| ... | Articles and Whitepapers |
| ... | Downloads |
| ... | Internet Resources |
| ... | Windows Server 2008 R2 Remote Desktop Services |
| ... | Windows Server 2003 Terminal Services |
| ... | About this Web Site |
| ... | Benny's Short Profile |
| ... | Benny's Biography |
| ... | Presentations 2010, 2009, 2008, 2007, 2006, 2005, 2004 and earlier |
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]
The test described here was performed in various instances from February to April, 2007 at the Microsoft GmbH subsidiary in Unterschleißheim.
Multiple Hewlett Packard ProLiant BL20p G3 servers, each with two dual core processors (2.8 GHz) and 4 or 8 GB RAM, were made available for each of the tests. These servers were operated in slightly different configurations:
The initial size of the page file was pre-configured to 2 GB on all servers.
Multiple PCs running under Windows Server 2003 or Windows XP, which were connected via a separate network to the servers to be tested, were used to generate the load. The measurement results are, however, independent of the absolute performance capacity of the hardware platforms for load generation. No performance limits were reached on the load generators during the test either in terms of storage or processor resources.
All servers are outfitted with a pre-installed English-language operating system. The terminal services were activated in the mode of an application server (terminal server). Office 2003 or 2007, respectively, was already pre-installed on the servers for the test on Microsoft Office. Special attention was paid to ensure that the installation mode was configured during the installation and that the common transform files for terminal servers from the corresponding Microsoft Office Resource Kits were used. Additional optimizations were not performed on the operating system or Microsoft Office.
120 local user accounts were set up with the following script. These accounts were used in the context of the test for interactive log on’s. All user accounts must be added to the “Users” and “Remote Desktop Users” local groups.
|
nMaxUsers = 121 |
Bginfo, a Microsoft Sysinternals tool, was pre-installed on all servers. The automatic start of Bginfo was removed from the startup sub-item of the start menu for all users in order to avoid negative side-effects on the test system.
During the test, no screen savers should be activated during the user sessions. For this reason, the screen savers for all user sessions were deactivated according to local policy. The accompanying configuration information may be found at Local Computer Policy | User Configuration | Administrative Templates | Control Panel | Display | Screen Saver.
Transform files were used to produce installations of Microsoft Office 2003 and Microsoft Office 2007 which were specially optimized for terminal servers. Transform files are a component of the Resource Kits for Office 2003 and Office 2007.
In order to stop the Configuration Wizard from opening during the first start up of an application on Microsoft Office 2003, Word, Excel, and PowerPoint will be opened for the first time in an administrator session, which was previously activated in the installation mode. In this manner, all associated configurations access the shadow-key section of the registry and will be passed on to all other user sessions. Automatically transferring the registry entries in the shadow-key section to a new user's profile prevents the Configuration Wizard from starting the first time that an application is opened in Office 2003. There is a corresponding administrative template for Microsoft Office 2007 that achieves this same result.
The installation of OpenOffice.org 2.1 on the server was quickly and easily configured in installation mode with an MSI file after unpacking the installation package on the local file system. The space occupied on the hard drive amounted to about 250 MB. There were color errors on the RDP 6 Client where the color depth was limited to 16 bit, but these errors were easily corrected by selecting a greater color depth.
The user-specific information from OpenOffice is stored in a folder hierarchy in the user profile instead of in the registry. The configuration files are based on XML, so they can also be modified as needed with a simple text editor. Upon the first start up of an OpenOffice application by a user, a series of basic settings must be made with a configuration wizard. This can be effectively avoided during a test if the entire directory hierarchy of the XML configuration files is copied to the user profile before starting OpenOffice for the first time. The hierarchy of a user’s XML configuration files is available as source material for the configuration, thus preemptively answering all of the configuration wizard’s queries.
During the installation of OpenOffice, the activation of the accompanying runtime environment is anchored in the startup folder. This accelerates the start up of the actual OpenOffice applications. However, this early start of the runtime environment for each new session does not permit pre-configuration by means of copying the XML configuration files. Consequently, the option to activate the runtime environment was removed from the startup folder for the purposes of the test. The runtime environment only started when the OpenOffice application was opened for the first time.