www.kingston.com/europe/server
Memory performance testing Research sponsored by Kingston Technology Europe and conducted by Mindcraft Labs
www.kingston.com/europe./server
Contents 1.0
Executive summary ...............................................................................................................3 Memory matters ..........................................................................................................................3
2.0
Background.............................................................................................................................4 More for less ................................................................................................................................4 Improving server life expectancy ...............................................................................................4
3.0
Objectives...............................................................................................................................5 Testing the evidence....................................................................................................................5
4.0
Web server test results.......................................................................................................6 Microsoft Windows 2000 Advanced Server..............................................................................6 Sun Solaris 9.0.............................................................................................................................8 Red Hat Linux 8.0 .....................................................................................................................10
5.0
Database Management System (DBMS) server results .............................................. 12 Microsoft Windows 2000 Advanced Server............................................................................12 Sun Solaris .................................................................................................................................14 Red Hat Linux 8.0 .....................................................................................................................16
6.0
Conclusion........................................................................................................................... 18 Closing the Gaps .......................................................................................................................18 Breaking the Price Performance Barrier ..................................................................................18
7.0
Appendix 1: Methodology ................................................................................................ 19 Test Server Configuration.........................................................................................................19
8.0
Appendix 2: Summary of Results .................................................................................... 21 SPECweb Analysis.....................................................................................................................21 OSDB Analysis...........................................................................................................................22
9.0
Appendix 3: Company information ................................................................................ 24 About Kingston Technology.....................................................................................................24 About Mindcraft ........................................................................................................................24
10.0
Appendix 3: Legal notice .................................................................................................. 25
2
www.kingston.com/europe./server
1.0 Executive summary Memory matters Businesses are constantly searching for technology solutions that offer better return on investment (ROI) and lower total cost of ownership (TCO). As line-of-business and web server applications continue to evolve and become more demanding of the hardware platforms on which they sit, the challenge facing IT departments is increasingly one of greater optimisation, squeezing every ounce of performance from a server for the lowest possible cost while increasing the period between server upgrade and replacement cycles. Research from Mindcraft Labs has revealed that adding memory rather than processors commonly represents a more cost-effective solution to the challenge of improving web server and DBMS server performance across operating platforms.
3
www.kingston.com/europe./server
2.0 Background More for less Computer Weekly has reported that since the beginning of 2001, businesses have constantly re-evaluated the size of IT budgets and have been exploring ways of leveraging their IT infrastructure in order to both reduce capital costs and to extend the working life of servers as far into the future as possible. Among larger companies, server consolidation and thin-client implementations have proved popular as cost-reduction measures but invariably these steps lead to a demand for more powerful servers, a consequence that challenges the ingenuity of the IT department in delivering better levels of performance from existing systems.
Improving server life expectancy In many examples, organisations simply purchase new servers rather than upgrade existing systems and with the arrival of Windows Server 2003 many companies, encouraged by leading server vendors, may choose to upgrade their servers in parallel with the upgrade to the new operating system. Purchasing new systems is rarely the most cost effective solution whether if be for improving applications performance or running a new operating system or even adding more users to the network, when a low-cost memory upgrade may offer a better return on investment.
4
www.kingston.com/europe./server
3.0 Objectives Testing the evidence To better understand the relationship between memory, cost of ownership and server performance, Kingston Technology Europe commissioned Mindcraft, an independent testing laboratory, to perform quantitative research. The objective of this research was to reveal how server performance might improve incrementally by adding more memory as an alternative to the installation of additional processors and two specific benchmark questions were put to the testing laboratory: 1.
How much does web server performance improve with additional memory?
2.
How much better does a database management system (DBMS) server perform with additional memory?
In conducting its research, Mindcraft discovered that the performance of Windows 2000 ™ Solaris ™ and Linux servers showed significant improved through the addition of incremental memory, leading to performance benefits of as much as 40% over servers with an additional processor and no increase in memory. The performance improvement could be as high as 1000% on dual processor Solaris DBMS systems with the memory expanded from 512Mb to 4 GB.
5
www.kingston.com/europe./server
4.0 Web server test results Microsoft Windows 2000 Advanced Server Single processor systems Mindcraft’s research clearly illustrates the benefits extra memory brings to Windows 2000 servers. For one-CPU systems, doubling the available memory from 512MB to 1GB gives 31% improvement in performance and is 26% faster than adding another CPU. Quadrupling the memory in a Windows 2000 server from 512 MB to 2GB is 40% more effective than simply adding another processor. Table 1: Percentage performance improvement with increased memory – 1 CPU (Windows 2000 Server)
Increase memory to Base memory size
1GB
2GB
4GB
512MB
31%
46%
48%
11%
13%
1GB Figure 1
512MB Base Memory (1 CPU) 60 50 40
46
48
2GB
4GB
31
% 30 20 10 0 1GB
6
www.kingston.com/europe./server
Dual processor systems Table 2: Percentage performance improvement with increased memory 2 CPUs (Windows 2000 Server) Increase memory to Base memory size
1GB
2GB
4GB
512MB
37%
76%
90%
29%
30%
1GB
As the results in Table 2 illustrate, in two-CPU Windows 2000 systems, doubling the memory from 512MB to 1GB increases performance by 37% and doubling the memory again from 1GB to 2GB improves performance by 29% Figure 2 512MB Base Memory (2 CPU) 100
90 76
80 60 % 40
37
20 0 1GB
2GB
4GB
7
www.kingston.com/europe./server
Sun Solaris 9.0 Single processor systems In single CPU Systems, doubling the memory from 1GB to 2GB gives 51% better performance while quadrupling memory to 4GB improves performance by 63%. Table 3: Percentage performance improvement with increased memory – 1CPU (Sun Solaris) Increase memory to Base memory size
2GB
4GB
1GB
51%
63%
2GB
63%
Figure 3: Percentage performance improvement with increased memory - 1CPU (Sun Solaris) 1GB Base Memory (1 CPU) Sun Solaris 70 60
63 51
50 %
40 30 20 10 0 2GB
4GB
Dual processor systems For two CPU Systems, doubling the memory from 1GB to 2GB increases performance by 66% and quadrupling the memory to 4GB improves performance by 82%. Table 4: percentage performance improvement with increased memory - 2CPUs (Sun Solaris) Increase memory to Base memory size
2GB
4GB
1GB
66%
82%
2GB
10%
8
www.kingston.com/europe./server
Figure 4 1GB Base Memory (1 CPU) Sun Solaris 70 60
63 51
50 %
40 30 20 10 0 2GB
4GB
9
www.kingston.com/europe./server
Red Hat Linux 8.0 Single processor systems Linux shows more modest performance improvements than Solaris 9.0 and with single processor systems, the results are closer to those of Windows 2000. Doubling the memory from 512 MB to 1GB gives 33% better performance. This is 30% faster than adding another CPU while quadrupling memory to 2GB is 45 % better than adding another CPU. Table 5: percentage performance improvement with increased memory - 1CPU (Linux) Increase memory to Base memory size
1GB
2GB
4GB
512MB
33%
48%
62%
11%
21%
1GB
512MB Base Memory (1 CPU) Linux 70
62
60 48
50 %
40
33
30 20 10 0 1GB
2GB
4GB
Figure 5
10
www.kingston.com/europe./server
Dual processor systems For two CPU Systems, doubling the memory from 1GB to 2GB increases performance by 66% and quadrupling the memory to 4GB improves performance by 82% Table 6: percentage performance improvement with increased memory - 2CPUs (Linux) Increase memory to Base memory size
1GB
2GB
4GB
512MB
53%
102%
125%
32%
47%
1GB
In two-CPU Linux systems, doubling the memory from 512 MB to 1GB returns a performance increase of 53% and quadrupling memory to 2GB raises performance by 47%. Doubling the memory again from 1GB to 2 GB improves performance by 32% Increasing the base memory from 512MB to 4GB in a two processor Linux server will increase performance by 125%. 512MB Base Memory (2 CPUs) Linux 130 120 110 100 90 80 % 70 60 50 40 30 20 10 0
125 102
53
1GB
2GB
4GB
Figure 6
11
www.kingston.com/europe./server
5.0 Database Management System (DBMS) server results For database management systems, having more memory means that the operating system and DBMS can keep more data in main memory thereby saving relatively slow disk accesses.
Microsoft Windows 2000 Advanced Server Single processor systems For one-CPU Systems: Doubling the memory from 512MB to 1GB gives 9% better performance and is 21% faster than adding another CPU while quadrupling the memory to 2GB is 939% better than adding another CPU*. Table 7: percentage performance improvement with increased memory - 1CPU (Windows 2002 Advanced Server) Increase memory to Base memory size
1GB
2GB
4GB
512MB
9%
837%
1458%
762%
1334%
1GB
512MB Base Memory (1 CPU) Windows 2000 Server 70 60
63 51
50 %
40 30 20 10 0 1GB
2GB
Figure 7
*
See appendix 2
12
www.kingston.com/europe./server
Dual processor systems In two-CPU systems, doubling the memory from 512MB to 1GB increases performance by 24% and doubling the memory again from 1GB to 2GB improves performance by 1202% Table 8: percentage performance improvement with increased memory – 2CPUs (Windows 2000 Server) Increase memory to Base memory size
1GB
2GB
4GB
512MB
24%
1510%
1559%
1202%
1241%
1GB
512MB Base Memory (2 CPU) Windows 2000 Server 1600
1510
1559
2GB
4GB
1400 1200 1000 % 800 600 400 200
24
0 1GB
Figure 8
13
www.kingston.com/europe./server
Sun Solaris In the Sun Solaris environment, adding extra memory produces remarkable performance improvements.
Single processor systems For single CPU systems, doubling the memory from 1GB to 2GB gives a 376% improvement in performance while quadrupling memory to 4GB improves performance by 439%. Table 9: percentage performance improvement with increased memory - 1CPU (Sun Solaris) Increase memory to Base memory size 1GB
2GB
4GB
376%
439%
2GB
439% 1GB Base Memory (1 CPU) Sun Solaris 500 400
439 376
300 % 200 100 0 2GB
4GB
Figure 9
14
www.kingston.com/europe./server
Dual processor systems Table 10: percentage performance improvement with increased memory2CPUs (Sun Solaris) Increase memory to Base memory size 1GB
2GB
4GB
775%
1010%
2GB
27%
In dual-processor Solaris systems, expanding the memory from 1GB to 2GB increases performance by 775% and quadrupling the memory to 4GB improves performance by 1010% 1GB Base Memory (2 CPU) Sun Solaris 1100 1000 900 800 700 600 % 500 400 300 200 100 0
1010 775
2GB
4GB
Figure 10
15
www.kingston.com/europe./server
Red Hat Linux 8.0 Single processor systems For one CPU systems, doubling the memory from 512 MB to 1GB gives 50% better performance and is 55% faster than adding another CPU while quadrupling memory to 2GB is 1018 % better than adding another CPU.* Table 11: percentage performance improvement with increased memory - 1CPU (Linux) Increase memory to Base memory size
1GB
2GB
4GB
512MB
50%
981%
1243%
620%
795%
1GB
512MB Base Memory (1 CPU) Linux 1500 1243
1250 981 1000 % 750 500 250
50
0 1GB
2GB
4GB
Figure 11
*
See appendix 2
16
www.kingston.com/europe./server
Dual processor systems In two CPU Systems, doubling the memory from 512 MB to 1GB increases performance by 52% and quadrupling memory to 2GB raises performance by 1357%. Doubling the memory from 1GB to 2 GB improves performance by 858%. Table 12: percentage performance improvement with increased memory - 2CPUs (Linux) Increase memory to Base memory size
1GB
2GB
4GB
512MB
52%
1357%
1477%
858%
937%
1GB
Increasing the base memory from 512MB to 4GB will increase performance by 1477%. 512MB Base Memory (2 CPU) Linux 1500
1357
1477
1250 1000 % 750 500 250
52
0 1GB
2GB
4GB
Figure 12
17
www.kingston.com/europe./server
6.0 Conclusion Closing the Gaps The Mindcraft research focused on the performance of web server and DBMS servers, because applications response times in both of these areas are among the most visible to the end user. Nobody likes having to work with a slow web server and where database applications are involved, the benchmark results clearly show that more memory available means that the Operating System and DBMS can keep more data in main memory, eliminating or reducing the frequency of relatively slow disk accesses, which in turn lead to slower applications response times.
Breaking the Price Performance Barrier One objective of the research was to establish a ‘test-point’, which could aid customers in determining the best and most cost-effective configuration for their servers. Many companies buy servers with a dual processor capability but only a relatively small percentage of owners actually choose to add a second CPU or more memory, rather than buy a completely new server when circumstances demand greater performance. The results clearly indicate that adding memory first, as a rule of thumb, offers the best solution until an application’s need for memory is completely satisfied. Once this objective has been achieved, all-important buffers and needed tables and data are present in the application's working set, adding further processors is a next logical step. Therefore, one should ensure that systems are fully memory efficient first before adding further processors or considering other upgrade options. The research results clearly illustrate the performance benefits achievable through simply adding extra memory to servers on the three most popular Operating Systems. Better performance at lower cost can also lead to greater productivity through faster application response. As a result of this information provided by Mindcraft Labs, businesses should now be able to budget with greater confidence, knowing that an investment in server memory is, in most cases, the most immediate and cost-effective means of optimising systems which might previously have been thought of as requiring extra processors.
18
www.kingston.com/europe./server
7.0 Appendix 1: Methodology Test Server Configuration For testing purposes, Mindcraft selected a HP ProLiant DL380 G3. This system uses 2.4GHz Xeon CPUs, and can be configured with one and two processors. The system was tested with 512MB of memory, and was expanded up to 4GB of memory. The machine was set up with Microsoft Windows 2000 Advanced Server and Red Hat Linux 8.0 For Solaris 9 Mindcraft selected a Sun Fire 280R configured with 1GB, 2GB and 4GB of memory.
Web Server Testing A web server houses HTML files that transfer to a client in response to HTTP requests. It can also use applications to generate HTML code. Mindcraft used SPECweb99 to test web server performance. We used Microsoft's IIS (Internet Information Server) 5.0 web server software. SPECweb99 measures a web server's performance while servicing a specific workload of static and dynamic page requests. Static requests return data stored in files while dynamic requests use a program to generate the data that is returned. This workload is intended to simulate the type of requests that an Internet Service Provider (ISP) would encounter on its web servers. As discussed below, even though SPECweb99 is focused at ISPs, the results may actually be more relevant to an enterprise's dedicated web servers.
DBMS Server Testing Mindcraft used the OSDB (Open Source Database Benchmark) test suite. OSDB has single-user tests that focus on database loading and index creation functions, as well as basic query access speed. Its multi-user tests model workloads, using simulated users to measure the performance of common information retrieval and online transaction processing (OLTP) environments. The Open Source Database Benchmark (OSDB) is based on the ANSI SQL Standard Scalable and Portable (AS3AP) benchmark for relational database systems. AS3AP was designed to: Provide a comprehensive, manageable set of tests for measuring database performance: •
Be scalable and portable to facilitate testing a broad range of systems;
•·
Minimize human interaction to implement and to run the benchmark tests.
•·
Create a uniform performance metric capable of unambiguous interpretation.
The OSDB implements much of the AS3AP benchmark. It implements both singleuser and multi-user tests for several types of data base management systems (DBMSs). The single-user tests focus on database loading and index creation functions. They also measure basic query and access method performance.
19
www.kingston.com/europe./server
The OSDB multi-user tests model four types of workloads using simulated foreground and background users: 1.
Information retrieval (IR), which tests all simulated users presenting queries for information on a single row using indexed values.
2.
IR Cross Section, where a foreground simulated user, whose performance is measured, executes a set of retrieval queries and updates while background simulated users execute the same IR query as in the first workload.
3.
On-line transaction processing (OLTP), which tests query performance while updating the database. All background-simulated users execute a single-row update on random rows in the same table while a foreground-simulated user makes the same IR queries as in the first workload.
4.
OLTP Cross Section, where a foreground simulated user, whose performance is measured, executes a set of retrieval queries and updates while the background simulated users execute the same OLTP queries and updates as in the third workload.
For all of the Kingston tests, Mindcraft used a database with 2,000,000 tuples (rows) per relation. This yielded a logical database size of 2.26 gigabytes. Mindcraft selected this size database because it could be created in 2.5 hours, could be saved and restored in under 20 minutes, and yielded an actual database that required over 6 GB of disk storage, which meant that it would not all fit in memory in the configurations tested. For the Kingston tests, Mindcraft simulated 100 active users in the background for all tests. This allowed a comparison of how different systems and configurations perform for the same user base accessing the same database. Mindcraft ran only the multi-user tests for Kingston. For the Kingston Technology tests, Mindcraft used the MySQL DBMS with InnoDBtype tables.
20
www.kingston.com/europe./server
8.0 Appendix 2: Summary of Results SPECweb Analysis SPECweb99
System/OS/Web Server HP ProLiant DL380 G3 1 x 2.4 GHz Xeon HP ProLiant DL380 G3 1 x 2.4 GHz Xeon HP ProLiant DL380 G3 1 x 2.4 GHz Xeon HP ProLiant DL380 G3 1 x 2.4 GHz Xeon HP ProLiant DL380 G3 2 x 2.4 GHz Xeon HP ProLiant DL380 G3 2 x 2.4 GHz Xeon HP ProLiant DL380 G3 2 x 2.4 GHz Xeon HP ProLiant DL380 G3 2 x 2.4 GHz Xeon HP ProLiant DL380 G3 1 x 2.4 GHz Xeon HP ProLiant DL380 G3 1 x 2.4 GHz Xeon HP ProLiant DL380 G3 1 x 2.4 GHz Xeon HP ProLiant DL380 G3 1 x 2.4 GHz Xeon HP ProLiant DL380 G3 2 x 2.4 GHz Xeon HP ProLiant DL380 G3 2 x 2.4 GHz Xeon HP ProLiant DL380 G3 2 x 2.4 GHz Xeon HP ProLiant DL380 G3 2 x 2.4 GHz Xeon Sun Sun Fire 280R 1 x 900 MHz UltraSparc III Sun Sun Fire 280R 1 x 900 MHz UltraSparc III Sun Sun Fire 280R 1 x 900 MHz UltraSparc III Sun Sun Fire 280R 2 x 900 MHz UltraSparc III Sun Sun Fire 280R 2 x 900 MHz UltraSparc III Sun Sun Fire 280R 2 x 900 MHz UltraSparc III
Memory Operating System Size (GB) # of CPUs Windows 2000 0.5 1 Advanced Server Windows 2000 1 1 Advanced Server Windows 2000 2 1 Advanced Server Windows 2000 4 1 Advanced Server Windows 2000 0.5 2 Advanced Server Windows 2000 1 2 Advanced Server Windows 2000 2 2 Advanced Server Windows 2000 4 2 Advanced Server
% Better than previous memory size
Performance improvement from lowest memory size
1032
30.80%
31%
1149
11.34%
SPECweb99
Performance improvement from second lowest memory size
% 2 x RAM Better than 1 more CPU
% 4 x RAM Better than 1 more CPU
789
1165
26%
46%
11%
48%
13%
2.59%
40% 4.02%
820 1120
36.59%
37%
1447
29.20%
76%
29%
1562
7.95%
90%
39%
Red Hat Linux 8.0
0.5
1
675
Red Hat Linux 8.0
1
1
900
33.33%
33%
Red Hat Linux 8.0
2
1
1002
11.33%
48%
11%
45%
Red Hat Linux 8.0
4
1
1092
8.98%
62%
21%
3.02%
Red Hat Linux 8.0
0.5
2
693
Red Hat Linux 8.0
1
2
1060
52.96%
53%
Red Hat Linux 8.0
2
2
1401
32.17%
102%
32%
Red Hat Linux 8.0
4
2
1559
11.28%
125%
47%
Solaris 9
1
1
384
Solaris 9
2
1
579
50.78%
51%
7.77%
63%
Solaris 9
4
1
624
Solaris 9
1
2
603
Solaris 9
2
2
1002
66.17%
66%
Solaris 9
4
2
1098
9.58%
82%
30%
63%
3.48%
10%
21
www.kingston.com/europe./server
OSDB Analysis Multi-user information retrieval Database OSDB Results
Performance improvement from second lowest memory % 2 x RAM Better than % 4 x RAM Better than size 1 more CPU 1 more CPU
Operating System
HP ProLiant DL380 G3 1 x 2.4 GHz Xeon
Windows 2000 Advanced Server
0.5
1
Windows 2000 Advanced Server
1
1
3.87
9%
9%
Windows 2000 Advanced Server
2
1
33.35
762%
837%
762%
740.05%
938.94%
Windows 2000 Advanced Server
4
1
55.48
66%
1458%
1334%
7.33%
1297.48%
Windows 2000 Advanced Server
0.5
2
3.21
Windows 2000 Advanced Server
1
2
3.97
24%
24%
Windows 2000 Advanced Server
2
2
51.69
1202%
1510%
1202%
Windows 2000 Advanced Server
4
2
53.24
3%
1559%
1241%
Red Hat Linux 8.0
0.5
1
4.21
Red Hat Linux 8.0
1
1
6.32
50%
50%
Red Hat Linux 8.0
2
1
45.49
620%
981%
620%
24%
1243%
795%
HP ProLiant DL380 G3 2 x 2.4 GHz Xeon HP ProLiant DL380 G3 1 x 2.4 GHz Xeon HP ProLiant DL380 G3 1 x 2.4 GHz Xeon HP ProLiant DL380 G3 1 x 2.4 GHz Xeon HP ProLiant DL380 G3 1 x 2.4 GHz Xeon HP ProLiant DL380 G3 2 x 2.4 GHz Xeon HP ProLiant DL380 G3 2 x 2.4 GHz Xeon HP ProLiant DL380 G3 2 x 2.4 GHz Xeon HP ProLiant DL380 G3 2 x 2.4 GHz Xeon Sun Sun Fire 280R 1 x 900 MHz UltraSparc III Sun Sun Fire 280R 1 x 900 MHz UltraSparc III Sun Sun Fire 280R 1 x 900 MHz UltraSparc III Sun Sun Fire 280R 2 x 900 MHz UltraSparc III Sun Sun Fire 280R 2 x 900 MHz UltraSparc III Sun Sun Fire 280R 2 x 900 MHz UltraSparc III
IR
Performance improvement from lowest memory size
System/OS/Web Server
HP ProLiant DL380 G3 1 x 2.4 GHz Xeon HP ProLiant DL380 G3 1 x 2.4 GHz Xeon HP ProLiant DL380 G3 1 x 2.4 GHz Xeon HP ProLiant DL380 G3 2 x 2.4 GHz Xeon HP ProLiant DL380 G3 2 x 2.4 GHz Xeon HP ProLiant DL380 G3 2 x 2.4 GHz Xeon
Memory Size (GB) # of CPUs
% Better Than Previous Memory Size
3.56 20.56%
55.28%
Red Hat Linux 8.0
4
1
56.56
Red Hat Linux 8.0
0.5
2
4.07
Red Hat Linux 8.0
1
2
6.19
52%
52%
Red Hat Linux 8.0
2
2
59.31
858%
1357%
858%
8%
1477%
937%
Red Hat Linux 8.0
4
2
64.17
Solaris 9
1
1
3.24
Solaris 9
2
1
15.41
376%
376%
Solaris 9
4
1
17.46
13%
439%
Solaris 9
1
2
2.88
Solaris 9
2
2
25.19
775%
775%
Solaris 9
4
2
31.96
27%
1010%
634.89%
1017.69% 813.73%
435.07% 439%
506.25%
27%
22
www.kingston.com/europe./server
Multi-user on-line transaction processing Database OSDB Results
System/OS/Web Server
Operating System
HP ProLiant DL380 G3 1 x 2.4 GHz Xeon
Windows 2000 Advanced Server
HP ProLiant DL380 G3 1 x 2.4 GHz Xeon HP ProLiant DL380 G3 1 x 2.4 GHz Xeon HP ProLiant DL380 G3 1 x 2.4 GHz Xeon HP ProLiant DL380 G3 2 x 2.4 GHz Xeon HP ProLiant DL380 G3 2 x 2.4 GHz Xeon HP ProLiant DL380 G3 2 x 2.4 GHz Xeon HP ProLiant DL380 G3 2 x 2.4 GHz Xeon HP ProLiant DL380 G3 1 x 2.4 GHz Xeon HP ProLiant DL380 G3 1 x 2.4 GHz Xeon HP ProLiant DL380 G3 1 x 2.4 GHz Xeon HP ProLiant DL380 G3 1 x 2.4 GHz Xeon HP ProLiant DL380 G3 2 x 2.4 GHz Xeon HP ProLiant DL380 G3 2 x 2.4 GHz Xeon HP ProLiant DL380 G3 2 x 2.4 GHz Xeon HP ProLiant DL380 G3 2 x 2.4 GHz Xeon Sun Sun Fire 280R 1 x 900 MHz UltraSparc III Sun Sun Fire 280R 1 x 900 MHz UltraSparc III Sun Sun Fire 280R 1 x 900 MHz UltraSparc III Sun Sun Fire 280R 2 x 900 MHz UltraSparc III Sun Sun Fire 280R 2 x 900 MHz UltraSparc III Sun Sun Fire 280R 2 x 900 MHz UltraSparc III
Memory Size (GB) # of CPUs 0.5
1
OLTP
% Better Than Previous Memory Size
Performance improvement from lowest memory size
Performance improvement from second lowest memory size
% 2 x RAM Better % 4 x RAM Better than 1 more CPU than 1 more CPU
0.60
Windows 2000 Advanced Server
1
1
0.61
2%
2%
Windows 2000 Advanced Server
2
1
0.75
22.95%
25%
23%
34%
15% 42%
Windows 2000 Advanced Server
4
1
0.90
20.00%
50%
48%
1.12%
61%
Windows 2000 Advanced Server
0.5
2
0.53
Windows 2000 Advanced Server
1
2
0.56
6%
6%
Windows 2000 Advanced Server
2
2
0.89
59%
68%
59%
Windows 2000 Advanced Server
4
2
0.90
1%
70%
61%
Red Hat Linux 8.0
0.5
1
0.46
Red Hat Linux 8.0
1
1
0.49
7%
7%
Red Hat Linux 8.0
2
1
0.65
33%
41%
33%
35%
48%
Red Hat Linux 8.0
4
1
0.73
12%
59%
49%
-31.78%
52%
Red Hat Linux 8.0
0.5
2
0.44
Red Hat Linux 8.0
1
2
0.48
9%
9%
123%
Red Hat Linux 8.0
2
2
1.07
Red Hat Linux 8.0
4
2
1.02
Solaris 9
1
1
0.43
Solaris 9
2
1
1.41
228%
228%
Solaris 9
4
1
1.45
3%
237%
11%
143%
123%
132%
113%
Solaris 9
1
2
0.47
Solaris 9
2
2
1.15
145%
145%
Solaris 9
4
2
1.36
18%
189%
200% 237%
26%
18%
23
209%
www.kingston.com/europe./server
9.0 Appendix 3: Company information About Kingston Technology Kingston Technology Europe Ltd. is the world's largest independent manufacturer of memory products. Kingston operates manufacturing facilities in Malaysia, Taiwan, China and Fountain Valley, California, including Payton Technology Corp., Kingston's back-end processing facility supporting memory packaging, test and logistics. With the advent of Payton, Kingston supports all memory processing functions from receipt of wafer to completed module. Kingston serves a network of distributors, OEMs, and retail customers in more than 3,000 locations worldwide. For more information on Kingston, call 0800 8888 0101 or visit the Kingston Technology at http://www.kingston.com/europe
About Mindcraft Mindcraft is a service-oriented, independent test laboratory. Mindcraft has been providing high-quality software, systems, and network products. Mindcraft produces performance-oriented reports for clients and develops widely accepted benchmarks for IT products such as the AuthMark Benchmark for access and identity management products and the DirectoryMark Benchmark for LDAP directories. Mindcraft participated in the development of the SPECweb99 benchmark and improved the OSDB benchmark by fixing bugs and adding functionality. For more information about Mindcraft, please visit http://www.mindcraft.com
24
www.kingston.com/europe./server
10.0 Appendix 3: Legal notice Products and corporate names are trademarks and/or registered trademarks of their respective owners. Kingston Technology and/or Mindcraft shall not be liable for errors or omissions contained herein, nor for incidental or consequential damages resulting from the use of this material.
25
