Accurately comparing the costs of compatible Windows- and Unix-based servers
can be an arduous task, especially since Unix and Windows systems do not offer
the same degree of scalability, performance, functionality, availability or
reliability, even for similarly configured servers. For example, the I/O,
backplane, complete error correcting code (ECC), component redundancy, hot-swapability,
dynamic reconfiguration and self-healing technologies that are often found in
enterprise-class Unix servers are typically not available on Windows-based
servers.
Customers often compare servers by number of CPUs, performance or cost when
making a purchasing decision. One problem with straight CPU-to-CPU price
comparisons between Windows and Unix servers is that, in general, nonclustered
Windows-based servers have trouble scaling beyond eight CPUs (with the exception
of the Unisys ES7000 server, which is scalable to 32 CPUs), whereas Unix
systems easily scale to 72 CPUs and beyond. The performance benchmarks in Table
1 provide the average current tpc-C performance ranges for nonclustered servers
by platform for the tiered classifications of Unix and Windows servers.
Table 1: Relative tpc-C Nonclustered Performance Benchmarks
(performance rates are in tpc-C throughput)
Winsows based servers |
Unix-RISC based servers |
||
Server type |
Avg. performance |
Server type |
Avg. performance |
Ultra-high(9-32 CPUs) |
71,000-170,000 |
Ultra-high(32+CPUs) |
135,000+ |
High-end(8 CPUs) |
36,000-70,000 |
High-end(16-31 CPUs) |
111,000-230,000 |
Mid-range(3-7 CPUs) |
16,000-35,000 |
Mid-range(5-15 CPUs) |
51,000-110,000 |
Entry-level(1-2 CPUs) |
9,000-15,000 |
Entry-level(1-4 CPUs) |
20,000-50,000 |
Source: Transaction Processing Performance Council, www.tpc.org,
and Giga Information Group
Evaluating performance and scalability
The server classifications provided in Table I should not be confused with
compatible servers. Due to the higher degree of scalability for Unix servers,
the classifications within Windows and Unix servers differ significantly. For
example, a one- to four-way RISC-based system would be considered an entry-level
Unix server, while an entry-level WinTel system typically only consists of one
to two CPUs, with a four-way WinTel server most likely categorized as a midrange
Windows server.
The intent of Table 1 is to show the difference between the current
performance levels and the classifications of Windows- and Unix-based servers.
There are no set standards for defining how to categorize a server, so the data
provided in Tables 1, 2 and 3 is based on the professional opinion of Giga
analysts using information from clients, server vendors, ISVs and performance
benchmarks.
The overall server classifications should be determined based on more than CPU
count and type (i.e., performance, workload, cost, etc.). However, should a
company need to compare systems based solely on CPUs, a "rule of
thumb" that can be used for general server classification is provided in
Table 2. Using the information in Table 2, users could infer that a one-to-two
CPU Unix RISC-based server is reasonably comparable to a one-to-four CPU
Intel-based server.
Table 2: "Rule of Thumb" Server Classification
Server configurations |
Unix RISC-Based servers |
Windows Intel(CISC)-based servers |
Tier 1- Workgroup |
1-2 processors |
1-4 processors |
Tier 2- Departmental |
3-4 processors |
5-8 processors |
Tier 3-Entry-level enterprise |
5-8 processors |
9-16 processors |
Tier 4-Midrange enterprise |
9-24 processors |
17-32 processors |
Tier 5- High-end enterprise |
25+ processors |
33+ processors |
Source: Giga Information Group
Note: Actual classifications will vary by vendor
Evaluating cost
There are numerous factors to consider when determining the total cost of a
server, such as the cost of hardware, software, support (including installation,
maintenance, training, etc. for internal and external operations), power
consumption, data center space, the impact of downtime, the average product life
cycle, and the ability to scale or meet application and availability
requirements. Additional Giga research will address the topic of evaluating the
total cost of server ownership in more detail. Table 3 focuses on providing a
general reference that evaluates the average differences in cost between Unix
and Windows servers and compares the hardware costs to the benefits of
scalability and performance.
Table 3: Comparison of Scalability, Performance and Cost of Unix to Windows
Servers
Server configurations |
Unix potential Scalability relative to |
Unix Performance relative to Windows system |
Unix HW Cost relative to Windows system |
Windows: 9-32 CPUs |
Ultra-high Unix servers are 2x-3x more scalable |
1.5x-4x performance |
3x-5x cost |
Windows &Unix servers with 8 CPUs |
High-end Unix servers are 2x-4x more scalable |
2x-3x performance |
2x-5x cost |
Windows & Unix servers with 6 CPUs |
Midrange Unix servers are 2x-3x more scalable |
2x-3x performance |
1.5x-4x cost |
Windows & Unix servers with 4CPUs |
Entry-level Unix servers are 2x more scalable |
1.5x-3x performance |
1x-3x cost |
Source: Giga Information Group
According to the information provided in Table 3, on average, a Unix server with
eight CPUs is about two to five times as expensive as a Windows server with
eight CPUs, but it has approximately two to three times the performance and two
to four times the potential scalability. Unix systems tend to excel in
scalability, manageability, flexibility and performance, particularly beyond
eight CPUs; whereas the strength of Windows systems is the price/performance
advantage. Unix systems are usually more expensive than their Windows
counterparts, but many Unix server vendors are beginning to scale down the
functionality of Unix systems to provide a more cost competitive product (i.e., Sun’s
new V880, Compaq’s AlphaServer ES40, Hewlett-Packard’s L2000
and IBM’s p610 servers).
There are advantages to both Unix and Windows-based servers. On average,
Windows-based servers are available at a lower cost and have competitive
performance benchmarks, making them suitable for most noncritical, front-end and
small-to-midsize applications. The scalability, performance throughput and
availability of Unix servers continue to make them the preferred platform for
larger, complex, back-end applications.