Understanding the Hardware Landscape [Advanced Windows Small Business Server 2003 Best Practices book excerpt]

Hiya folks – I am the co-author and publisher of the Advanced SBS 2003 Best Practices book and I like to hold virtual book readings by posting up. So let me know how you like today’s passage on hardware.  BTW – my SBS 2008 book is here – so think about it 🙂


Understanding the Hardward Landscape

What makes an Opteron better than a Xeon, or vice versa? Truth be told, each processor architecture, memory technology, and platform design lends itself to particular environments. Understanding hardware’s strengths and weaknesses is key to making the right infrastructure choices.


Today’s Processor Technology

Don’t feel bad if you aren’t up to speed on the latest processor architectures. Both Intel and AMD are continually making adjustments to optimize for this or that, and it’s a full-time job keeping up with their advancements. Fortunately, despite divergent names, AMD’s Opteron and Athlon 64 families are built using very similar blueprints, as are Intel’s Xeon and Pentium 4 processors.

AMD Opteron

For the sake of alphabetical convenience, let’s start with AMD and its now-popular Opteron—centerpiece for many higher-end servers and workstations, such as HP’s ProLiant DL145. The Opteron is available in several different models: the 100 series, for use in single-processor systems; the 200 family, capable of dual- processing; and the 800 series, with support for up to eight-way configurations.

There are actually quite a few reasons why the Opteron is attractive in a small business environment. To begin, it offers perhaps the most compelling performance at its price inflection point, due to a number of distinct architectural features.

CHAPTER Chaper 2 1 So Understanding You Want to Hardwre Be an in SMB the SBS Consultant?!?! Environment

·           Before the Opteron emerged, processors communicated with RAM through another motherboard component called a north bridge, which housed the platform’s memory controller. AMD sought to minimize the latency overhead associated with the north bridge by integrating a memory controller directly on the processor die. In cutting out the figu­rative middle-man, AMD maximized real-world memory bandwidth and increased the Opteron’s responsiveness. The Opteron supports up to two channels of DDR400 memory for up to 6.4 gigabytes per sec­ond of bandwidth.

·                       The Opteron also boasts large caches (the small, temporary data re­positories located on the processor). Whereas Intel’s Xeon comes with 16 KB of L1 (level one) data cache and enough storage to hold 12,000 micro-ops, the Opteron features 64 KB each of L1 data and instruction cache. It also comes with 1 MB of L2 (level two) memory, similar to the latest models of Intel’s Xeon.

·                       In single-processor computers, all of the front-side and memory bus bandwidth is focused to just the one processor. But when you start adding extra processors, there needs to be a coordinated effort to man­age traffic between each CPU. Intel’s Xeon employs a shared bus, where only one processor can control the bus at a time, reducing effective throughput to each processing core. In contrast, the Opteron resides on a point-to-point bus. Each processor has a dedicated path to system memory, resulting in greater overall bandwidth in multi-processor sys­tems. As a result, the Opteron is considered to be highly-scalable—a great attribute for growing businesses.

·                       One of AMD’s principal selling points is 64-bit computing, supported by both its Opteron and Athlon 64 processors. Now, the most notable benefit of a 64-bit chip is the ability to address, or recognize, more than 4 GB of RAM, which today’s 32-bit processors cannot do. However, in order for a system to run in 64-bit mode, it requires a compatible pro­cessor, BIOS, operating system, and set of device drivers. Because Windows Server 2003 is a 32-bit operating system, the Opteron won’t employ the benefits of 64-bit processing. In all honesty, few small busi‑

y SECTIONBrelsford1 1 SBS 2003Consuting BestDeploymentPrac

nesses need more than 4 GB of RAM anyway, but it’s nice to know that the Opteron leaves the door open for a future migration to 64-bit tech­nology. Note that Microsoft’s plans for 64-bit SBSing (a future release of SBS that would natively support 64-bit processing) are not known as of this writing. This type of information is reported, as it becomes available, in the free SBS newsletter that you can sign up for at www.smbnation.com.

·                       In a small business environment with limited on-site IT support, a low- maintenance infrastructure is invaluable, and that includes a secure back- end that won’t succumb to malicious viruses propagated through security vulnerabilities. AMD’s Opteron includes a special hardware bit that prevents the execution of code in what’s called a “buffer overrun.” AMD claims that its Enhanced Virus Protection technology, used in conjunc­tion with Windows XP Service Pack 2 (or Windows Server 2003 SP1), would have protected systems against the MSBlaster and Slammer bugs.

·                       Beyond its most attractive features, the Opteron also supports the MMX, SSE, and SSE2 instruction sets pioneered by Intel. Figure 2-1 illus­trates the basic layout of Opteron’s core, and how the chip is able to work so efficiently.

Figure 2-1

With its on-die memory controller and HyperTransport connection, Opteron

enjoys low latency and high performance. Image courtesy of AMD.

CHAPTER Chaper 2 1 So Understanding You Want to Hardwre Be an in SMB the SBS Consultant?!?! Environment

For all of its advanced features, the underlying platform infrastructure for the Opteron is maturing more slowly than Intel’s competing Xeon. AMD doesn’t devote a lot of its time to chipset development, so most workstations rely on core logic from either NVIDIA or VIA, AMD’s most prominent chipset partners. Emerging technologies, such as PCI Express, aren’t yet prevalent in either chipset manufacturers’ products. Fortunately, PCI Express graphics cards and high­speed networking products aren’t yet necessities in small business servers or clients. A stable, integrated graphics processor works wonders, and any number of Gigabit Ethernet products should suffice in the networking department.

AMD Athlon 64

As I mentioned previously, the Opteron shares a number of common features with the Athlon 64, its desktop counterpart. It has the same integrated memory controller, similar caches (with a couple of exceptions), 64-bit computing extensions, the Enhanced Virus Protection feature, and built-in support for the most popular extended instruction sets, such as SSE and SSE2. Because the chip only runs in single-processor configurations, the whole debate about shared versus point-to-point buses doesn’t apply here.

The Opteron’s nomenclature is admittedly obscure (Opteron Model 250 doesn’t say a whole lot about what the chip can do or how it compares to Intel’s wares). Anticipating a more mainstream audience for its Athlon 64, AMD employed a different naming scheme. Each model is differentiated by a relative performance rating, such as 3800+ or 3500+, which roughly corresponds to Intel’s own Pentium 4 frequencies. An Athlon 64 3400+, for example, actually operates at 2.2 GHz and competes with Intel’s 3.4 GHz Pentium 4.

Along with its somewhat simpler rating system, the Athlon 64 differs from the Opteron in a couple of regards.

• When AMD first introduced the Athlon 64, it fit into a 754-pin socket. Those processors featured 1 MB of L2 cache, like the Opteron, but only supported one channel of DDR memory, effectively cutting the available bandwidth in half. Over time, AMD improved the platform and eventually released a 939-pin version of the processor, simulta­neously adding a second channel of DDR memory and cutting the L2 cache down to 512 KB. Today’s Athlon 64 comes in both 939-pin and 754-pin varieties, with a number of clock speed and cache memory


y SECTIONBrelsford1 1 SBS 2003Consuting BestDeploymentPrac


variations. Just remember that platforms centering on the new 939-pin interface will generally yield higher performance, despite the smaller L2 cache.

·         As processors evolve, they naturally get more complex. Shrinking lithography processes and ambitious manufacturers, eager to cram more features onto tiny processor dies, invariably lead to components that consume lots of power and dissipate plenty of heat. The Athlon 64 includes a feature called Cool’n’Quiet, which dynamically throttles the processor clock frequency according to computing demands. When an Athlon 64 system is under heavy load, the processor runs at full speed, and when it idles, Cool’n’Quiet reduces its speed, cutting back on both power consumption and heat. This is an especially useful feature, par­ticularly in small spaces that are intolerant to heat buildup.

·         There’s one aberration in the Athlon 64 family—AMD’s Athlon 64 FX. Functionally similar to the Opteron, the FX is a high-end enthusiast component destined for the very fastest desktop machines. It features an unlocked clock multiplier, a boon to enthusiasts dedicated to tweak­ing their hardware. Overall, it isn’t of interest to SMBs; simply be aware of this top-dollar flagship and its intended audience.


The Athlon 64 suffers from the same slow platform development as the Opteron, though the most modern chipsets really don’t lack any features other than PCI Express connectivity. Even still, don’t expect to see many Athlon 64-based servers. AMD’s desktop processor is most at home in a high-end workstation or client machine, such as a content creation or graphics rendering system.

Intel Xeon

The Xeon family has a distinguished heritage, beginning in 1998 with the 400 MHz Pentium II Xeon. Today’s Xeon is much more diverse, available in several different configurations, operating at a wide range of speeds, and supporting varying features. The uppermost models are very similar to their desktop Pentium 4 counterparts, manufactured on a 90 nanometer (nm) lithography process and armed with 1 MB of L2 cache memory. Of the few differences between the Xeon and Pentium 4 processor families, dual-processing support is perhaps the Xeon’s most notable attribute. Incidentally, Intel’s Xeon

CHAPTER Chaper 2 1 So Understanding You Want to Hardwre Be an in SMB the SBS Consultant?!?! Environment

family is one of the last of the modern processors to rely on its clock frequency for naming purposes.

The NetBurst micro-architecture on which the Xeon is centered has been a topic for debate since it emerged in 2000. Right from the start it was designed to be highly scalable by virtue of its incredibly long execution pipeline. And while that vision has only partially materialized, there’s little doubt that Intel is a bona fide veteran in the server and workstation space, with the experience to build solid server hardware. Though the previous incarnation of Intel’s Xeon was ill-prepared to compete with the Opteron, recent changes have infused the Xeon with much-needed horsepower.

·                       Because Intel employs a shared bus architecture, dual-processor sys­tems are sensitive to front-side bus performance. The latest Xeon chips support an 800 MHz front-side bus, enabling throughput of up to 6.4 GB per second. Of course, the shared bus isn’t an issue in single processor configurations.

·                       AMD might be planning to manufacture processors with two cores mid­way through 2005, but Intel already has technology for augmenting per­formance in particularly taxing situations. Hyper-Threading Technology enables the Xeon to execute threads in parallel, encouraging more effi­cient utilization of the processor’s execution resources. It isn’t quite the same as using two physical processors; however, Windows Server 2003 will recognize a Xeon with Hyper-Threading as two logical chips. A dual-processor setup will accordingly identify four logical processors. Bottom line—regardless of how it works, Hyper-Threading effectively improves overall responsiveness in a multi-tasked environment.

·                       Marketing folks get knocked around for unnecessarily complicating simple concepts with silly names. DBS (Demand Based Switching) is a prime example of why. Essentially SpeedStep technology in different colored wrapping paper, DBS allows the Xeon to run at lower clock speeds with a reduced operating voltage during periods of light load. The result is diminished power consumption, resulting in less thermal output, and quieter operation. The cumulative power savings is quantifiable in a


y SECTIONBrelsford1 1 SBS 2003Consuting BestDeploymentPrac


larger business environment, but most SMBs won’t notice a significant reduction in their electricity bills.

• Under extreme pressure from the AMD64 initiative, Intel reluctantly added 64-bit extensions to the Xeon family in the form of EMT64 (Extended Memory 64 Technology). The raw mechanics of Intel’s imple­mentation must be fairly similar, because 64-bit applications designed to run on the Opteron should also be compatible with Intel’s EMT64. But of course, since Windows Server 2003 is a 32-bit operating sys­tem, 64-bit computing isn’t yet an issue for SBS 2003. As I mentioned though, it’s good to have that sort of hardware flexibility should the need for 64-bit arise in the future.

There’s also a new chipset, the E7525, to complement Intel’s reworked Xeon processor. In addition to supporting Xeon’s architectural revisions, it introduces new memory and serial I/O technologies, though there’s still support for older PCI and 64-bit PCI-X technologies through the 6700PXH PCI hub. Figure 2-2 shows the four components that are required for the Xeon architecture to function properly, including the processor itself, the E7525 workstation MCH, the ICH5- R (named 82801ER), and the 6700PXH PCI hub.

Figure 2-2

The E7525 platform is designed for high-end workstations that require Xeon’s

power. Image courtesy of Intel.


CHAPTER Chaper 2 1 So Understanding You Want to Hardwre Be an in SMB the SBS Consultant?!?! Environment

The serial I/O bus used on the E7525 is called PCI Express and is designed to work around some of the bottlenecks imposed by the parallel PCI bus to which we’ve grown accustomed. For example, PCI uses a shared bus topology for communication amongst devices populating the bus. Though initially relatively simple and inexpensive, PCI gets much more complicated as you tweak it for extra performance, as many powerful server-class components tend to do. High- end 133 MHz 64-bit PCI devices (such as SCSI RAID cards) are significantly more expensive due to strict manufacturing requirements and complex board designs. Bottom line—while it delivered ample performance back in 1993, the parallel nature of today’s PCI bus limits its scalability.

In contrast, PCI Express employs a point-to-point topology, where each device is directly connected to a shared switch that routes bus traffic, and uses a serialzed approach to establish communication between connected devices. That is, rather than move data in multiple streams, PCI Express utilizes a single stream that moves much faster and is more scalable than a parallel connection. Each stream is capable of transmitting up to 2.5 gigabits per second of data in each direction at the same time. Even more impressive, up to 32 streams (referred to as lanes) can be aggregated into a single, 128 gigabit link. That’s 16 gigabytes per second of bandwidth compared to one gigabyte per second for a 64-bit, 133 MHz PCI-X device.

Those figures only illustrate the scalability of PCI Express. Intel’s E7525 chipset offers one x1 6 PCI Express connection, accommodating the latest generation of graphics cards (ideal in high-end workstations) and one configurable x8 interface that is divisible into two x4 slots. What might all of that PCI Express connectivity be used for, you ask? Well, the x1 6 slot is exclusively for use with a graphics card, and the x8 interface could, for instance, support a pair of Intel’s 8257 1EB Gigabit Ethernet controllers plus its 6700PXH PCI-X bridge for a SCSI RAID card. The permutations will undoubtedly multiply as more PCI Express devices emerge.

Intel Pentium 4

The Pentium 4 dates back to 2000, when Intel gave its micro-architecture an overhaul in order to procure more operating frequency from its processors. The Pentium 4 began its life at 1.5 GHz, a figure that has more than doubled since then, thanks to lots of work under the chip’s proverbial hood.


y SECTIONBrelsford1 1 SBS 2003Consuting BestDeploymentPrac


That doesn’t necessarily mean that the chip’s performance has multiplied. The latest incarnation of the Pentium 4, centering on a 90 nm manufacturing process, employs a longer execution pipeline, which makes it a little less efficient. But there’s also a laundry list of new features that overshadow that minor drawback.

·         The previous Pentium 4 processor came with 512 KB of L2 cache and 8 KB of the faster, more expensive L 1 data cache. The latest imple­mentation actually boasts a full 1 MB of L2 and 16 KB of L1 memory. The larger repositories are just about enough to make up for Pentium 4’s elongated pipeline.

·         One of Intel’s unquestionable strengths is the way it rallies software developers to its causes. Perhaps you’ve heard of the SSE2 instruc­tion set, supported by the original Pentium 4 and designed to enhance a number of operations that center on multimedia. On its own, SSE2 is worthless. However, applications that are properly optimized to recognize SSE2 are able to benefit from the performance increases that are inherent to the instruction set. But that’s old news; even AMD’s Athlon 64 supports SSE2. The latest Pentium 4 features SSE3, a small upgrade consisting of 13 new instructions that improve the efficiency of processing complex arithmetic, video encoding, graphics, and thread synchronization.

·         The Xeon is a server-class processor, and as such it emphasizes perfor­mance over stability. Intel’s E7525 chipset, Xeon’s core logic back­bone, consequently supports up to DDR2 memory at 400 MHz. The performance-oriented Pentium 4 goes a bit further, with sanctioned support for DDR2 memory at 533 MHz, with some motherboard manu­facturers advertising 600 MHz memory and above.


Most of the Pentium 4 features are shared in common with the Xeon. Both processors boast very similar core architectures for processing information, related value-added features such as Hyper-Threading and support for the SSE3 instruction set, and platforms with like-minded capabilities like the 800 MHz front side bus and the prominence of PCI Express.

Then there’s Intel’s reputation for reliability. While AMD has made fantastic
strides in improving its image as a corporate contender, there’s little doubt

CHAPTER Chaper 2 1 So Understanding You Want to Hardwre Be an in SMB the SBS Consultant?!?! Environment

that Intel still rules the roost. It then goes without saying that Pentium 4 client systems generally go over well, especially given the integration options of Intel’s latest chipsets.

Figure 2-3

The 925X chipset is intended for desktop workstations, enabling all of Intel’s best technology. Image courtesy of Intel.

The top-end platform is Intel’s 925X, a performance juggernaut that features a 775-pin socket to accommodate the latest Pentium 4 processors. DDR2 memory support is of course included, as is a PCI Express x1 6 slot for the latest graphics cards. According to Intel representatives, Accelerated Graphics Port (AGP) is on the outs, making PCI Express the latest in a series of new selling points.

There’s also expanded Serial ATA support through Intel’s new ICH6 controller hub that features an important second-generation feature called Native Command Queuing (NCQ). Especially in servers and workstations, NCQ improves performance by intelligently managing data requests, though it requires compatible core logic on both the controller and the hard drive. One variant of the ICH6, bearing a –R suffix, adds RAID 0 and RAID 1 support


y SECTIONBrelsford1 1 MBSBS 2003Consuting BestDeploymentPrac


for either enhanced performance or extra I/O security. Finally, the ICH6-W features wireless networking, though it requires an add-on PCI card to expose wireless functionality.

Processor Summary

At the end of the day, there are two principal products for servers—AMD Opteron and Intel Xeon—and two processors that are more oriented for client machines— AMD Athlon 64 and Intel Pentium 4. Now, a Pentium 4 works just fine in server environments, and an Opteron would perform superbly in a graphics workstation, so don’t feel constrained by each manufacturer’s target applications. Just keep in mind that a Pentium 4 server and its complementary platform won’t sport the same manageability and reliability features as a higher-end Xeon machine. At the same time though, smaller businesses won’t always tax the Pentium 4 anyway.

When you choose the processor for your SBS 2003 server, keep three things in mind: budget, expected load, and headroom. Of course, you should only buy what you can afford, so don’t go over the top if it isn’t necessary. Further, businesses with only 20 client systems won’t need the raw horsepower that shops three times that size would require. But if you project significant growth in the next five years, it’s easier to buy the better hardware today rather than have to replace your server a few years down the road.


Harry Brelsford, CEO at SMB Nation (www.smbnation.com)

MBA, MCSE, CNE, CLSE, CNP, MCP, MCT, SBSC (Microsoft Small Business Specialist)

PS – did you know my Windows Small Business Server 2008 (SBS 2008) book is almost here? Yes!


Leave a comment

Filed under Book

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s