Microsoft® Exchange Server 2003 Scalability with SP1 and SP2

4.2 System Processing 187

Chapter 4

These are two examples of choices that you will be confronted with, and

many others will come up, undoubtedly. This chapter covers some of these

critical technologies for Exchange 2003 and provides some useful recom-

mendations. Nevertheless, even in your server and systems architecture

model, you will need to justify the need for what you plan to acquire

upfront and proactively make sure that you make the best use of the tech-

nology that the financial department and the business units let you acquire.

This proactive monitoring can be done by the means of third-party tools

or by simply using Windows Server out-of-the-box utilities such as SysMon

or PerfMon, and proper analysis of meaningful data. This aspect is covered

later on in the book.

4.2 System Processing

Above all, technology has evolved in an unbalanced manner over time. Cer-

tainly computing processing is following Moore’s law, and doubles almost

every 18 months; 2005 brought more on this trend with the introduction

of dual-core processors, and we are now looking at quad-core processors

being delivered from the key industry players that Intel and AMD are.

Exchange 2003 is a great application for taking advantage of processing

capabilities, being used with either fast processors or multiple processors, or

processors with advanced parallelism.

Among deployments already observed, the sweet spot for Exchange

2003 was in 4-processor systems. Scaling up (i.e., adding more processing

capability) often does not help due to the overhead of context switches and

contention at the system bus level. From a system processing viewpoint and

for Exchange 2003 in particular, I propose you review some of the key sys-

tem components and attributes, such that you can actually realize which of

these technologies matter to Microsoft Exchange.

4.2.1 Systems architectures: Processors and buses

Microsoft Exchange 2003 largely benefits from the increasing speed of pro-

cessors provided by Intel and other chip manufacturers such as AMD.

Across the years, the “megahertz war” transformed into a “gigahertz war.” As

found often in desktop computing, you will benefit from fast clock rates.

However, there are variants in industry standard processor architectures that

offer better improvement than just a high core processor clock rate. Today,

industry standard processors, provided, for the majority, by Intel and AMD

come with varying attributes beyond the simple clock rate (the GHz figure),

188 4.2 System Processing

which include inner components such as cache, hyper-threading, inner and

outer clock rates, and addressing/computing capabilities (32-bit vs. 64-bit).

One particular advantage of processors that run slower is that they demand

less power and as a result, dissipate less heat, significantly decreasing the

cooling requirements for the computing platform and environment.

Beyond the processor, which you might consider as a black-box, you

need to understand how well does the processor communicate with the

outer parts of the core server and motherboard that equips a server. These

parts include:



memory bus



network and storage interfaces (I/O)



peer-processor communication

This point is critical to understand, because no matter how fast a proces-

sor can execute instructions, if the data does not come in and out fast

enough, all you obtain is an idle computing system that keeps on waiting

on slow interfaces. The objective, therefore, does not only consist in using

powerful computing units, it also aims at providing data fast enough to

keep the CPU busy.

For example, processors can operate beyond 3-GHz, based on a system

bus architecture that operates four times as fast as the current system

buses that normally operate between 400 MHz and 533 MHz. Further-

more, newer system models take advantage of PCI-Express, a serial evolu-

tion from the PCI/PCI-X parallel architecture. This brings a new

standard for high-speed bus access, joining the more traditional PCI/PCI-

X and AGP bus technologies together into a much faster transport, and

opening the opportunity to advanced CPU communication, possibly

amongst other peer CPU.

Initially co-founded between Intel and Dell, PCI-Express has an oppor-

tunity for scaling-up architectures that use PCI-Express serial interconnect

for gathering several 2-way or 4-way servers; the catch is how well the

underlying operating system can take advantage of this underlying capabil-

ity, and eventually, the application.

Figure 4.1 shows a flow diagram of data and basic components interac-

tion of an industry standard server. Each of these components can indeed

operate at varying clock speeds. Although a single component of the