Scalability of Office Suites on Terminal Servers

Posted by Benny Tritsch on May 10, 2007

[Introduction] [Methodology] [Performing the Test] [Measurements] [Results] [Conclusions]
[Appendix 1: Configuration Details] [Appendix 2: Step-by-Step Description]

Methodology – Test Procedure and Measuring Methods

Systematic

The performance of load simulations under standardized conditions is often designated as “benchmarking“; however, no generally accepted standardized test procedure exists for applications and terminal servers. For this reason, the term “load test” should be used for the analyses referenced here.

The analyses and results of this load test should not be interpreted in isolation. The selection of the applications launched in terminal server sessions during the test ensured that the load generated on the server resources would be reproducible. In addition, the test possesses a static quality because users only logged on to the system, started three applications and their accompanying documentation, remaining inactive for the rest of the test. This feature allows for the generation of reproducible results, on the one hand, but on the other hand the results can vary depending on the applications activated as well on the documents used.

Nevertheless, this type of test provides clear results that emerge from the comparison of multiple pre-defined scenarios. The absolute numbers, such as the maximum number of user sessions, are therefore less relevant; the relative relationships between two scenarios are much more relevant than the numbers taken individually. The result of this test should above all provide greater insight as to which scenario requires proportionately more or less system resources in comparison with other scenarios.

Tools Used

Unfortunately, there is no standardized and easily reproduced procedure that can be used to test the performance of applications on terminal servers. Manufacturers such as Microsoft and Citrix do provide tools for this purpose, although they have not successfully established any generally recognized system for comparison.

Only the tools that were easiest to use were implemented for the test described here. This facilitates the reproducibility of the reference scenarios and results.

• vRD Load Edition, a tool for the timed control of logon sequences by test users on a terminal server and for the representation of all related sessions.
• Scripts on the basis of the Cmd.exe command line interpreters including the Sleep.exe tool from the Windows Server 2003 Resource Kit.
• Scripts on the basis of the Windows Script Host, which can be executed from the Windows command line with Cscript.exe.
• Msinfo32.exe to determine the exact hardware configuration of the server.
• The Pslist.exe Information Lister from Sysinternals to list all started processes at the end of a test sequence.

Next