship, only one of the two drives can be active at any given time. Moreover, if the master drive fails, both drives are down for the count. You can include up to four drives in a PC by using two IDE adapters. The four-drive limit can be a problem for servers that need to add more drives eventually.

EIDE, which stands for Enhanced Integrated Drive Electronics, is an extension of IDE that supports faster transfer rates, larger disk capacities, and storage devices other than hard disks. Unfortunately, it suffers from the same serialized drive access as IDE.

If you plan on using any of NT Server’s fault-tolerance RAID options for data redundancy or improved performance, forget using IDE or EIDE drives. Writes to and reads from multiple drives are serialized, causing the performance of mirrors and stripes to go down the tubes rather quickly. Even if you’re not using RAID features, using multiple IDE or EIDE drives can cause real server performance bottlenecks, as one drive locks the other out during disk accesses.

The Case for SCSI

The best high-performance disk solution for network servers is based on SCSI, the Small Computer Systems Interface. SCSI overcomes the serialization problem by allowing independent and concurrent access to each attached drive, and it allows more drives per adapter. You’ll need SCSI drives if you want to implement the RAID features of Windows NT Server. (SCSI’s ability to start an I/O operation, disconnect until the I/O is complete, and allow other devices to perform I/O during the disconnection make it ideal for RAID implementations.) Table 1-7 summarizes the characteristics of the SCSI technologies available at the time of this writing.

TABLE 1-7   CHARACTERISTICS OF AVAILABLE SCSI TECHNOLOGIES
SCSI Technology	Interface Data Path Width	Data Transfer Rate	Approach	Maximum Cable Length
SCSI-1: SCSI	8-bit (narrow)	5MBps	Parallel data transfer	20 ft
SCSI-2: Fast SCSI	8-bit (narrow)	10MBps	Doubles bus clock rate of SCSI-1	20 ft
SCSI-2: Wide SCSI	16-bit (wide)	20MBps	Doubles data path width of Fast SCSI	20 ft
SCSI-2: Fast & Wide SCSI	16-bit (wide)	40MBps	Dual Wide 20MBps channels	10 ft
UltraSCSI Fast 20	8-bit (narrow)	20MBps	Doubles clock rate of FAst SCSI	5 ft
UltraSCSI Fast 40	16-bit (wide)	40MBps	Doubles data path of UltraSCSI FAst 20	5 ft
UltraSCSI Fast 80 (emerging)	16-bit (wide)	80MBps	Dual UltraSCSI Fast 40 channels	5 ft

Note: You may see references to SCSI-3, sometimes erroneously equated with UltraSCSI. They aren’t the same. UltraSCSI is probably the last step in parallel SCSI technology. SCSI-3 refers to a high-performance serial technology, which will likely be used primarily for digital video. FireWire is one implementation of SCSI-3. It’s currently used by some digital camcorder manufacturers.

A SCSI bus can support as many devices as it has bits in its data path. In reality, one of these device addresses is automatically consumed by the SCSI adapter itself, since it’s considered a SCSI device on the bus. So, an 8-bit SCSI bus can support up to seven devices attached to its adapter. A 16-bit wide SCSI bus can support up to 15 devices. Dual-channel SCSI adapters appear to the operating system as two separate adapters. Together, the two channels can support twice as many devices as single-channel adapters, consuming one device address for each channel on the adapter. For example, Fast and Wide SCSI adapters can support up to 30 devices attached to the bus.

Note: Windows NT Server currently supports only eight SCSI addresses per adapter, including the adapter’s address itself. So, 16-bit wide SCSI adapters on NT can only address 7 drives, not 15. Dual-channel adapters appear to NT as two distinct adapters, so if each channel supports up to 7 devices, a dual-channel adapter in NT can support up to 14 devices.

If you’ve had much experience with SCSI or IDE drives, you may have encountered an annoying behavior that seems to defy explanation. Occasionally, the drive disappears or goes to sleep for a few seconds, although it sounds as if the heads are seeking across the disk. This can happen when the drive is idle or just as the drive is being accessed. If the latter occurs, you notice a delay in responsiveness from the operating system or application that’s trying to access the disk.

What’s going on during this period when the drive seems to have a mind of its own? It’s going through a thermal recalibration cycle, sometimes called T-cal. This is an internal housekeeping task in which the drive makes sure that all of its heads can still correctly read and write data, even though temperature changes have altered the mechanics of the drive. Some disk manufacturers now offer AV (audiovisual) drives, which have been tuned to minimize or completely eliminate this time-consuming operation. Although designed originally to prevent dropped video frames and sound samples in audiovisual applications, these drives provide excellent, consistent performance on servers.

Cabling SCSI Drives

Next to IRQ conflicts, the most common hardware configuration problems on PC servers stem from improper SCSI cabling and termination. If you’re maintaining an NT Server, you should become an expert on these SCSI topics.

There are two types of wiring used in SCSI buses. The most common is called single-ended, which uses one wire for each signal on the bus. Single-ended SCSI is so widespread that it’s often referred to as normal SCSI.

The less common approach is called differential. This approach uses two wires for each signal, sending the actual data on one wire and its logical inverse on the other. The receiving end uses the difference (or differential) between the two signals to determine and verify what was sent on the bus. Differential SCSI is less susceptible to external noise, allowing you to use longer cables for reliable data transfer. Although differential SCSI is a bit hard to find today, it’s worth considering if you plan on placing a large number of devices on the SCSI bus, if you intend to push the SCSI cable length to its limits, or both.

Caution: Never mix single-ended SCSI peripherals or adapters with differential SCSI equipment on the same SCSI bus. Not only will you see strange behavior, but you could also lose valuable data on your disks.

SCSI bus termination is absolutely critical to your success, both in terms of performance and reliability. An improperly terminated SCSI bus will behave in ways that will make you lose lots of sleep. Sometimes, it will work just fine. Other times, you’ll get high error rates and retries. In other instances, the computer won’t even see some of the SCSI devices at boot time. Basically, both ends of the SCSI bus must be properly terminated. Figure 1-1 illustrates the three most common configurations and how they must be terminated.

The cabling and termination requirements become a bit more complicated when you’re dealing with some of the new Wide SCSI adapters that include three connectors. A good example of this is the Adaptec AHA-2940W. Internally, the adapter includes both Fast SCSI-2 and Wide SCSI connectors. Externally, it provides a Wide SCSI connector. You can use any combination of two of the connectors, but not all three. So, the SCSI bus still has only two ends, and the devices at the far ends of the two cables that you use must be terminated. If you use only one connector, the SCSI adapter provides its own termination.

Figure 1-1: Proper termination is required at both ends of the SCSI bus.

Tip: When you remove termination resistor packs from a SCSI peripheral or adapter, don’t bend the pins. Tape the resistor packs for a specific device together, and label the bundle to identify from which device it came. If you ever want to reconfigure your SCSI bus, you may need to resurrect these resistor packs for reinsertion. Devices with termination switches are preferable, since you don’t have to worry about keeping track of these loose small parts.

 

Some more recent SCSI adapter offerings provide an automatic termination feature. These adapters automatically sense the drives on the bus and establish their termination at boot time. If you can find these adapters, they do make SCSI life a bit easier for you.

Don’t scrimp when it comes to SCSI cable quality. You need tight-fitting, well-shielded cables with high-quality (preferably gold) connectors. Otherwise, you’ll invisibly waste some of your server bandwidth recovering from data errors that you don’t know you have. If you experience reliability problems, even with good cabling, consider a small investment in active SCSI terminators, which provide higher reliability than passive terminators in some environments.

Worksheet Entry

In Worksheet 1-3 (Windows NT Server Setup) on line 3, list the SCSI adapters in your computer. Also, list any special storage devices that NT doesn’t support directly. If you’re using storage devices (other than SCSI, IDE, or EIDE hard disks) that aren’t on the NT Hardware Compatibility List, circle YES on line 4. Have the NT driver floppy disk supplied by the hardware manufacturer available during installation. If you don’t have a driver floppy disk, look on your Windows NT Server 4.0 CD-ROM in the \DRVLIB\STORAGE\RETIRED\<platform>\README.TXT file to see if a driver for your device is supplied. (Replace <platform> with I386, MIPS, ALPHA, or PPC, depending on what CPU platform you’re installing.) Follow the instructions in the README.TXT file to create a driver floppy disk before you commence NT installation.

Selecting Your CD-ROM Drive

I strongly recommend that you attach an NT-compatible CD-ROM drive to the computer on which you plan to install Windows NT Server. Doing this will make the installation process in Chapter 3 much easier for you.

Unless you already have a non-SCSI CD-ROM drive attached to your server, I recommend that you use a SCSI CD-ROM drive. Having a SCSI drive will make it easier for you to install Windows NT and configure your system, since most such drives will work when attached to an NT-compatible SCSI adapter. Non-SCSI CD-ROM drives often require a special driver, either supplied with Windows NT or by the drive manufacturer. If your server is a RISC-based computer, Windows NT Server requires that you have a SCSI CD-ROM drive attached.

Stay away from portable CD-ROM drives that hang off the parallel port of your computer. Although they’re quite convenient, driver support for NT is often lacking. Even with driver support, these drives are typically very slow. Likewise, if you’re thinking about using a SCSI adapter that attaches to the parallel port, avoid this temptation for the same reasons.

A wide range of CD-ROM drive performance is available today. At the time of this writing, 2X drives are installed in low-end systems, and the drives themselves are down to commodity pricing. Many desktop computers ship today with 4X drives installed, 6X and 8X drives occupy the high end in powerful multimedia workstations and servers, and 10X drives are just now emerging on the market. (There are a few oddball 4.5X drives on the market. I think they were designed as a Candid Camera prank to confuse patrons of computer stores.) For your Windows NT Server, if you’re just going to use the drive to install NT and application software, I recommend a minimum speed of 2X. If you’re going to share the drive over the network with multiple users, opt for the 4X speed or faster.

Worksheet Entry

In Worksheet 1-3 (Windows NT Server Setup) on line 1, circle the type of CD-ROM drive that you’re going to use for Windows NT Server installation. Circle SCSI, IDE, or Other. If you have non-SCSI CD-ROM drives, list them on line 3. If you’ve circled SCSI, and you have more than one SCSI CD-ROM drive attached to your computer, determine the SCSI device address of each drive, and then identify which one has the lowest SCSI device address on line 2 of the worksheet.

Tip: CD-ROM changers, which shuffle three to six discs in a single drive, are great for keeping several CD-ROMs ready to run. However, don’t use them to share CD-ROMs from your network server. The time required to change CD-ROMs mechanically is typically high, and two users accessing different CD-ROMs will drive each other (and your server) crazy as the discs are constantly swapped. If you need to share multiple CD-ROMs on the network, use separate drives attached to your server.

If you’re planning to share your CD-ROM drive over the network, consider a drive with a large internal cache. CD-ROM drive caches range from 64KB to 1MB today. If all you’re going to do is install NT and other CD-based software but aren’t going to use the CD-ROM drive otherwise, a cache doesn’t buy you much. If you’re going to have multiple users accessing your drive, go for the biggest cache you can get.

Selecting Your Tape Drive

You’ll need at least one NT-compatible tape drive to back up data on your Windows NT Servers and other computers on your network. Again, select a drive from the NT Hardware Compatibility List. Based on personal experience, I suggest that you shy away from tape drives that attach to the floppy controller. They’re excruciatingly slow and often less reliable than SCSI tape drives.

I highly recommend SCSI tape drives for use with NT Server. If you can afford it, pick a 4mm DAT drive. These units are typically very fast, reliable, and capable of recording huge amounts of data on a single tape. If you need to perform unattended backup operations that span multiple tapes, there are a handful of SCSI robotic autoloader tape units available for NT. (NT’s built-in NTBACKUP utility doesn’t support autoloaders. You’ll need third-party backup software, such as BackupExec, to take advantage of these high-end devices.

Tip: The NTBACKUP utility that comes with Windows NT Server doesn’t perform software compression, so if you plan to use this program, expect to get about half the tape capacity advertised. You can either invest in a third-party backup utility that performs software compression or pick a tape drive that does its own compression in hardware. Either way, there’s an extra cost that you need to consider.

Selecting Your System Unit

For a Windows NT server computer, a full-height tower case is best for maximum expandability and heat dissipation. Even if you don’t need all of it right away, opt for at least a 300-watt power supply and as many open drive bays as possible. As your server needs grow, you’ll be able to add disks and other devices easily. Also, make sure that your case can accommodate the largest motherboards (typically called standard AT size), in case you ever have to replace the motherboard.

Do You Really Need a UPS?

The answer to this question is a resounding “yes” for NT Server computers. Windows NT Server has full support for several UPS models. It allows you to connect the UPS to a serial port, take specific actions when the power fails, send notification messages to other nodes on the network, and perform an orderly shutdown of the server. When selecting a UPS, either pick one from the compatibility list, or make sure that the UPS manufacturer supplies the appropriate cables, software, and information to work with Windows NT Server.

Selecting Your Video Subsystem

The minimum requirement for video on Windows NT Server is VGA (640×480, 16 colors). This is perfectly fine for most servers out there, especially if you plan to perform your network and server administration tasks remotely from a different computer. In this situation, you’ll only use the monitor attached to the server for infrequent maintenance and configuration tasks.

Note: You may have been exposed to the concept of headless servers, which have no video monitor at all. Although some operating systems support this configuration, Windows NT Server doesn’t. You need at least a VGA adapter and monitor installed on your NT computer. (You can use an electronic switch box to share your monitor between multiple computers, but that’s the closest that NT comes to a headless server.)

If you plan to do your server administration at the server console, you’ll probably want to consider higher-resolution video. You’ll want to have the ability to run multiple administration and performance-monitoring applications and have them visible at the same time on the screen. In this situation, I recommend a minimum of a 17’’ monitor and an SVGA video adapter with 1024×768 resolution. You can certainly go up from there, but it might be difficult to justify. I advise against using a fancy graphics accelerator card that may burn one of your precious high-speed PCI slots. On a server, there are better uses for this slot, such as network and SCSI adapters.

Tip: I’m a firm believer in using screen savers to protect your investment in your video monitor. However, when selecting a screen saver to run on your Windows NT Server, think about its impact on your users.

Worksheet Entry

In Worksheet 1-3 (Windows NT Server Setup) on line 29, record the make and model of your video adapter. On line 30, record the video chip set that your adapter uses, if your adapter documentation includes this information. On line 31, indicate the number of colors that you want to use. Finally, on line 32, enter the screen resolution that you plan to use (for example, 640×480, 800×600, or 1024×768). You’ll need all of this information toward the end of the NT installation process.

Always select a screen saver that has very low CPU impact. Don’t use OpenGL screen savers or third-party products that send kitchen appliances or bathroom fixtures careening across your server screen. Your server customers’ needs come first. Be kind to them by using the simpler screen savers. The Blank Screen choice is always an excellent one.

Regardless of what type of video adapter and monitor you select, Windows NT Server Setup will use VGA mode during the installation process. At the end of this process, you’ll have an opportunity to select the appropriate resolution and color depth for your video adapter.

Selecting Your Input Devices

Keyboards are really a matter of personal preference. Most PC keyboards today are fully compatible with Windows NT, so the choice is pretty wide open. If you’re going to spend significant time at the server performing network administration functions, you may want to opt for one of the ergonomic keyboards at an additional cost of $60 to $100. My fingers tend to slide off the slanted keys, but you may have better luck with them.

You need a mouse or some other NT-compatible pointing device on your server. In theory, you could do everything from the keyboard, using the command prompt for text and shortcut keys in windowed applications. However, you’ll operate much faster and more efficiently if you use a mouse.

Worksheet Entry

In Worksheet 1-3 (Windows NT Server Setup) on lines 8 and 9, record the make and model of your keyboard and the special keyboard layout setting for your part of the world.

Worksheet Entry

In Worksheet 1-3 (Windows NT Server Setup) on line 10, record the make and model of your pointing device.

I recommend using an inport or PS/2 mouse that doesn’t tie up one of your serial ports. Serial and inport mice work equally well, but if you take my earlier advice and attach your server to a UPS that requires a serial port, you’ll only have one standard serial port left for a modem or serial printer. You don’t want to waste that precious port on a serial mouse. Many motherboards today have mouse ports built into them in addition to two serial ports, so the decision has already been made for you on these computers.

In Windows NT Server 4.0, you’ll definitely need two buttons on the mouse. The user interface consistently makes use of both buttons, unlike earlier versions of Windows and Windows NT. So, if you’re considering using that one-button mouse that’s hiding at the bottom of your hardware bin, don’t.

CHOOSING YOUR FILE SYSTEM

During Windows NT Server 4.0 installation, you’ll need to specify which file system you want to use for the NT operating system files. If you have multiple hard disks and partitions, you’ll be able to use a combination of file systems on your computer. However, during NT Setup, you’ll need to select the file system on which NT itself will reside. Table 1-8 summarizes the pros and cons of the FAT/VFAT and NTFS file systems.

TABLE 1-8   CHARACTERISTICS OF WINDOWS NT FILE SYSTEMS
File System Characteristic	FAT/VFAT	NTFS
Supports undelete (after rebooting to DOS and running UNDELETE)	Yes	No
Accessible by dual-booting to DOS, Windows 3.x, Windows 95, and OS/2	Yes	No
Allowed on system partition of RISC-based computer	Yes	No
Can be used on a floppy disk	Yes	No
Uses space efficiently on partitions under 200MB	Yes	No
Performance degrades as partition size increases	Yes	No
Supports long file names	Yes	Yes
Supports file security	No	Yes
Offers file system recoverability	No	Yes
Supports hot-fixing when disk sectors fail	No	Yes
Supports large partitions	No	Yes
Supports large files	No	Yes
Uses space efficiently on partitions over 400MB	No	Yes
Supports POSIX applications	No	Yes
Supports Macintosh files	No	Yes
Supports file compression on Windows NT	No	Yes
Faster for small directories	Yes	No
Faster for large directories	No	Yes
Supports volume sets to expand existing partitions	No	Yes
Cluster sizes	512 bytes to 16KB	512 bytes to 4KB
Maximum partition size	2GB(DOS)/4GB(NT)	16 exabytes

Securing Your Data

If you’re planning to dual-boot your server between Windows NT Server and other operating systems such as DOS or OS/2, and you want these other operating systems to have access to your files, select the FAT file system. Only NT can access files on partition formatted as NTFS. Other operating systems can’t.

From a data security viewpoint, selecting NTFS is a better choice than FAT for two reasons. First, NTFS offers you file-by-file security, allowing you to restrict and grant individual users and groups different levels of access to files on the partition. Second, NTFS partitions can’t be read directly if you run a different operating system on the computer.

However, don’t assume that your data is secure from prying eyes if you use NTFS on your server and impose tight file security. If someone has physical access to your server computer, reboots it under DOS using a bootable floppy disk, has intimate knowledge of NTFS on-disk structures, and uses a low-level disk editing utility, they can read your NTFS files. NTFS doesn’t encrypt your data.

Caution: It now requires even less knowledge of NTFS to access files from DOS. Students at the University of Oregon have developed a utility called NTFSDOS that can read NTFS files while booted under DOS, Windows 3.x, or Windows 95. They’ve made this program available on the Internet. So assume that data thieves have it in their toolbox, as you plan your security measures to thwart their efforts.

You can limit your exposure to this type of security breach. First, restrict physical access to your server by locking it in a room that only authorized people can enter. Second, configure the computer to boot only NT from hard disk, rather than presenting a dual-boot menu. In addition, on some computers, you can change the server’s CMOS settings to prevent booting from a DOS floppy disk. Finally, you can use a RISC-based computer that can’t boot to DOS at all.

Physical security of your server is critical. If a data marauder does get access to your server and can make off with your hard disk, they can attach it to another computer to access the data. To help thwart this plan, use internal hard disks rather than easily transportable external drives.

Converting FAT to NTFS

You can convert from FAT to NTFS either during the NT installation process or after installation using the CONVERT command line utility. You’ll need about 100MB of free space on the partition to be converted, because NT needs this to write temporary information during the conversion process. Depending on the size and content of your FAT partition, conversion to NTFS can be a lengthy process. Conversion of a 1GB partition, for example, can take several hours. If and when you decide to convert a FAT partition to an NTFS partition, allow sufficient time to complete the process. Since the partition is inaccessible during the conversion, start the process in the evening and allow it to grind overnight.

Caution: Converting from FAT to NTFS is a one-way door. You can’t go directly back to FAT. The only way back is to back up your data elsewhere, reformat the partition as FAT, and copy the data back to the partition.

Using Space Efficiently

On small disk partitions (say, less than 200MB), NTFS imposes more on-disk overhead than FAT does. Thus, NTFS burns a larger percentage of your disk space just to handle housekeeping chores. On disk partitions of this size, FAT uses space more efficiently. If your partition is between 200MB and 400MB, the two file systems are about on par in terms of on-disk overhead. As the partition size climbs above 400MB, NTFS becomes increasingly much more efficient since its on-disk overhead doesn’t increase as the partition size grows. FAT, on the other hand, gobbles up more space for its own use as the partition size climbs.

NTFS supports compression on a file-by-file basis. Unlike DOS compression schemes that compress an entire disk partition into one huge file, you can select individual files and directories to be compressed automatically by NTFS. NTFS was designed to perform fast decompression of files, at the expense of having files that are slightly larger than those created by other compression schemes. You specify what you want to compress after you’ve installed Windows NT Server.

Comparing Performance

There are no easy answers to which file system is the faster one. The relative speeds observed are highly dependent on the operations being performed, the size of the directories, how badly fragmented the files are, and so on.

NTFS recoverability and security both add overhead to file access that FAT doesn’t require. On the other hand, NTFS lays out its directories in a way that minimizes the number of disk accesses required to find a file. Moreover, NTFS handles small files much more efficiently than FAT.

On FAT, random access to a badly fragmented file typically takes longer than it does on NTFS. Creating new files and listing directories are quicker on FAT. Opening a file on FAT is sometimes faster than NTFS, unless the file is near the end of the directory. Then it’s slower than NTFS.

In summary, performance differences between FAT and NTFS vary from one operation to the next and depend on the location of the file, size of the directory, and amount of fragmentation. Overall, the performance of the two file systems is on par, even though NTFS is more robust and secure than FAT.

Accounting for the RISC Factor

The boot partition on RISC-based computers must be formatted as FAT. This partition contains a few hardware-specific NT files. You can install Windows NT Server on this partition (if there’s room), or on another one. I highly recommend installing NT on a separate partition, formatted with NTFS. This approach will give you maximum reliability and security of your NT system files.

Handling Unknown File Systems

Windows NT doesn’t support DOS file compression, HPFS, or non-FAT file systems created by other operating systems. In addition, NT may not correctly recognize disks if they’ve never been partitioned or formatted. If you have any of these conditions on your computer, you’ll need to do a bit of preparatory work before attempting to install Windows NT Server. In this section, I discuss what to do in each of these situations.

Compressed Partitions

Windows NT doesn’t recognize or support DoubleSpace, DriveSpace, Stacker, or other disk compression schemes implemented for FAT under DOS. If you have compressed partitions, do the following on each one on your computers before commencing Windows NT installation:

1. Back up the data on the compressed partition.
You’ll need to back up either to tape or to another disk. Keep in mind that the data will occupy up to twice as much space on the backup, since it will be uncompressed.
2. Remove the compressed drive file(s) or reformat the partition as a normal FAT partition.
If you choose to remove the compressed drive file(s), refer to your disk compression product documentation for details on what files to remove.
3. Restore the data from the backup.
Again, keep in mind that the data will occupy up to twice as much space as it did on the original compressed partition.

HPFS Partitions

Support for the HPFS file system, originally designed for OS/2, has been phased out over the last few releases of Windows NT. Version 3.1 and 3.5 fully supported HPFS. Version 3.51 of NT supported it only to boot from an existing HPFS partition. Windows NT 4.0 has completely removed HPFS support. If you have one or more HPFS partitions and you want to use them under NT, do the following for each HPFS partition on your computer:

1. Back up the data on the HPFS partition.
2. Boot the computer to DOS and reformat the partition as a normal FAT partition.
3. Restore the data from the backup.

Unrecognized Partitions

If your computer’s hard disks have never been formatted or they were formatted under another operating system such as UNIX or VMS, NT Setup may complain that it has detected a possible virus. Although it’s possible that there really is a virus, Setup typically means that it found a nonstandard boot sector on one or more of your hard disks. (The boot sector is a common hiding place for viruses, so Setup concludes there might be a virus lurking in there. Since Setup doesn’t want to take the chance of installing NT on an infected system, it stops. Disks formatted under DOS, OS/2, or NT won’t generate this complaint, unless they actually do have a virus.)

If it’s possible that your boot sector has a virus, use a virus scanner to detect it and remove it. You can avoid Setup’s virtual virus complaint by doing a little extra work beforehand. If there’s any data on the drives that you want to save, back it up. Boot to DOS and partition and format the potentially offending drives to FAT. If you’re installing on a RISC platform, you can’t boot to DOS. To succeed on a RISC-based computer, you may have to unplug all potentially offending drives except the system drive, install NT Server, and then plug the drives back in once NT is installed.

Worksheet Entry

In Worksheet 1-3 (Windows NT Server Setup) on line 13, circle your file system choice for the partition on which you plan to install the NT operating system.

Making the File System Decision

I highly recommend that you use the NTFS file system on your server. Its excellent security, recoverability, and support of very large files and disks are critically important for your server now and as your system grows.

However, for the reasons I’ve discussed, you may decide to use FAT for some or all of the partitions on your server computer. If you do, I recommend using NTFS on the partition where you install the NT operating system itself, to protect your system files.

PLANNING FOR FAULT TOLERANCE

Although you don’t establish fault-tolerance associations between disk partitions during installation, it’s helpful to plan which, if any, of these features you’ll be using. NT Setup allows you to review, delete, and create disk partitions in preparation for NT installation.

If you’re planning on setting up a disk mirror (RAID 1), you’ll need two equal-sized partitions on two different hard disk drives. A stripe set without parity (RAID 0) has the same requirement. If you’re going to create a stripe set with parity (RAID 5), you’ll need at least three equal-sized partitions on different drives.

Cross-Reference: If you don’t deal with managing disk partitions during the installation process, you’ll be able to manipulate them using Disk Administrator, which I describe in detail in Chapter 9. See Chapter 10 for details about fault tolerance.

Worksheet Entry

In Worksheet 1-3 (Windows NT Server Setup) on line 12, record your plans for deleting and creating disk partitions in preparation for NT installation. Also, indicate which partition you want to use for the NT operating system files.

CHOOSING YOUR TRANSPORT PROTOCOLS

Before you install Windows NT Server 4.0, you’ll need to select the set of protocols that you want to run on the server. It’s best to make an on-purpose decision up front, although you’ll certainly be able to change your mind after installation. As with any major decision, you need to be armed with information. Table 1-9 summarizes the three main Windows NT Server protocols in terms of their general characteristics. You’ll need to select at least one of them to run on your server.

TABLE 1-9   CHARACTERISTICS OF THE THREE MAIN NT PROTOCOLS
Protocol Characteristic	NetBEUI	IPX/SPX	TCP/IP
Best performer on small LANs	Yes	No	No
Best performer for file and print sharing	No	Yes	No
Best performer for application servers	No	No	Yes
Easy client configuration	Yes	Yes	No
Easy network administration	Yes	Yes	No
Breadth of use in the industry	Microsoft and IBM networks only	Most popular of all PC network protocols	Most popular network protocol across all networks
Open protocol specifications	No	No	Yes
Network interoperability	Microsoft and IBM networks only	Available on several platforms	Available on almost all platforms
Ability to send packets across routers	No	Yes	Yes

Why not install all of the protocols for maximum flexibility? In theory, this is a fine idea, and it’s supported by Windows NT Server. In practice, however, each transport protocol eats up resources on your server, both in terms of disk space and memory. Using multiple protocols on client computers is even more painful, since disk space and memory are even more limited there. Also, the more protocols you’ve loaded, the more difficult it is to troubleshoot the network when a problem occurs. So, select and install only the protocols that you need.

For example, if you need Internet access, you want to use NT Server as an application server; if you have several Macintosh clients and one HP JetDirect network printer, you should opt for TCP/IP, AppleTalk, and DLC. In this situation, stay away from NetBEUI and IPX/SPX, and use TCP/IP as the primary protocol between PC nodes on your network. As another example, if you are integrating NT Server into an existing NetWare network, you don’t need Internet access; if your network is small, you could get away with IPX/SPX as your only transport protocol.

Table 1-10 shows the protocols that you’ll need to install on Windows NT Server to satisfy a variety of network types and requirements. I can’t tell you exactly what protocols to install, since networking needs differ. Moreover, your requirements will change over time. What I’ve done here is provide you with the kinds of criteria to consider when selecting a protocol.

TABLE 1-10   PROTOCOL SELECTION CRITERIA
Network Type or Requirement	NetBEUI (NBF)	IPX/SPX (NWLink)	TCP/IP	DLC	AppleTalk (SFM)
Pure Microsoft [small] network	ü
Wide area network			ü
Large network			ü
Small network	ü	ü
Integration with NetWare		ü
Integration with UNIX			ü
Integration with mainframe				ü
Integration with Apple Macintosh clients					ü
Connection to the Internet			ü
Connection to network printers such as HP JetDirect				ü

Notice that TCP/IP appears to be the most flexible. However, the cost of that flexibility can mean added complexity for you. Remember that you’ll need outside help to get your network addresses assigned. If you don’t need Internet access right away and your network is fairly small, consider using IPX/SPX for now. When you need Internet access in the future, you can easily convert to TCP/IP.

Worksheet Entry

In Worksheet 1-3 (Windows NT Server Setup) on line 24, circle the protocol or protocols that you plan to install. If you want to install AppleTalk (SFM) or DLC, you’ll be able to do so after you install Windows NT Server.

NAMING YOUR NT SERVER

Each computer in your network needs a unique name, with a maximum of 15 characters. Windows NT Server computers are no exception. Once you assign a name to a server, it’s difficult to change it.

Tip: Although you can include spaces in computer names, don’t do it. First, if you have Windows for Workgroups computers in your network, they’ll have trouble accessing servers with spaces in their names. Second, whenever you include the name of the computer in a command line, you’ll have to remember to surround the name with quotes because of the spaces. So stay away from spaces.

It’s important to give your servers names that won’t have to change as your organization changes. Your network users need to be able to access server resources consistently. If you’ve changed a server name before, you’re familiar with the downstream effect that it has on users who must change their batch files and other settings to point to the new server. Don’t base server names on the names of people in your organization. As reorganizations occur and people leave, are promoted, or move from one department to another, server names shouldn’t have to change. Likewise, you don’t want to use a physical location as the computer name, since you may need to move the server to another office, building, or site.

If you’ve been using your server computer as a peer-to-peer server, and you’re planning to install Windows NT Server on it, consider giving the computer the same name it had before. That way, other computers on the network will see this computer and be able to connect to it. They’ll have no idea that you changed operating systems out from under their server.

Worksheet Entry

In Worksheet 1-3 (Windows NT Server Setup) on line 18, write the computer name for the NT Server that you’re going to install.

CHOOSING NT SERVER’S ROLE

You have two alternatives in designing the overall structure of your Windows NT Server network. You can set up a workgroup (or peer-to-peer) environment, much like what Windows for Workgroups offers, or you can establish one or more network domains. Table 1-11 summarizes the differences between these two approaches to establishing your network.

TABLE 1-11   DOMAINS VERSUS WORKGROUPS
Characteristic	Domain Model	Workgroup (Peer-to-Peer) Model
Good for very small networks		ü
Good for larger networks	ü
Easy to establish		ü
Requires planning to establish	ü
Centralized user account management	ü
Centralized security management	ü
Single user has one account for the entire network (single network logon)	ü
Single user has separate accounts on each server accessed		ü
User accounts scattered across multiple servers		ü
Potential for different user passwords on different servers		ü

When you install Windows NT Server, you’ll need to specify whether the computer is to act as a stand-alone server or as a domain controller. Even for small networks, I recommend taking the domain approach with Windows NT Server. Small networks have a way of growing into larger ones, usually more quickly than planned. If you start with a centralized approach to administration, you won’t have a mess of user accounts and servers to clean up later. Moreover, you won’t have to train yourself and your users twice if you begin with the domain model.

In the following sections, I cover the requirements for installing NT Server as a stand-alone (peer-to-peer) server, as a primary domain controller (PDC), and as a backup domain controller (BDC).

Creating a Stand-Alone Server

If you elect to install Windows NT Server as a stand-alone server, it can participate in a domain or workgroup just like Windows NT Workstation or Windows for Workgroups computers. It will be able to share resources with other nodes on the network. It cannot act as either a PDC or BDC. A stand-alone server is preferable if it’s going to be administered separately from the rest of the domain or if you plan to move it to another domain.

Caution: Make sure that you really don’t want this Windows NT Server computer to be a PDC or BDC in your domain. If you change your mind later, you’ll have to reinstall the operating system from scratch. There’s no way to upgrade a stand-alone server to a BDC or PDC role.

Establishing a Domain—The PDC

A domain requires one and only one Windows NT Server computer playing the role of primary domain controller (PDC). You create the domain by installing Windows NT Server as a PDC on the network. The first Windows NT server computer in the domain must be the PDC. Once the domain is established, additional NT server computers can be added to the domain, to act as backup domain controllers (BDCs) or as stand-alone servers.

Before installing NT Server on your computer as a PDC, you need to make sure that it contains a working network adapter and is properly attached to the network. (You might not have any other computers on the network at this point, but that’s OK.) During installation, the computer will check the network to assure that you aren’t trying to duplicate an existing domain. If Setup can’t scan the network, PDC installation will fail.

You’ll need to select a name for your domain. Once you’ve established your domain, it’s a nightmare to have to change its name. If you’re setting up a domain for each department in your organization, you may want to use the department name as the domain name. However, if your department names change frequently, don’t use this model. Changing a domain name is extremely painful, so pick a naming convention for your domains that will remain stable over time. See the previous section on “Naming Your NT Server” for more hints on naming strategies.

Creating BDCs in Your Domain

You’ll need to plan on having at least one other Windows NT server computer on the network to play the role of BDC, in case your PDC computer becomes unavailable. When you install a BDC, you specify which existing domain it will join. (If you install more than one version of Windows NT Server 4.0 on your network, you’ll need licenses to cover each of the copies that you install.)

Worksheet Entry

In Worksheet 1-3 (Windows NT Server Setup) on line 19, circle the role of this computer: PDC, BDC, or stand-alone server. Line 20 is there to remind you that you’ll need the domain administrator password. For security reasons, don’t write the password on the worksheet, but be sure to have it available during installation. On line 27, write the new domain name that you’ve selected, if you’re installing a PDC. If you’re installing a BDC, write the name of the domain in which this computer will participate. If you’re installing a stand-alone server, write the name of the domain or workgroup in which this computer will participate.

How many BDCs does your domain need? If you have less than 2,000 users on your network, you can get by with one PDC and just one BDC. However, the more BDCs you add to your domain, the better responsiveness your users will see when they attempt to log on to the domain simultaneously. Extra BDCs spread the load of user authentication and provide an additional level of fault tolerance for your network, should the PDC or some of the BDCs go down.

Cross-Reference: If you have more than 2,000 users or you’re considering establishing multiple domains in your enterprise, take a look at Chapter 4 for the scoop on how to plan larger domain-based networks.

Before you install NT Server on your computer as a BDC, you’ll need to make sure that several things are in place. First, your computer must be attached to a working network on which your existing PDC resides. Second, the PDC must already be up and running to allow the BDC to contact it during installation. Finally, you’ll need to be armed with the domain administrator password in order to add the BDC to the domain.

CHOOSING YOUR LICENSING OPTION

Microsoft’s license agreement for Windows NT Server and other BackOffice products requires a license for each server and each client on the network. For Windows NT Server, there are two licensing options from which to choose—namely, Per Seat licensing and Per Server licensing. Making the right choice requires some knowledge of how client computers will use your NT server computer.

Per Seat Licensing

Per Seat licensing is the easier to understand. You need a license for each server and a license for each client (seat) on the network. If you add a client, you pay only for one additional client license. If you add a server, you pay only for one additional server license. A Per Seat license is tied to the client computer and gives the client the right to access any number of NT server computers. The Per Seat license option is the more cost-effective approach if all of your network clients need to access your NT server at the same time or if your clients tend to connect to multiple NT server computers.

Per Server Licensing

Per Server licensing is a little more complicated. It’s designed to save you money in certain situations. In this option, you license the server for the maximum number of clients connecting to your NT Server at the same time. Let’s consider an example. Say you have 100 computers on your network, but you know that no more than 50 (any 50) of those computers need to access your NT server at any given time. With Per Seat licensing, you’d have to purchase 100 client licenses. With Per Server licensing, you’d save money by buying only 50 licenses. A Per Server license is tied to the NT Server computer and gives you the right to make a connection from any client on the network. The Per Server licensing option is the more cost-effective approach if only a subset of your client computers connect to your NT server at any given time, clients connect to only one server, or they connect infrequently.

Tip: If you’re not sure which licensing option is best for you, select the Per Server option. If you decide later that Per Seat licensing is the way to go, you’ll get a one-time free opportunity to convert to Per Seat licensing. However, you can’t go the other direction for free.

If you have only one NT server on your network, opt for Per Server licensing. You can always change to Per Seat licensing later, if the number of servers in your network grows. If you have multiple NT servers, and the number of Per Server client licenses tied to these servers exceeds the total number of client computers in your network, you’ve reached the point where Per Seat licensing is cheaper.

Worksheet Entry

In Worksheet 1-3 (Windows NT Server Setup) on line 17, circle the licensing option you’ve chosen.

UPGRADING TO WINDOWS NT SERVER 4.0

During NT Server installation, you’ll have the opportunity to upgrade your existing operating system or install NT in a separate directory. If you choose the latter, you’ll typically be able to dual-boot between your old and new operating systems. In this section, I fill you in on what to expect from each type of upgrade.

Upgrading from Previous NT Versions

If you decide not to upgrade an existing NT installation, NT Setup will create a new option in the system boot menu, allowing you to dual-boot between the old and new versions of NT.

If you elect to have Windows NT Server 4.0 upgrade an existing version of Windows NT Server on your computer, NT Setup will overwrite the old NT installation. You’ll no longer be able to boot to the old version of NT that you upgraded, since system files will have changed. Setup will migrate your NT 3.x Program Manager settings into the NT 4.0 Start Menu and will bring forward just about everything else. It will retain your desktop settings, user account and security information, drive letters, fault-tolerance volume settings, and so forth.

Caution: If you decide to have Setup upgrade from Windows NT 3.1 or Windows NT Workstation 3.5x to Windows NT Server 4.0, the computer won’t be able to act as either a PDC or BDC for your domain. If you want your Windows NT Server 4.0 computer to play the role of domain controller, you’ll need to opt for a fresh installation of NT 4.0.

Upgrading from Windows 3.x and Windows for Workgroups

If you decide not to upgrade over 16-bit Windows, you’ll probably have to reinstall your Win16 applications while running NT. Moreover, your desktop, Program Manager, and other settings won’t appear in your NT 4.0 installation.

If you decide to install NT in your 16-bit Windows directory tree, NT Server will put most of its files into the SYSTEM32 directory under your Windows root directory. It will also put a few innocuous files (such as bitmap files) into the Windows root directory. Setup will migrate your Program Manager groups to the Start Programs menu in NT 4.0. Setup will also migrate application .INI files.

If your Win16 applications are compatible with Windows NT, you’ll be able to run them from the Start Programs . . . menu just as if you had clicked on icons in Program Manager. Some applications require OLE and DDE files that they installed in the Windows System directory. NT Setup might not migrate these files and thus may not find them when you run your Win16 application. To work around this, add the Windows directory to your NT search path. This will allow NT to find the missing files.

The first time that you log on to the newly installed NT, it will ask if you want to migrate your WIN.INI, CONTROL.INI, and Program Manager groups. Your Windows desktop settings will be brought forward.

Note: WINFILE.INI won’t get migrated, so if you’re a diehard File Manager user, you’ll need to tweak your File Manager settings when you use the program for the first time under NT.

Because of their complexity, there are a few other settings that NT Setup won’t migrate for you:

• Any persistent net shares and net uses from Windows for Workgroups
• The default domain and user ID from Windows for Workgroups
• Any changes that you’ve made to the Main, Startup, Games, or Accessories Program Manager groups
• Font information for DOS character-mode command windows

Upgrading from Windows 95

If you’re considering installing NT 4.0 as an upgrade over Windows 95, don’t do it. At the time of this writing, Microsoft doesn’t support installing Windows NT Server 4.0 on top of a Windows 95 installation. NT doesn’t detect a Windows 95 installation on your computer, unless you installed Windows 95 in your Windows 3.x directory. Even if NT did detect it, though, NT’s ability to migrate Windows 95 settings is extremely limited, because of a lack of standards in how Windows 95 applications install themselves. Microsoft is working with software developers to nail this down. But for now, I advise against installing NT 4.0 as an upgrade over a Windows 95 installation.

If you have Windows 95 on your computer, install a fresh copy of NT Server 4.0 in a separate directory, rather than attempting to upgrade. That way, you’ll be able to dual-boot between Windows 95 and Windows NT Server 4.0. Note that you’ll have to install your applications separately for each operating system—once while running Windows 95 and once while running NT. None of your Windows 95 application information will be brought forward into your Windows NT installation.

Tip: Although the bad news is that you’ll have to reinstall all applications that you want to run under both Windows 95 and Windows NT, the good news is that you can probably install an application in the same directory both times. As long as the application’s configuration information is limited to the registry and to .INI files that don’t reside in the application’s installation directory, you can install in the same place each time. However, if the application keeps configuration data in its installation directory, the second installation will overwrite the settings created by the first installation. Determine where your application’s configuration files are kept, or consult your software manufacturer.

Worksheet Entry

In Worksheet 1-3 (Windows NT Server Setup) on line 5, circle YES if you’re going to upgrade an older version of Windows NT. Otherwise, circle NO (because you want to dual-boot between the old and new versions). On line 11, circle YES if you’re going to upgrade from a previous version of Windows. Circle NO if you aren’t. If you answered NO to both upgrade options, select a destination directory name where the NT system files should be installed. It must be an 8.3 file name, even if you’re installing to NTFS. Write this name on line 14.

Upgrading from NetWare

Novell employees might argue that converting a NetWare server to a Windows NT server isn’t an upgrade, but that’s what a number of corporations are doing today. To make this process as smooth as possible, Microsoft provides a conversion utility that migrates NetWare account information to the Windows NT Server account database. The NWCONV utility supports upgrading from either NetWare 2.x or NetWare 3.x. It doesn’t support conversion of NetWare 4.x servers to NT.

Upgrading from a NetWare server is a bit different from the other upgrades discussed so far. You’ll need to have Windows NT Server installed on a computer that’s different from your NetWare server. You’ll then perform the migration over the network, between the two servers. Because there’s no direct way to translate some of the account information from the NetWare world to the NT world, you should count on some manual efforts to complete the conversion.

Cross-Reference: I cover this entire process in detail in Chapter 6.

THE WINDOWS NT SERVER SETUP WORKSHEET

If you’ve been making written entries at the points that I’ve suggested throughout this chapter, you’re already quite familiar with the worksheet presented in this section. If you haven’t, be sure to fill it out now. With the completed worksheet at your side during Windows NT Server installation, the process will go more smoothly for you—and you won’t have to repeat it in order to correct bad decisions.

SUMMARY

In this chapter, I’ve given you advice about how to design your Windows NT server computer. In addition, you’ve made nearly all the decisions required prior to installation of the operating system. You’ve completed the Windows NT Server Setup Worksheet, which will help you successfully install NT. In Chapter 2, you’ll learn the nuts and bolts of designing and installing your network hardware, including configuration of your network adapters. Then in Chapter 3, I’ll walk you through the actual NT installation process.

<< Previous Page Next Page >>